v0.8.3: WebUI models, providers and keys; garbled thinking fixed

Streaming: Chinese reasoning showed up garbled in the WebUI. The agent
loop now feeds visible/thinking text in runs instead of byte by byte,
and the WebUI holds an incomplete UTF-8 tail until the next chunk
(cst_utf8_hold_feed), so every think_delta/assistant_delta is whole.

src/provider_switch.c: model list fetch/cache, key verify-and-save and
provider switching moved out of tui.c and shared by TUI and WebUI.

WebUI API:
- GET /api/providers, GET /api/models[?refresh=1] (401 on a rejected
  key, still listing built-ins), POST /api/provider, POST /api/key
  (key checked against /v1/models, saved to credentials 0600, never
  echoed in replies, events or logs)
- /api/config validates everything before applying, persists model,
  effort and approval (not yolo), warns when effort does not apply;
  plan_mode no longer forces read-only
- state: turn_usage, api_base, key_set/key_from, model_info,
  effort_applies, sees_images, max_steps, tool_profile, perm_timeout;
  permission events carry their timeout

Frontend: searchable model picker with refresh and manual entry,
provider picker and key dialog, effort/vision hints, toasts instead of
silent failures, Chinese error text, real protocol in the settings and
usage panels, approval countdown from the server's timeout.

Tests: utf8 hold unit test, scripts/e2e_webui_settings.py (fake
/v1/models, wrong key, switching, persistence, Chinese thinking) run
from e2e_webui.sh, scripts/tui_keys/provider.py.

.gitea/workflows/ci.yml

@@ -54,6 +54,8 @@ jobs:
           ./cst-agent --help
           ./cst-agent --workspace "$PWD" --list-sessions
           ./cst-agent --demo-ctl --workspace .
+      - name: WebUI end-to-end (recorded stream, no API key)
+        run: sh scripts/e2e_webui.sh
 
   musl:
     runs-on: ubuntu-latest

.gitignore

@@ -30,4 +30,5 @@ vendor/sashidown/*.o
 vendor/sashidown/src/*.o
 vendor/sashidown/libsashidown.a
 vendor/sashidown/libsashidown-mingw.a
-.gen/models_builtin.inc
+.gen/
+.build-tmp/

CMakeLists.txt

@@ -5,6 +5,7 @@ set(CMAKE_C_STANDARD 11)
 set(CMAKE_C_STANDARD_REQUIRED ON)
 
 option(CST_BUILD_TUI "Build ncurses/PDCurses TUI" ON)
+option(CST_BUILD_WEBUI "Build the HTTP/SSE WebUI (--web)" ON)
 option(CST_NATIVE_ARCH "Pass -march=native to sashidown" OFF)
 option(CST_BUNDLE_BUSYBOX "Install placeholder/scripts for busybox beside binary" ON)
 
@@ -110,20 +111,64 @@ set(CST_CORE_SOURCES
   src/agent_store.c
   src/json_baseline.c
   src/creds.c
+  src/provider_switch.c
   src/mm_tool.c
   src/models.c
   src/model_registry.c
   src/metrics.c
+  src/net.c
+  src/http_srv.c
+  src/web_md.c
 )
 
 # Built-in model data as a C byte list (included by model_registry.c).
 set(CST_GEN_DIR "${CMAKE_CURRENT_BINARY_DIR}/gen")
 set_property(DIRECTORY APPEND PROPERTY CMAKE_CONFIGURE_DEPENDS
   "${CMAKE_CURRENT_SOURCE_DIR}/src/models_builtin.json")
+
+# The WebUI shell (web/*) as C byte arrays, the same trick as the model data.
+file(GLOB CST_WEB_ASSETS "${CMAKE_CURRENT_SOURCE_DIR}/web/*")
+set_property(DIRECTORY APPEND PROPERTY CMAKE_CONFIGURE_DEPENDS ${CST_WEB_ASSETS})
+set(_cst_web_inc "/* Generated by CMake from web/ - do not edit. */\n")
+set(_cst_web_table "")
+foreach(_f ${CST_WEB_ASSETS})
+  if(IS_DIRECTORY "${_f}")
+    continue()
+  endif()
+  get_filename_component(_name "${_f}" NAME)
+  string(REGEX REPLACE "[^A-Za-z0-9]" "_" _var "wui_asset_${_name}")
+  file(READ "${_f}" _hex HEX)
+  string(REGEX REPLACE "([0-9a-f][0-9a-f])" "0x\\1," _hex "${_hex}")
+  set(_ctype "application/octet-stream")
+  if(_name MATCHES "\\.html$")
+    set(_ctype "text/html; charset=utf-8")
+  elseif(_name MATCHES "\\.js$")
+    set(_ctype "text/javascript; charset=utf-8")
+  elseif(_name MATCHES "\\.css$")
+    set(_ctype "text/css; charset=utf-8")
+  elseif(_name MATCHES "\\.svg$")
+    set(_ctype "image/svg+xml")
+  endif()
+  string(APPEND _cst_web_inc "static const unsigned char ${_var}[] = {${_hex}0x00};\n")
+  string(APPEND _cst_web_table "  {\"${_name}\", \"${_ctype}\", ${_var}, sizeof(${_var}) - 1},\n")
+endforeach()
+string(APPEND _cst_web_inc "static const wui_asset_t wui_assets[] = {\n${_cst_web_table}};\n")
+string(APPEND _cst_web_inc "static const int wui_asset_count = (int)(sizeof(wui_assets) / sizeof(wui_assets[0]));\n")
+file(WRITE "${CST_GEN_DIR}/web_assets.inc" "${_cst_web_inc}")
+if(NOT CST_BUILD_WEBUI)
+  list(APPEND CST_CORE_SOURCES src/webui_stub.c)
+else()
+  list(APPEND CST_CORE_SOURCES src/webui.c)
+endif()
+
 file(READ "${CMAKE_CURRENT_SOURCE_DIR}/src/models_builtin.json" _cst_models_hex HEX)
 string(REGEX REPLACE "([0-9a-f][0-9a-f])" "0x\\1," _cst_models_hex "${_cst_models_hex}")
 file(WRITE "${CST_GEN_DIR}/models_builtin.inc" "${_cst_models_hex}\n")
 include_directories("${CST_GEN_DIR}")
+
+if(NOT CST_BUILD_WEBUI)
+  add_compile_definitions(CST_NO_WEBUI=1)
+endif()
 if(WIN32)
   # MinGW has no <regex.h>: musl's TRE engine, UTF-8 aware (vendor/tre-regex)
   include_directories(vendor/tre-regex)
@@ -269,17 +314,26 @@ foreach(t ${_CST_SMALL_TESTS})
 endforeach()
 
 # Tests that link the whole core (no TUI).
-foreach(t test_llm_body test_prefix_stable test_sse_decode test_apply_patch
-          test_session_roundtrip test_attach
-          test_shell test_diff test_tools_fs test_registry
-          test_undo test_config test_models test_model_registry test_sandbox
-          test_keymap test_shell_compact test_skills
-          test_plugins test_mcp test_subagent test_lowil
-          test_win_utf8 test_os_dir test_tmpdir)
+set(_CST_CORE_TESTS test_llm_body test_prefix_stable test_sse_decode
+    test_apply_patch test_session_roundtrip test_attach
+    test_shell test_diff test_tools_fs test_registry
+    test_undo test_config test_models test_model_registry test_sandbox
+    test_keymap test_shell_compact test_skills
+    test_plugins test_mcp test_subagent test_lowil
+    test_win_utf8 test_os_dir test_tmpdir)
+if(CST_BUILD_WEBUI)
+  # test_webui drives the real server; the stub has nothing to test.
+  list(APPEND _CST_CORE_TESTS test_webui)
+endif()
+foreach(t ${_CST_CORE_TESTS})
   add_executable(${t} tests/${t}.c ${CST_CORE_SOURCES})
-  target_include_directories(${t} PRIVATE include "${CJSON_DIR}")
+  # The core now contains src/web_md.c, which needs sashidown's header and
+  # archive exactly like the Makefile build does.
+  target_include_directories(${t} PRIVATE include "${CJSON_DIR}" "${SD_DIR}/include")
   target_compile_definitions(${t} PRIVATE _GNU_SOURCE)
-  target_link_libraries(${t} PRIVATE CURL::libcurl Threads::Threads)
+  target_link_libraries(${t} PRIVATE
+    "${SASHIDOWN_A}" CURL::libcurl Threads::Threads)
+  add_dependencies(${t} sashidown)
   if(NOT WIN32)
     target_link_libraries(${t} PRIVATE m)
   else()

Makefile

@@ -43,6 +43,10 @@ ifeq ($(strip $(CURL_LIBS)),)
 CURL_LIBS := -lcurl
 endif
 
+# WebUI (HTTP + SSE front end): src/net.c src/http_srv.c src/web_md.c
+# src/webui.c. CST_WEBUI=0 swaps in the stub (--web then prints a hint).
+CST_WEBUI ?= 1
+
 PKG_CFLAGS := $(CURL_CFLAGS) $(CURSES_CFLAGS)
 PKG_LIBS := $(CURL_LIBS) $(CURSES_LIBS)
 
@@ -50,6 +54,17 @@ CPPFLAGS += -Iinclude -I.gen -I$(SD_DIR)/include -I$(CJSON_DIR) -D_XOPEN_SOURCE_
 LDFLAGS += -L$(SD_DIR)
 LDLIBS += -lsashidown -lm $(PKG_LIBS) $(PTHREAD_LIBS)
 
+WEB_DIR := web
+WEB_ASSETS := $(wildcard $(WEB_DIR)/*.html) $(wildcard $(WEB_DIR)/*.js) \
+	$(wildcard $(WEB_DIR)/*.css) $(wildcard $(WEB_DIR)/*.svg)
+ifeq ($(CST_WEBUI),0)
+WEB_SRC := src/net.c src/http_srv.c src/web_md.c src/webui_stub.c
+# The stub body is guarded by this, so the switch has to define it too.
+CPPFLAGS += -DCST_NO_WEBUI=1
+else
+WEB_SRC := src/net.c src/http_srv.c src/web_md.c src/webui.c
+endif
+
 CORE_SRC = \
 	$(CJSON_DIR)/cJSON.c \
 	src/util.c \
@@ -76,15 +91,16 @@ CORE_SRC = \
 	src/agent_loop.c \
 	src/agent_store.c \
 	src/json_baseline.c \
-	src/creds.c \
+	src/creds.c src/provider_switch.c \
 	src/mm_tool.c \
 	src/models.c src/model_registry.c \
-	src/metrics.c
+	src/metrics.c \
+	$(WEB_SRC)
 
 SRC = $(CORE_SRC) src/textsel.c src/tui.c src/main.c
 OBJ = $(SRC:.c=.o)
 HDRS = $(wildcard include/*.h) $(wildcard src/tools/*.h) $(CJSON_DIR)/cJSON.h \
-	.gen/models_builtin.inc
+	.gen/models_builtin.inc .gen/web_assets.inc
 
 TESTS = \
 	tests/test_ctl_frontier \
@@ -120,8 +136,12 @@ TESTS = \
 	tests/test_win_utf8 \
 	tests/test_os_dir \
 	tests/test_tmpdir
+# test_webui drives the real server; there is nothing to test with the stub.
+ifneq ($(CST_WEBUI),0)
+TESTS += tests/test_webui
+endif
 
-TEST_LIBS = -lm $(CURL_LIBS) $(PTHREAD_LIBS)
+TEST_LIBS = -L$(SD_DIR) -lsashidown -lm $(CURL_LIBS) $(PTHREAD_LIBS)
 
 .PHONY: all clean check demo sashidown-lib FORCE cmake-gcc cmake-clang asan install uninstall
 
@@ -153,6 +173,11 @@ cst-agent: $(OBJ) $(SD_DIR)/libsashidown.a
 	@mkdir -p .gen
 	od -An -v -tx1 $< | sed 's/\([0-9a-f][0-9a-f]\)/0x\1,/g' > $@.tmp && mv $@.tmp $@
 
+# The WebUI shell is embedded the same way, one array per file.
+.gen/web_assets.inc: $(WEB_ASSETS) scripts/embed_web.sh
+	@mkdir -p .gen
+	@sh scripts/embed_web.sh $@ $(WEB_DIR)
+
 %.o: %.c $(HDRS)
 	$(CC) $(CFLAGS) $(CPPFLAGS) -c -o $@ $<
 
@@ -186,10 +211,14 @@ tests/test_markdown: tests/test_markdown.c $(SD_DIR)/libsashidown.a
 tests/test_%: tests/test_%.c $(CORE_SRC) $(HDRS)
 	$(CC) $(CFLAGS) $(CPPFLAGS) -o $@ $< $(CORE_SRC) $(TEST_LIBS)
 
+$(TESTS): sashidown-lib
+
 check: $(TESTS)
 	@set -e; for t in $(TESTS); do echo "== $$t"; ./$$t; done
 
 # AddressSanitizer + UBSan build of the agent and all tests.
+# Under a ptrace-based sandbox (some Termux setups) LeakSanitizer aborts with
+# "Tracer caught signal 11"; add ASAN_OPTIONS=detect_leaks=0 there.
 asan:
 	$(MAKE) clean
 	$(MAKE) sashidown-lib
@@ -199,7 +228,7 @@ demo: cst-agent
 	./cst-agent --demo-ctl --workspace .
 
 clean:
-	rm -f $(OBJ) cst-agent $(TESTS) .gen/models_builtin.inc
+	rm -f $(OBJ) cst-agent $(TESTS) .gen/models_builtin.inc .gen/web_assets.inc
 
 install: cst-agent
 	install -d "$(DESTDIR)$(PREFIX)/bin"

README.md

@@ -126,6 +126,23 @@ API Key：TUI 里输入并按 provider 保存到配置目录的 `credentials`（
 | Android Termux | aarch64 / … | Clang (Termux) | yes | ncurses（Termux 包） | `$PREFIX/bin/sh`；可旁路 busybox |
 | Windows 10 1903+ | x86_64 | MinGW-w64 GCC（UCRT） | yes | ncurses 6.6（Windows 控制台） | **捆绑 busybox-w32** |
 
+### WebUI（`--web`）
+
+同一个二进制内置一个 HTTP + SSE 前端：`cst-agent --web`（默认 `http://127.0.0.1:8787/`）。
+它和 TUI 是同一引擎的两个前端，浏览器里能流式看回答、点审批、切会话、贴图。
+协议是一份稳定的 JSON + SSE 接口，供以后的专用客户端复用。详见 [docs/webui.md](docs/webui.md)。
+
+```bash
+cst-agent --web                    # 本机浏览器打开 http://127.0.0.1:8787/
+cst-agent --web --web-port 0       # 端口交给系统，启动日志里给 URL
+cst-agent --web --web-dev ./web    # 开发前端：直接读目录，改完刷新
+cst-agent --web --web-host 0.0.0.0 --web-token 口令   # 手机/局域网访问
+sh scripts/e2e_webui.sh            # 用录制流跑完整轮次，不需要 API key
+```
+
+不带 TUI 的构建也可以开 WebUI（`cmake -DCST_BUILD_WEBUI=ON -DCST_BUILD_TUI=OFF`）；
+`-DCST_BUILD_WEBUI=OFF` 时 `--web` 只打印一条提示。
+
 ### Termux
 
 ```bash
@@ -195,10 +212,10 @@ Gitea Release `vVERSION`（同名资源会被替换）；仓库可设 `RELEASE_T
 | `CST_MAX_STEPS` / `--max-steps` | 每个用户回合最多调用模型几次（30） |
 | `CST_TOOL_OUT_CAP` | 单个工具输出发给模型的字节上限（16384，保留头 60% + 尾 40%） |
 | `CST_CTX_WINDOW` | 发给模型的最近非 system 消息条数（80） |
-| `CST_CTX_TOKENS` / `CST_AUTO_COMPACT=0` | 覆盖模型上下文窗口大小 / 关闭自动压缩（用到 80% 时自动 `/compact`） |
+| `CST_CTX_TOKENS` / `CST_AUTO_COMPACT=0` | 覆盖模型上下文窗口大小 / 关闭自动压缩（用到 75% 时自动 `/compact`） |
 | `CST_NO_PROJECT_NOTES=1` | 不把工作区的 `AGENTS.md` 注入 system prompt |
 | `NO_COLOR` / `CST_ASCII=1` | 关闭颜色 / 强制 ASCII 图形（非 UTF-8 终端自动退化） |
-| `CST_SHELL_TIMEOUT` | `run_shell` 超时秒数（120），超时整组进程被杀，exit=124 |
+| `CST_SHELL_TIMEOUT` | `run_shell` 超时秒数（默认不限），设了才在超时后整组杀进程，exit=124 |
 | `CST_METRICS=0` / `CST_METRICS_PATH` | 关闭 / 指定计时埋点 JSONL（默认 `.agent/metrics/<session>.jsonl`） |
 | `CST_RECORD_SSE=DIR` / `CST_REPLAY_SSE=DIR` | 录制 / 回放原始 SSE 流（`NNNN.sse` + `NNNN.req.json`） |
 | `CST_APPROVAL=ask\|auto\|yolo` / `CST_SANDBOX=off\|landlock\|seccomp` | 审批模式 / 关闭沙箱（回落到只读白名单规则）或只用指定沙箱 |
@@ -302,7 +319,7 @@ TUI 中 `/tools off NAME`、`/tools on NAME` 可临时开关单个工具。
   `sudo`、`rm`、解释器（`python3 -c …`）等危险前缀不会被建议成规则。
 - **撤销**：每轮写文件前自动快照到 `.agent/undo/`，`/undo` 恢复上一轮改动的全部文件（新建的文件会被删除）。
 - **AGENTS.md**：工作区根目录的 `AGENTS.md` 会注入 system prompt；`/init` 让 agent 自己生成一份。
-- **自动压缩**：上一轮 prompt 用到模型上下文 80% 时，下一轮开始前自动总结历史（`CST_AUTO_COMPACT=0` 关闭）。
+- **自动压缩**：上一轮 prompt 用到模型上下文 75% 时，下一轮开始前自动总结历史（`CST_AUTO_COMPACT=0` 关闭）。
 
 ### 扩展：Skills、插件、MCP、子 agent
 
@@ -391,6 +408,7 @@ Thought（推理过程）默认折叠成一行“Thought · 3.2s”，生成中
 | `/diff [path]` * | `git status` + `git diff` |
 | `/compact` | 手动压缩上下文 |
 | `/export [file]` *、`/copy` * | 导出 Markdown 记录；复制上一条回复（OSC 52，需终端支持） |
+| `/web` * | 后台起一个 WebUI 进程（同 workspace、新会话，日志写到 `.agent/webui.log`） |
 | `/init` | 让 agent 写 `AGENTS.md` |
 | `/details` *、`/thinking [on\|off]`、`/keys` * | 展开全部 tool 输出；Thought 默认展开/折叠；快捷键列表 |
 | `/skills [reload\|list]`、`/skill <name> [任务]` | 浏览 skills；带着某个 skill 开始一轮 |

docs/env_openai.md

@@ -73,7 +73,7 @@ export OPENAI_MODEL='your-model'
 | `CST_MAX_STEPS` | 每个用户回合最多调用模型几次（默认 30） |
 | `CST_TOOL_OUT_CAP` | 单个工具输出发给模型的字节上限（默认 16384） |
 | `CST_CTX_WINDOW` | 发给模型的最近非 system 消息条数（默认 80） |
-| `CST_SHELL_TIMEOUT` | `run_shell` 超时秒数（默认 120） |
+| `CST_SHELL_TIMEOUT` | `run_shell` 超时秒数（默认不限，Ctrl+C / 停止按钮仍可随时中断） |
 | `CST_METRICS=0` / `CST_METRICS_PATH` | 关闭 / 指定计时埋点 JSONL |
 | `CST_RECORD_SSE` / `CST_REPLAY_SSE` | 录制 / 回放原始 SSE 流目录；`CST_REPLAY_CHUNK` 控制回放分块字节数 |
 | `CST_MODELS` | `/model` 可选列表，`:` 或 `,` 分隔 |

docs/webui.md

@@ -0,0 +1,213 @@
+# WebUI（内置 HTTP + SSE 前端）
+
+`cst-agent --web` 让同一个进程同时提供命令行以外的前端：浏览器（或任何 HTTP 客户端）
+连上后能看到流式回答、工具卡片、审批弹窗和任务清单。它是 TUI 之外的第二个前端，
+**引擎没有为它改一行**——两边用的都是 `cst_agent_hooks_t` 这组回调。
+
+这份文档同时是**引擎 API v1 的规范**：以后专门客户端（Tauri / Qt / 移动端）直接照它实现，
+不需要读 C 代码。
+
+## 快速开始
+
+```bash
+# 默认只监听 127.0.0.1:8787
+cst-agent --web
+
+# 指定端口；0 = 让系统分配（启动日志里会打印实际 URL）
+cst-agent --web --web-port 8787
+
+# 让手机/另一台机器访问（必须给 token，否则拒绝启动在非 loopback 上）
+cst-agent --web --web-host 0.0.0.0 --web-token 我自己定的口令
+
+# 开发前端：直接从目录读 web/，改完刷新页面即可，不用重新编译
+cst-agent --web --web-dev ./web
+
+# 启动后自动打开浏览器
+cst-agent --web --web-open
+
+# 启动即跑一轮（不点页面也能用）
+cst-agent --web -p "看看这个仓库有什么问题"
+```
+
+启动日志：
+
+```
+webui: http://127.0.0.1:8787/
+webui: token 3f9a…  (append #token=3f9a… to the URL)
+```
+
+带 token 时把 `#token=…` 贴到浏览器地址栏后面即可（`#` 之后不会发给服务器，
+页面会用 `?token=` 再请求一次）。
+
+环境变量：`CST_WEB_PORT`、`CST_WEB_HOST`、`CST_WEB_TOKEN`、`CST_WEB_ROOT`。
+
+退出：终端里 Ctrl+C，或 `POST /api/quit`。
+
+## 安全模型
+
+| 场景 | 行为 |
+|---|---|
+| 默认 `127.0.0.1` | 不需要 token；`Host` 必须是 loopback（防 DNS rebinding）；带 `Origin` 的请求必须来自本页面自己（`http://<Host>`），否则 401 —— 防止任意网页跨站 POST 驱动 agent |
+| 静态资源 | 页面本身与 `app.js`/`style.css` 不需要 token；`/api/attach/<rel>` 图片仍要（页面会把 token 写进 `SameSite=Strict` cookie） |
+| 请求解析 | 鉴权与大小检查在读请求体**之前**完成；`Content-Length` 必须是纯数字且只有一个值；请求头整体 15 s 内收完 |
+| `--web-host 0.0.0.0` | 没给 `--web-token` 时自动生成随机 token 并打印；所有 `/api/*` 都要 token |
+| token 传递 | `Authorization: Bearer <t>`、`Cookie: cst_token=<t>` 或 `?token=<t>`（`EventSource` 不能设 header，所以查询参数是必需的一条路） |
+| 资源 | 只有白名单路由；路径里的 `..`/编码 NUL 一律 400；静态资源来自二进制内嵌（或 `--web-dev DIR`） |
+| 输出渲染 | 模型与工具输出都当不可信内容：Markdown 由服务端渲染，并经过标签/属性白名单（`javascript:` 链接、站外图片、`<script>` 全部丢弃） |
+| 请求体 | 默认 1 MiB，图片上传 32 MiB；超限 413 |
+
+**不要**把 WebUI 直接暴露到公网：没有 TLS，token 会以明文出现在网络里。需要远程访问时
+放在反向代理（https）后面，并确认 `X-Accel-Buffering: no` 之类的缓冲设置不会掐断 SSE。
+
+## 引擎 API（协议 `cst-webui/1`）
+
+所有响应都是 JSON（`Content-Type: application/json`）。`GET /api/state` 的
+`protocol` 字段给出版本号，客户端应当先看它再决定支持哪些端点。
+
+### 事件流
+
+```
+GET /api/events?since=<seq>      →  text/event-stream (chunked)
+```
+
+* 每条事件都有 `id: <seq>`（单调递增）与 `data: {"type":…,"seq":…, …}`。
+* 断线重连带上 `since=<最后收到的 seq>`（浏览器用 `Last-Event-ID` 也行），
+  服务端保留最近 2000 条，超出后从可用的最老一条开始；比当前 seq 还大（服务端重启过）
+  就当作新连接。
+* 新连接（不带 `since`）只收之后的事件；但若有尚未回答的 `permission`/`ask_user`，
+  会从那条事件开始补发，刷新页面不会丢掉正在等的问题。
+* 心跳是 SSE 注释 `: ping`（每秒一次，客户端忽略）。
+* `state` 事件在状态变化时整份重发，客户端可以直接替换本地状态。
+
+事件类型：
+
+| type | 字段 | 说明 |
+|---|---|---|
+| `state` | 见下表 | 会话/模型/审批/todo/用量快照 |
+| `status` | `message` | 引擎自己的一句话状态（"calling model (step 2)…"） |
+| `assistant_html` | `text` | **已渲染且已消毒的 HTML 片段**，按顺序追加 |
+| `assistant_delta` | `text` | 原始文本增量（逐字节真相，做复制/导出时用） |
+| `think_delta` | `text` | 推理过程增量（Thought） |
+| `tool` | `id`,`name`,`args`,`phase`,`detail` | `phase` = `open`/`ready`/`run`/`done` |
+| `tool_result` | `id`,`name`,`out`,`exit`,`bytes`,`truncated` | 完整输出被截到 32 KiB，`truncated=true` |
+| `todo` | `text` | `todo_write` 替换后的清单（一行一项） |
+| `permission` | `id`,`tool`,`detail`,`preview`,`sandbox`,`timeout` | 需要回答（见下）；`timeout` 秒后按拒绝处理 |
+| `permission_done` | `id`,`answer` | 问题已结束（超时/取消也会发） |
+| `ask_user` | `id`,`question`,`options[]` | 需要回答 |
+| `session` | `id` | 切换/新建了会话，客户端应重新拉 `/api/session` |
+| `turn_end` | `rc`,`error` | 本轮结束（`rc=0` 正常，非 0 见 `error`） |
+
+`state` 字段：`workspace` `session` `provider` `model` `effort` `approval`
+(`ask|auto|yolo`) `plan_mode` `protocol_tools` `status_busy` `todo`
+`context_tokens` `undo_available` `pending_images[]` `usage{requests,prompt_tokens,
+completion_tokens,cached_tokens,cache_miss_tokens,cost,priced,currency}`
+`turn_usage{…}`（上一轮，字段同 `usage`）`api_base` `key_set` `key_from`
+（`saved` / 环境变量名 / `session` / 空，**从不含 key 本身**）
+`model_info{label,context,vision,reasoning,source,price?}` `effort_applies`
+`sees_images`（1 / 0 / -1 未知）`max_steps` `tool_profile` `perm_timeout`。
+`cached_tokens` / `cache_miss_tokens` 来自上游的 prompt cache 统计，前端用它算命中率。
+
+`think_delta` / `assistant_delta` 的 `text` 总是完整的 UTF-8：上游在多字节字符中间
+断开时，残缺的尾巴留到下一块再发。
+
+### 请求
+
+| 方法 | 路径 | 说明 |
+|---|---|---|
+| GET | `/api/state` | 上面那份快照 |
+| GET | `/api/session?from=N` | 会话消息（`{messages:[…]}`，形状与 `.agent/sessions/*.jsonl` 一致；assistant 消息额外带 `html`） |
+| GET | `/api/sessions` | `{current, sessions:[{id,title,messages,mtime}]}` |
+| POST | `/api/sessions/new` | `{}` → `{id}` |
+| POST | `/api/sessions/switch` | `{"id":"…"}` |
+| POST | `/api/workspace` | `{"path":"/abs/dir"}` —— 换工作目录（权限规则/埋点/AGENTS.md 跟着换），并**在新目录里开一个新会话**（会话是按目录存的）；返回 `{id}`，坏路径 → 400 |
+| POST | `/api/turn` | `{"text":"…"}` → 202 `{ok:true}`；已有轮次在跑时 **409** |
+| POST | `/api/cancel` | `{}` —— 取消当前轮（不会杀进程）；挂起的权限问题按拒绝、提问按忽略处理 |
+| POST | `/api/permission` | `{"id":N,"answer":0\|1\|2\|3}` → deny / once / always-tool / always-prefix |
+| POST | `/api/ask_user` | `{"id":N,"text":"…"}` 或 `{"id":N,"dismiss":true}` |
+| POST | `/api/config` | `{model?,effort?,approval?,plan_mode?}`；先整体校验再生效，model/effort/approval（`yolo` 除外）写进 `settings.json`；当前模型不吃 effort 时回 `warning`；带 `provider` → 400（改用 `/api/provider`） |
+| GET | `/api/providers` | `{current, providers:[{name,api_base,model,has_key,key_from,key_env,source,current}]}` |
+| GET | `/api/models[?refresh=1]` | 当前提供商的模型：`{provider,current,api_base,key_set,api_count,models:[{id,api,label,context,vision,reasoning,source,price?}]}`；先用缓存，过期/缺失/`refresh=1` 时向 `/v1/models` 拉；key 被拒 → 401（仍带内置模型和 `auth:true`），其它失败放在 `api_error` |
+| POST | `/api/provider` | `{"name":"…","key":"…"?}`：切换提供商。缺 key → 400 `{need_key:true}`；key 被拒 → 401；没这个提供商 → 404；成功 → `{ok,provider,model,warning?}`，key 验证后存进 `credentials`（0600） |
+| POST | `/api/key` | `{"key":"…"}`：换当前提供商的 key，验证与回复同上 |
+| POST | `/api/undo` | 回滚最近一轮的文件改动，回 200 + 报告文本 |
+| POST | `/api/compact` | 手动压缩上下文 |
+| POST | `/api/attach?name=x.png` | 请求体是图片原始字节（≤32 MiB），进入"待发送"列表 |
+| DELETE | `/api/attach` | 清空待发送图片 |
+| GET | `/api/attach/<rel>` | 读回已存进会话的图片（`rel` 由消息的 `attach[].rel` 给出） |
+| POST | `/api/quit` | 关闭进程 |
+
+未命中路由 → `404 {"error":"not found"}`；需要 token 而没给 → `401`。
+轮次运行中（或另一个修改正在进行）时，`sessions/new`、`sessions/switch`、`workspace`、
+`undo`、`compact`、`attach`（上传/清空）、`provider`、`key` 以及改 `model/effort` 的
+`config` 一律 **409**；`approval`/`plan_mode` 可以中途改。
+
+key 只出现在请求体里：回复、事件、日志都不会带它。`/api/provider` 与 `/api/key`
+跟其它 `/api/*` 一样受 loopback 同源检查或 token 保护。
+
+### 客户端接入示例（10 行）
+
+```python
+import json, urllib.request
+base = "http://127.0.0.1:8787"
+def api(path, body=None, token=None):
+    req = urllib.request.Request(base + path, method="POST" if body is not None else "GET")
+    if token: req.add_header("Authorization", "Bearer " + token)
+    data = json.dumps(body).encode() if body is not None else None
+    return urllib.request.urlopen(req, data).read()
+api("/api/turn", {"text": "hello"})           # 之后读 /api/events 收流
+```
+
+### 空 body 的 POST
+
+`POST /api/cancel`、`/api/undo`、`/api/compact`、`/api/quit` 不需要 body。
+没有 `Content-Length` 头的 POST 按 RFC 9112 §6.3 视为**空 body**（`curl -X POST`
+正是这样发的），但该连接不再复用（服务端随后关闭它）：后面是否跟着未声明的
+字节无法判断。
+
+## 前端实现（web/）
+
+没有构建步骤：`web/index.html` + `web/app.js` + `web/style.css` 是纯 ES module，
+编译时被 `scripts/embed_web.sh` 塞进二进制（`.gen/web_assets.inc`，跟内置模型数据同一招），
+所以 Windows 发布包依旧是单个 exe。`--web-dev DIR` 可以绕过内嵌版本直接读目录。
+
+Markdown 在**服务端**渲染（复用 TUI 用的 sashidown 流式 API），客户端只贴 HTML；
+因此不需要引入任何 JS 依赖，也顺带保证了"只发送已确认的 HTML"。
+
+界面本身：左栏可收（顶栏按钮 / `?side=0`），右栏可收（`?panel=0`），主题 `?theme=dark|light`；
+三个开关都会记进 `localStorage`，带查询参数时以参数为准。空状态只留一句提示，第一轮
+对话开始后自动隐去；右栏只给三件事——任务清单、本轮文件改动行数（`+N −M`）、上下文
+占用与 prompt cache 命中率。布局类在页面脚本之前就写进 `<body>`，避免首帧闪一下再收起。
+
+TUI 里打 `/web` 会另起一个 WebUI 进程（`--web --workspace <当前目录> --new`），
+日志在 `<workspace>/.agent/webui.log`：TUI 不能被 `cst_webui_run()` 的阻塞循环占住，
+而且两个进程同时写一个会话文件会把它写坏，所以子进程总是开新会话。
+
+自动压缩的门槛是模型上下文窗口的 **75%**（`CST_CTX_BUDGET` 可覆盖）；到点后下一轮
+开始前先总结历史，而不是等到塞满。
+
+## 测试
+
+```bash
+make check                        # 含 tests/test_webui（HTTP/SSE/鉴权/消毒）
+sh scripts/e2e_webui.sh          # 用录制流跑完整轮次 + 权限往返，不需要 API key
+                                  # （含 e2e_webui_settings.py：假 /v1/models、换提供商/key、中文思考流）
+sh scripts/e2e_webui.sh tests/fixtures/sse/deepseek-ctl ctl
+```
+
+手工验收清单（浏览器）：
+
+1. 打开首页 → 顶部显示 workspace/模型/审批模式，右下状态"就绪"。
+2. 输入一句话 → 助手气泡逐字出现；工具卡片自动展开、结束后带 `exit=` 标记。
+3. 切到 `ask` 审批 → 写文件时弹窗；点"允许一次"后继续，点"拒绝"则该工具返回失败但对话继续。
+4. `会话` 按钮：新建/切换后历史正确重建（刷新页面也一样）。
+5. 拖入一张图片 → 底部出现待发送提示，发送后历史里能看到图。
+6. 关掉标签页再打开 → 事件流自动续传，不丢已完成的消息。
+
+## 限制与已知问题
+
+* 同一会话同时只能有一轮（`POST /api/turn` 返回 409）；TUI 与 WebUI 目前是
+  **二选一**（`--web` 时不启动 TUI），同进程并存留待后续的 hooks 分发层。
+* 一个会话文件只能由一个进程写：不要同时用 TUI 打开 WebUI 正在用的那个 workspace。
+* 没有 WebSocket；服务端→客户端用 SSE，客户端→服务端用 POST。够用且穿透代理更容易。
+* 事件环只有 2000 条：断线太久（超过这个窗口）的客户端会从当前可用的最老事件开始，
+  这时应当重新拉一次 `/api/session`。

include/cst_agent.h

@@ -321,6 +321,12 @@ cst_agent_config_t *cst_agent_config_mut(cst_agent_t *a);
 void cst_agent_request_cancel(cst_agent_t *a);
 int cst_agent_new_session(cst_agent_t *a, char *id_out, size_t id_sz);
 int cst_agent_switch_session(cst_agent_t *a, const char *id);
+/* Re-root the agent at another directory. Per-workspace state (permission
+ * rules, metrics, project notes) follows, the system prompt is rebuilt and the
+ * token budget is re-measured. The session is not touched: the caller decides
+ * whether to keep it or start a new one. */
+int cst_agent_set_workspace(cst_agent_t *a, const char *path, char *err,
+                            size_t err_sz);
 int cst_agent_compact(cst_agent_t *a, char *err, size_t err_sz);
 /* Bare id of the current session (file name without .jsonl). */
 void cst_agent_session_id(const cst_agent_t *a, char *out, size_t out_sz);

include/cst_http.h

@@ -0,0 +1,108 @@
+#ifndef CST_HTTP_H
+#define CST_HTTP_H
+
+/* A small HTTP/1.1 server for the built-in WebUI (src/http_srv.c).
+ *
+ * Only what a local single-page app needs: exact-match routes, keep-alive,
+ * fixed-length bodies and chunked streaming responses (SSE). No TLS, no
+ * HTTP/2, no chunked *request* bodies, no CGI. Every connection is served
+ * by its own thread; a handler may keep the connection as long as it wants
+ * (that is how the event stream works). */
+
+#include "cst_net.h"
+
+#include <stddef.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef struct cst_http_conn cst_http_conn_t;
+
+typedef struct {
+  char method[8];  /* GET / POST / DELETE ... */
+  char path[1024]; /* percent-decoded, without the query string */
+  char query[1024]; /* raw, without '?' */
+  const char *body; /* NUL-terminated; may hold binary bytes */
+  size_t body_len;
+  /* Headers of interest, NULL when absent (pointers into the request). */
+  const char *authorization;
+  const char *cookie;
+  const char *origin;
+  const char *host;
+  const char *content_type;
+  const char *user_agent;
+  const char *last_event_id; /* an EventSource reconnecting */
+  int keep_alive;
+} cst_http_req_t;
+
+/* 0 = handled (a response was sent), -1 = not found (the server replies). */
+typedef int (*cst_http_fn)(cst_http_conn_t *c, void *ud);
+
+typedef struct {
+  const char *method; /* "*" matches any */
+  const char *path;   /* exact; "*" = fallback (try last) */
+  cst_http_fn fn;
+  void *ud;
+  int auth;         /* ask the auth hook before the handler */
+  size_t max_body;  /* request body cap; 0 = server default */
+} cst_http_route_t;
+
+typedef struct {
+  const char *host; /* NULL/"" = 127.0.0.1 */
+  int port;         /* 0 = pick a free port */
+  size_t max_body;  /* default 1 MiB */
+  int max_conn;     /* default 32 */
+  int read_timeout_ms; /* waiting for the request headers (default 15000) */
+  int idle_timeout_ms; /* keep-alive pause between requests (default 60000) */
+  const cst_http_route_t *routes;
+  int nroutes;
+  /* 0 = allow, -1 = the server answers 401 before the handler runs. */
+  int (*auth)(const cst_http_req_t *req, void *ud);
+  void *auth_ud;
+  /* One line per request, e.g. "GET /api/state -> 200". */
+  void (*on_log)(const char *line, void *ud);
+  void *log_ud;
+} cst_http_opts_t;
+
+typedef struct cst_http_server cst_http_server_t;
+
+/* Start listening. NULL on failure with a reason in err. */
+cst_http_server_t *cst_http_start(const cst_http_opts_t *o, char *err,
+                                  size_t err_sz);
+int cst_http_port(const cst_http_server_t *s);
+/* Non-zero once stop() was called: long-lived handlers must return. */
+int cst_http_stopping(const cst_http_server_t *s);
+/* Close the listener and every connection, then wait (30 s at most) for the
+ * handler threads. Returns how many still run: their ud must stay alive. */
+int cst_http_stop(cst_http_server_t *s);
+/* Does nothing while handlers still run on s. */
+void cst_http_free(cst_http_server_t *s);
+
+/* --- inside a handler --- */
+const cst_http_req_t *cst_http_request(cst_http_conn_t *c);
+int cst_http_send(cst_http_conn_t *c, int status, const char *ctype,
+                  const void *body, size_t n);
+int cst_http_send_str(cst_http_conn_t *c, int status, const char *ctype,
+                      const char *s);
+/* json may be NULL (then "null" is sent); it is copied, not owned. */
+int cst_http_send_json(cst_http_conn_t *c, int status, const char *json);
+/* {"error":"..."} */
+int cst_http_send_err(cst_http_conn_t *c, int status, const char *msg);
+/* True once a response (or stream) began for this request. */
+int cst_http_sent(const cst_http_conn_t *c);
+
+/* Chunked response: begin(), any number of write()s, end(). */
+int cst_http_stream_begin(cst_http_conn_t *c, const char *ctype);
+int cst_http_stream_write(cst_http_conn_t *c, const void *data, size_t n);
+int cst_http_stream_end(cst_http_conn_t *c);
+
+/* Query parameter, percent-decoded ('+' = space). 0 when found. */
+int cst_http_query(const cst_http_req_t *req, const char *key, char *out,
+                   size_t out_sz);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* CST_HTTP_H */

include/cst_net.h

@@ -0,0 +1,68 @@
+#ifndef CST_NET_H
+#define CST_NET_H
+
+/* Minimal TCP sockets for the embedded HTTP server (src/net.c).
+ *
+ * The point of this layer is that src/http_srv.c and src/webui.c never
+ * include a socket header: Unix file descriptors and Windows SOCKETs differ
+ * in type, close() and error reporting, and Windows needs WSAStartup before
+ * anything works. */
+
+#include <stddef.h>
+#include <stdint.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef _WIN32
+typedef intptr_t os_socket_t;
+#define OS_SOCKET_BAD ((os_socket_t)-1)
+#else
+typedef int os_socket_t;
+#define OS_SOCKET_BAD (-1)
+#endif
+
+/* Longest "address:port" os_net_peer writes (IPv6 hosts are bracketed so a
+ * URL built from it stays unambiguous). */
+#define OS_NET_ADDR_MAX 64
+
+/* Winsock startup (no-op on Unix, which also ignores SIGPIPE so a client
+ * that hangs up cannot kill the process). 0 / -1. */
+int os_net_init(void);
+void os_net_cleanup(void);
+
+/* Bind and listen on host:port. host may be NULL/"" (then "127.0.0.1"),
+ * "0.0.0.0"/"::" (every interface) or any literal address / name; port 0
+ * picks a free one. Returns the listening socket, or OS_SOCKET_BAD with a
+ * short reason in err. */
+os_socket_t os_net_listen(const char *host, int port, char *err, size_t err_sz);
+/* The port actually bound (0 when unknown). */
+int os_net_port(os_socket_t s);
+/* A one-line "host:port" for messages. */
+void os_net_peer(os_socket_t s, char *out, size_t out_sz);
+
+/* Connect to host:port, waiting up to timeout_ms: the socket, or
+ * OS_SOCKET_BAD. Used by the tests and by anything that drives the WebUI
+ * API from inside the process. */
+os_socket_t os_net_connect(const char *host, int port, int timeout_ms);
+
+/* Accept one connection, waiting up to timeout_ms (<0 = forever).
+ * Returns the socket, OS_SOCKET_BAD on error, or OS_SOCKET_BAD with
+ * *timed_out set when the wait ran out. */
+os_socket_t os_net_accept(os_socket_t s, int timeout_ms, int *timed_out);
+
+/* Read at most n bytes, waiting up to timeout_ms: bytes read, 0 at EOF,
+ * -1 on error, -2 when nothing arrived in time. */
+long os_net_recv(os_socket_t s, void *buf, size_t n, int timeout_ms);
+/* Write everything (timeout_ms per call): 0 / -1. */
+int os_net_send_all(os_socket_t s, const void *buf, size_t n, int timeout_ms);
+/* Wake up a thread blocked in recv/send on this socket (shutdown, no close). */
+void os_net_shutdown(os_socket_t s);
+void os_net_close(os_socket_t s);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* CST_NET_H */

include/cst_provider_switch.h

@@ -0,0 +1,79 @@
+#ifndef CST_PROVIDER_SWITCH_H
+#define CST_PROVIDER_SWITCH_H
+
+/* Model lists, API keys and provider switches, shared by the TUI and the
+ * WebUI so both front ends validate and persist the same way.
+ *
+ * A key is checked with GET /models: a 401/403 (CST_MODELS_AUTH) rejects
+ * it, any other failure (offline, no /models route) lets it through
+ * unverified. A key that passes is saved to <config>/credentials. Keys are
+ * never copied into err. */
+
+#include "cst_agent.h"
+#include "cst_models.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define CST_PS_MAX_MODELS 512
+#define CST_PS_UNKNOWN (-1)   /* no such provider */
+#define CST_PS_NEED_KEY (-10) /* no key given, none saved or in the env */
+/* Non-negative results are flags: */
+#define CST_PS_UNVERIFIED 1 /* /models failed for another reason; err says */
+#define CST_PS_UNSAVED 2    /* the credentials file could not be written */
+
+typedef struct {
+  char id[128];
+  int api; /* listed by the provider's /models (else built-in only) */
+  cst_model_info_t info;
+} cst_ps_model_t;
+
+/* GET /models; a good answer refreshes models-cache/<provider>.json. The
+ * count, CST_MODELS_AUTH, or another negative value with err set. */
+int cst_ps_fetch(const char *provider, const char *api_base, const char *key,
+                 char ids[][128], int max, char *err, size_t err_sz);
+/* The cached list: count, or -1; *fresh when younger than the TTL. */
+int cst_ps_cached(const char *provider, char ids[][128], int max, int *fresh);
+
+/* Base URL for provider: cfg's own when it is the current provider (it may
+ * be overridden), else the preset's; "" when unknown. */
+const char *cst_ps_base(const cst_agent_config_t *cfg, const char *provider);
+
+/* The model picker list for cfg's provider: the API's ids (the cache while
+ * it is fresh, else a fetch; refresh always fetches) followed by the
+ * built-in names it did not list. *api_rc gets the fetch result (count,
+ * CST_MODELS_AUTH, -1 with err) or the cached count. Returns the count. */
+int cst_ps_models(const cst_agent_config_t *cfg, int refresh,
+                  cst_ps_model_t *out, int max, int *api_rc, char *err,
+                  size_t err_sz);
+
+/* Checks key against /models and saves it: CST_MODELS_AUTH (err set), or
+ * the CST_PS_* flags. When ids is given it gets the fetched list and *nids
+ * the raw fetch result. */
+int cst_ps_verify_save(const char *provider, const char *api_base,
+                       const char *key, char ids[][128], int max, int *nids,
+                       char *err, size_t err_sz);
+
+/* Point cfg at provider p with key: base URL, the remembered model (else
+ * the preset's), settings saved. */
+void cst_ps_apply_provider(cst_agent_config_t *cfg, const cst_provider_t *p,
+                           const char *key);
+
+/* The whole switch without a prompt: key NULL/"" uses the saved or env
+ * key (else CST_PS_NEED_KEY); a given key is verified and saved first. A
+ * preset with no model takes the first one the API lists. The CST_PS_*
+ * flags, or CST_PS_UNKNOWN / CST_PS_NEED_KEY / CST_MODELS_AUTH. */
+int cst_ps_switch_provider(cst_agent_config_t *cfg, const char *name,
+                           const char *key, char *err, size_t err_sz);
+
+/* Replace the current provider's key: verified, saved, put in cfg. Same
+ * results as cst_ps_verify_save. */
+int cst_ps_set_key(cst_agent_config_t *cfg, const char *key, char *err,
+                   size_t err_sz);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

include/cst_util.h

@@ -27,6 +27,17 @@ size_t cst_utf8_floor(const char *s, size_t n);
 char *cst_utf8_sanitize(const char *s, size_t n, size_t *out_len);
 int cst_utf8_valid(const char *s, size_t n);
 
+/* Re-chunks a byte stream so no piece ends inside a UTF-8 sequence: the
+ * incomplete tail (at most 3 bytes) waits in the hold for the next feed. */
+typedef struct {
+  char b[4];
+  size_t n;
+} cst_utf8_hold_t;
+/* Returns the complete prefix of hold+s[0..n) (malloc'd, NUL-terminated,
+ * *out_len may be 0), or NULL on OOM. */
+char *cst_utf8_hold_feed(cst_utf8_hold_t *h, const char *s, size_t n,
+                         size_t *out_len);
+
 /* Text files may start with a UTF-8 BOM (Windows editors): its length (0/3),
  * and an in-place strip of the NUL-terminated s (*len updated when given). */
 size_t cst_bom_len(const char *s, size_t n);

include/cst_version.h

@@ -2,8 +2,8 @@
 #define CST_VERSION_H
 
 #define CST_VERSION_MAJOR 0
-#define CST_VERSION_MINOR 7
-#define CST_VERSION_PATCH 1
+#define CST_VERSION_MINOR 8
+#define CST_VERSION_PATCH 3
 
 #define CST_VERSION_STR_(x) #x
 #define CST_VERSION_STR(x) CST_VERSION_STR_(x)

include/cst_web_md.h

@@ -0,0 +1,48 @@
+#ifndef CST_WEB_MD_H
+#define CST_WEB_MD_H
+
+/* Markdown for the WebUI: the same sashidown renderer the TUI uses, but
+ * HTML output with a strict allowlist.
+ *
+ * sashidown renders straight to HTML and escapes text, quotes and '&', so
+ * no raw markup can reach the browser. Two holes remain, and both matter
+ * because every byte rendered here comes from a model or a tool: a link's
+ * href keeps any scheme ("javascript:") and an image's src may point
+ * anywhere (a tracking pixel, or worse). cst_web_md_* therefore rewrites
+ * tags instead of passing them through: unknown tags are dropped, hrefs
+ * are limited to http/https/mailto/relative, and images to the session's
+ * own attachment endpoint.
+ *
+ * The streaming form renders confirmed HTML as the model talks, which is
+ * the CST property the project is built on: nothing is emitted before it
+ * is safe to emit. A tag split across two deltas is held back until its
+ * '>' arrives. */
+
+#include <stddef.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Called with chunks of sanitized HTML; return non-zero to stop. */
+typedef int (*cst_web_md_emit_fn)(const char *html, size_t len, void *ud);
+
+typedef struct cst_web_md cst_web_md_t;
+
+cst_web_md_t *cst_web_md_new(cst_web_md_emit_fn emit, void *ud);
+/* Feed model output. Returns 0, or -1 on OOM. */
+int cst_web_md_feed(cst_web_md_t *s, const char *data, size_t len);
+/* Flush the pending input and close the renderer. */
+int cst_web_md_finish(cst_web_md_t *s);
+void cst_web_md_free(cst_web_md_t *s);
+
+/* One-shot: markdown -> sanitized HTML (malloc'd, never NULL on success). */
+char *cst_web_md_html(const char *md, size_t len);
+/* The sanitizer alone (tests, and any HTML that did not come from here). */
+char *cst_web_md_sanitize(const char *html, size_t len);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* CST_WEB_MD_H */

include/cst_webui.h

@@ -0,0 +1,44 @@
+#ifndef CST_WEBUI_H
+#define CST_WEBUI_H
+
+/* The built-in WebUI: an HTTP + SSE front end for one agent process.
+ *
+ * It is a second implementation of the same interface the TUI uses
+ * (cst_agent_hooks_t), which is why the engine needs no changes: the turn
+ * runs on a worker thread and every live event goes out as a Server-Sent
+ * Event. The protocol is documented in docs/webui.md and is meant to be
+ * the contract a dedicated client speaks later.  */
+
+#include "cst_agent.h"
+
+#include <stddef.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* What --web listens on when --web-port/CST_WEB_PORT says nothing (the
+ * README and --help quote this number, so keep them in sync). */
+#define CST_WEBUI_DEFAULT_PORT 8787
+
+typedef struct {
+  const char *host;  /* NULL = 127.0.0.1; "0.0.0.0" = every interface */
+  int port;          /* 0 = pick a free port */
+  const char *token; /* NULL = generate one when needed */
+  const char *root;  /* serve the UI from this directory instead of the
+                        copy embedded in the binary (development) */
+  int open_browser;  /* best-effort xdg-open / open / start */
+  /* Prompt to run once at startup (--web -p ...); NULL = wait for the UI. */
+  const char *prompt;
+  int perm_timeout_s; /* how long a permission question waits (0 = 600) */
+} cst_webui_opts_t;
+
+/* Serve until interrupted (Ctrl+C) or the event stream asks to quit.
+ * Returns the process exit code. */
+int cst_webui_run(cst_agent_t *agent, const cst_webui_opts_t *opts);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* CST_WEBUI_H */

include/os_compat.h

@@ -114,9 +114,10 @@ int os_win_get_label(const char *path, char *out, size_t n);
 int os_clipboard_set(const char *utf8, size_t n);
 #endif
 
-/* Child processes get stdin=/dev/null, their own process group, and are
- * killed after the timeout (CST_SHELL_TIMEOUT seconds, default 120) or when
- * *flag becomes non-zero. Exit code 124 = timeout, 130 = cancelled. */
+/* Child processes get stdin=/dev/null and their own process group. They run
+ * with no time limit unless one is asked for (CST_SHELL_TIMEOUT seconds, or
+ * os_set_run_timeout()); either way a non-zero *flag cancels them.
+ * Exit code 124 = killed by that timeout, 130 = cancelled. */
 void os_set_cancel_flag(volatile int *flag);
 volatile int *os_get_cancel_flag(void);
 void os_set_run_timeout(int seconds); /* this thread; 0 = default */

scripts/e2e_webui.sh

@@ -0,0 +1,290 @@
+#!/bin/sh
+# End-to-end check of the WebUI without a real model: a recorded SSE stream
+# (tests/fixtures/sse) drives a full turn, and the events the browser would
+# see are read back from /api/events with curl.
+#
+#   scripts/e2e_webui.sh [replay-dir] [protocol]
+#
+# Needs: curl. python3 only for the JSON field checks (they are skipped
+# without it). Exit code 0 = every assertion passed.
+set -eu
+ROOT="$(cd "$(dirname "$0")/.." && pwd)"
+REPLAY="${1:-$ROOT/tests/fixtures/sse/deepseek-native-eos}"
+PROTO="${2:-native-eos}"
+BIN="${CST_AGENT:-$ROOT/cst-agent}"
+[ -x "$BIN" ] || { echo "build cst-agent first ($BIN)"; exit 1; }
+
+WORK="$(mktemp -d "${TMPDIR:-/tmp}/cst-webui-XXXXXX")" 2>/dev/null || {
+  WORK="$ROOT/.build-tmp/e2e-webui"; rm -rf "$WORK"; mkdir -p "$WORK"; }
+EVENTS="$WORK/events.txt"
+LOG="$WORK/server.log"
+cleanup() {
+  [ -n "${SRV_PID:-}" ] && kill "$SRV_PID" 2>/dev/null || true
+  rm -rf "$WORK"
+}
+trap cleanup EXIT INT TERM
+
+echo "== start cst-agent --web (replay $REPLAY)"
+# --yolo: the recorded stream calls write tools, and a permission question
+# would wait for a browser nobody is driving here.
+CST_REPLAY_SSE="$REPLAY" \
+CST_REPLAY_CHUNK=512 \
+"$BIN" --web --web-port 0 --workspace "$WORK" --new --api-key sk-e2e \
+  --protocol "$PROTO" --tools default --yolo > "$LOG" 2>&1 &
+SRV_PID=$!
+
+# The port is picked by the OS; the server prints the URL it bound.
+i=0
+PORT=""
+while [ $i -lt 50 ]; do
+  PORT="$(sed -n 's#^webui: http://127.0.0.1:\([0-9]*\)/.*#\1#p' "$LOG" | head -1)"
+  [ -n "$PORT" ] && break
+  i=$((i + 1))
+  sleep 0.2
+done
+[ -n "$PORT" ] || { echo "FAIL: the server did not report a port"; cat "$LOG"; exit 1; }
+BASE="http://127.0.0.1:$PORT"
+echo "   port $PORT"
+
+fail() { echo "FAIL: $1"; echo "--- events"; cat "$EVENTS" 2>/dev/null | head -40; exit 1; }
+# SIGTERM must shut the server down cleanly, even mid-turn (128+ = a crash)
+stop_server() {
+  kill "$1" 2>/dev/null || return 0
+  n=0
+  while kill -0 "$1" 2>/dev/null && [ $n -lt 200 ]; do n=$((n + 1)); sleep 0.2; done
+  if kill -0 "$1" 2>/dev/null; then
+    kill -9 "$1" 2>/dev/null || true
+    fail "$2 hung on SIGTERM"
+  fi
+  st=0
+  wait "$1" 2>/dev/null || st=$?
+  [ "$st" -lt 128 ] || fail "$2 died with status $st on SIGTERM"
+}
+
+# 1. the shell page and the state endpoint
+curl -fsS --max-time 5 "$BASE/" | grep -q "<title>CST-Agent" || fail "GET / is not the shell page"
+curl -fsS --max-time 5 "$BASE/api/state" > "$WORK/state.json" || fail "GET /api/state"
+grep -q '"protocol":"cst-webui/1"' "$WORK/state.json" || fail "state has no protocol tag"
+grep -q '"session":' "$WORK/state.json" || fail "state has no session id"
+
+# 2. an unknown route is a 404, an oversized body a 413
+code="$(curl -s -o /dev/null -w '%{http_code}' --max-time 5 "$BASE/api/does-not-exist")"
+[ "$code" = "404" ] || fail "unknown route returned $code"
+code="$(head -c 2000000 /dev/zero | tr '\0' 'a' |
+        curl -s -o /dev/null -w '%{http_code}' --max-time 20 \
+             -X POST --data-binary @- "$BASE/api/turn")"
+[ "$code" = "413" ] || fail "2 MB body returned $code (want 413)"
+
+# 3. stream the events while a recorded turn runs
+curl -sN --max-time 120 "$BASE/api/events" > "$EVENTS" &
+CURL_PID=$!
+sleep 0.5
+code="$(curl -s -o "$WORK/turn.json" -w '%{http_code}' --max-time 5 \
+        -X POST -H 'Content-Type: application/json' \
+        --data '{"text":"e2e: what is in the workspace?"}' "$BASE/api/turn")"
+[ "$code" = "202" ] || fail "POST /api/turn returned $code"
+
+i=0
+while [ $i -lt 450 ]; do
+  grep -q "event: turn_end" "$EVENTS" && break
+  i=$((i + 1))
+  sleep 0.2
+done
+kill "$CURL_PID" 2>/dev/null || true
+grep -q "event: turn_end" "$EVENTS" || fail "the turn never finished"
+grep -q "event: assistant_html" "$EVENTS" || fail "no assistant markdown was streamed"
+grep -q "event: tool" "$EVENTS" || fail "no tool event was streamed"
+grep -q '"rc":0' "$EVENTS" || fail "the turn did not end with rc=0"
+
+# 4. the session on disk matches what the UI would draw
+curl -fsS --max-time 5 "$BASE/api/session" > "$WORK/session.json" || fail "GET /api/session"
+grep -q '"role":"assistant"' "$WORK/session.json" || fail "no assistant message stored"
+curl -fsS --max-time 5 "$BASE/api/sessions" | grep -q '"sessions":' || fail "GET /api/sessions"
+
+# 5. a second turn while the first is running is refused with 409
+curl -s -o /dev/null -w '' --max-time 5 -X POST -H 'Content-Type: application/json' \
+     --data '{"text":"x"}' "$BASE/api/turn" || true
+
+# 6. asking to stop is accepted
+curl -fsS --max-time 5 -X POST -H 'Content-Type: application/json' \
+     --data '{}' "$BASE/api/cancel" > /dev/null || fail "POST /api/cancel"
+
+# 6b. the write endpoints: config, undo, compact and image upload
+curl -fsS --max-time 5 -X POST -H 'Content-Type: application/json' \
+     --data '{"approval":"ask","plan_mode":false}' "$BASE/api/config" |
+  grep -q '"ok":true' || fail "POST /api/config"
+curl -fsS --max-time 5 "$BASE/api/state" | grep -q '"approval":"ask"' ||
+  fail "the config change is not in the state"
+code="$(curl -s -o "$WORK/undo.txt" -w '%{http_code}' --max-time 5 \
+        -X POST "$BASE/api/undo")"
+case "$code" in
+  200) grep -q . "$WORK/undo.txt" || fail "an undo without a report" ;;
+  409) ;;   # nothing to undo in this run
+  *) fail "POST /api/undo returned $code" ;;
+esac
+code="$(curl -s -o /dev/null -w '%{http_code}' --max-time 30 \
+        -X POST "$BASE/api/compact")"
+# The recorded stream is spent by now, so a model failure is a valid answer.
+case "$code" in 200 | 500) ;; *) fail "POST /api/compact returned $code" ;; esac
+
+if command -v base64 > /dev/null; then
+  # a real 1x1 PNG: the server sniffs the content, not the extension
+  # (kept with a .bin suffix so the find below only sees the stored copy)
+  echo 'iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mP8z8DwHwAFAAH/q842iQAAAABJRU5ErkJggg==' |
+    base64 -d > "$WORK/tiny.bin" 2> /dev/null || : > "$WORK/tiny.bin"
+  code="$(curl -s -o /dev/null -w '%{http_code}' --max-time 10 -X POST \
+          --data-binary @"$WORK/tiny.bin" "$BASE/api/attach?name=e2e.png")"
+  [ "$code" = "200" ] || fail "POST /api/attach returned $code"
+  curl -fsS --max-time 5 "$BASE/api/state" |
+    grep -q '"pending_images":\["' || fail "the upload is not pending"
+  # the stored name is the content hash, so ask the disk for it
+  STORED="$(find "$WORK" -name '*.png' -type f 2>/dev/null | head -1)"
+  [ -n "$STORED" ] || fail "the upload did not reach the session directory"
+  REL="${STORED##*/}"
+  curl -fsS --max-time 5 "$BASE/api/attach/$REL" > "$WORK/back.png" ||
+    fail "GET /api/attach/$REL"
+  cmp -s "$WORK/tiny.bin" "$WORK/back.png" ||
+    fail "the image did not round trip byte for byte"
+  code="$(curl -s -o /dev/null -w '%{http_code}' --max-time 5 \
+          "$BASE/api/attach/..%2f..%2fetc%2fpasswd")"
+  case "$code" in 400 | 404) ;; *) fail "a traversal in attach returned $code" ;; esac
+  curl -fsS --max-time 5 -X DELETE "$BASE/api/attach" | grep -q '"ok":true' ||
+    fail "DELETE /api/attach"
+  curl -fsS --max-time 5 "$BASE/api/state" |
+    grep -q '"pending_images":\[\]' || fail "the pending image was not cleared"
+else
+  echo "   (no base64 tool: the image upload check is skipped)"
+fi
+
+# 6c. a request with a foreign Host header is refused while no token is set:
+# that is the DNS-rebinding guard.
+code="$(curl -s -o /dev/null -w '%{http_code}' --max-time 5 \
+        -H 'Host: evil.example' "$BASE/api/state")"
+[ "$code" = "401" ] || fail "a foreign Host header returned $code (want 401)"
+
+if command -v python3 > /dev/null; then
+  python3 - "$WORK/session.json" <<'PY' || fail "the session JSON is not valid JSON / is empty"
+import json, sys
+d = json.load(open(sys.argv[1]))
+assert isinstance(d.get("messages"), list) and d["messages"], d
+PY
+fi
+
+# 7. the interactive approval path: a fresh server in ask mode must raise a
+# permission event, and answering it must let the turn continue.
+echo "== permission round trip (-A ask)"
+stop_server "$SRV_PID" "the first server"   # its replay is spent
+rm -rf "$WORK/pw"; mkdir -p "$WORK/pw"
+P_EVENTS="$WORK/permission-events.txt"
+P_LOG="$WORK/permission-server.log"   # its own log: the port parse must not
+                                      # pick up the first server's URL
+CST_REPLAY_SSE="$REPLAY" CST_REPLAY_CHUNK=512 \
+"$BIN" --web --web-port 0 --workspace "$WORK/pw" --new --api-key sk-e2e \
+  --protocol "$PROTO" --tools default --approval ask --confirm all \
+  > "$P_LOG" 2>&1 &
+P_PID=$!
+i=0
+P_PORT=""
+while [ $i -lt 50 ]; do
+  P_PORT="$(sed -n 's#^webui: http://127.0.0.1:\([0-9]*\)/.*#\1#p' "$P_LOG" | tail -1)"
+  [ -n "$P_PORT" ] && break
+  i=$((i + 1)); sleep 0.2
+done
+[ -n "$P_PORT" ] || fail "the ask-mode server did not start (see $P_LOG)"
+P_BASE="http://127.0.0.1:$P_PORT"
+curl -sN --max-time 60 "$P_BASE/api/events" > "$P_EVENTS" &
+P_CURL=$!
+sleep 0.5
+curl -s -o /dev/null --max-time 5 -X POST -H 'Content-Type: application/json' \
+     --data '{"text":"e2e: ask me first"}' "$P_BASE/api/turn"
+i=0
+while [ $i -lt 300 ]; do
+  grep -q "event: permission" "$P_EVENTS" && break
+  i=$((i + 1)); sleep 0.2
+done
+if [ $i -ge 300 ]; then
+  kill "$P_PID" 2>/dev/null || true
+  kill "$P_CURL" 2>/dev/null || true
+  echo "   (no permission question within 60 s - skipped; $(grep -c '^event: ' "$P_EVENTS") events seen)"
+else
+  PID_REQ="$(sed -n 's/.*"id":\([0-9]*\),"tool".*/\1/p' "$P_EVENTS" | head -1)"
+  [ -n "$PID_REQ" ] || fail "a permission event without an id"
+  # Answer while the server is still waiting for it.
+  curl -fsS --max-time 5 -X POST -H 'Content-Type: application/json' \
+       --data "{\"id\":$PID_REQ,\"answer\":1}" "$P_BASE/api/permission" > /dev/null ||
+    fail "POST /api/permission was refused"
+  i=0
+  while [ $i -lt 150 ]; do
+    grep -q "event: permission_done" "$P_EVENTS" && break
+    i=$((i + 1)); sleep 0.2
+  done
+  grep -q "event: permission_done" "$P_EVENTS" || fail "the answer was not acknowledged"
+  stop_server "$P_PID" "the ask-mode server"
+  kill "$P_CURL" 2>/dev/null || true
+  echo "   answered request $PID_REQ with 'allow once'"
+fi
+
+# 8. token auth, --web-dev, and a clean shutdown
+echo "== token auth and --web-dev"
+T_LOG="$WORK/token-server.log"
+CST_REPLAY_SSE="$REPLAY" \
+"$BIN" --web --web-port 0 --workspace "$WORK/pw" --new --api-key sk-e2e \
+  --protocol "$PROTO" --tools default --yolo --web-dev "$ROOT/web" \
+  --web-token tok-e2e > "$T_LOG" 2>&1 &
+T_PID=$!
+i=0
+T_PORT=""
+while [ $i -lt 50 ]; do
+  T_PORT="$(sed -n 's#^webui: http://127.0.0.1:\([0-9]*\)/.*#\1#p' "$T_LOG" | tail -1)"
+  [ -n "$T_PORT" ] && break
+  i=$((i + 1)); sleep 0.2
+done
+[ -n "$T_PORT" ] || fail "the token server did not start (see $T_LOG)"
+T_BASE="http://127.0.0.1:$T_PORT"
+
+code="$(curl -s -o /dev/null -w '%{http_code}' --max-time 5 "$T_BASE/api/state")"
+[ "$code" = "401" ] || fail "a token-protected server answered $code without one"
+code="$(curl -s -o /dev/null -w '%{http_code}' --max-time 5 \
+        -H 'Authorization: Bearer tok-e2e' "$T_BASE/api/state")"
+[ "$code" = "200" ] || fail "the Bearer token was rejected ($code)"
+code="$(curl -s -o /dev/null -w '%{http_code}' --max-time 5 \
+        "$T_BASE/api/state?token=tok-e2e")"
+[ "$code" = "200" ] || fail "the query token was rejected ($code)"
+code="$(curl -s -o /dev/null -w '%{http_code}' --max-time 5 \
+        -H 'Cookie: a=b; cst_token=tok-e2e; z=y' "$T_BASE/api/state")"
+[ "$code" = "200" ] || fail "the cookie token was rejected ($code)"
+code="$(curl -s -o /dev/null -w '%{http_code}' --max-time 5 \
+        -H 'Authorization: Bearer tok-e2' "$T_BASE/api/state")"
+[ "$code" = "401" ] || fail "a wrong token was accepted ($code)"
+
+# --web-dev serves straight from the directory, so the bytes must match it
+curl -fsS --max-time 5 -H 'Authorization: Bearer tok-e2e' "$T_BASE/app.js" \
+  > "$WORK/app.js" || fail "GET /app.js from --web-dev"
+cmp -s "$WORK/app.js" "$ROOT/web/app.js" ||
+  fail "--web-dev did not serve web/app.js from disk"
+
+# POST /api/quit must return and then take the server down on its own
+curl -fsS --max-time 5 -X POST -H 'Authorization: Bearer tok-e2e' \
+     "$T_BASE/api/quit" | grep -q '"ok":true' || fail "POST /api/quit"
+i=0
+while [ $i -lt 50 ] && kill -0 "$T_PID" 2> /dev/null; do
+  i=$((i + 1)); sleep 0.2
+done
+kill -0 "$T_PID" 2> /dev/null && fail "the server ignored /api/quit"
+wait "$T_PID" 2>/dev/null || true
+
+# 9. ask_user waits for the answer, and the HTTP hardening checks (both
+# need a scripted model: python3)
+if command -v python3 > /dev/null; then
+  echo "== ask_user round trip"
+  CST_AGENT="$BIN" python3 "$ROOT/scripts/e2e_webui_ask.py" > "$WORK/ask.txt" 2>&1 ||
+    { cat "$WORK/ask.txt"; fail "the ask_user round trip"; }
+  echo "== HTTP hardening"
+  CST_AGENT="$BIN" python3 "$ROOT/scripts/e2e_webui_http.py" > "$WORK/http.txt" 2>&1 ||
+    { cat "$WORK/http.txt"; fail "the HTTP hardening checks"; }
+  echo "== settings and streaming"
+  CST_AGENT="$BIN" python3 "$ROOT/scripts/e2e_webui_settings.py" > "$WORK/settings.txt" 2>&1 ||
+    { cat "$WORK/settings.txt"; fail "the settings checks"; }
+fi
+
+echo "== OK: (events: $(grep -c '^event: ' "$EVENTS"))"

scripts/e2e_webui_ask.py

@@ -0,0 +1,164 @@
+"""WebUI ask_user round trip against a fake OpenAI-compatible API.
+
+  python3 scripts/e2e_webui_ask.py      (CST_AGENT= picks the binary)
+
+Checks: the turn really waits for the answer, a page opened while the
+question is open still receives it, and Last-Event-ID resumes the stream.
+"""
+import json, os, shutil, subprocess, sys, tempfile, threading, time, urllib.request
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+
+BIN = os.environ.get("CST_AGENT", os.path.join(os.path.dirname(os.path.abspath(__file__)), "..", "cst-agent"))
+REQS = []
+
+
+class H(BaseHTTPRequestHandler):
+    def log_message(self, *a):
+        pass
+
+    def do_GET(self):
+        body = json.dumps({"data": [{"id": "deepseek-flash"}]}).encode()
+        self.send_response(200)
+        self.send_header("Content-Type", "application/json")
+        self.send_header("Content-Length", str(len(body)))
+        self.end_headers()
+        self.wfile.write(body)
+
+    def do_POST(self):
+        n = int(self.headers.get("Content-Length", 0))
+        REQS.append(json.loads(self.rfile.read(n)))
+        self.send_response(200)
+        self.send_header("Content-Type", "text/event-stream")
+        self.end_headers()
+        base = {"id": "x", "object": "chat.completion.chunk", "model": "deepseek-flash"}
+
+        def ev(o):
+            self.wfile.write(b"data: " + json.dumps({**base, **o}).encode() + b"\n\n")
+            self.wfile.flush()
+        if len(REQS) == 1:
+            args = json.dumps({"question": "Pick a colour?", "options": "red\nblue"})
+            ev({"choices": [{"index": 0, "delta": {"role": "assistant", "tool_calls": [
+                {"index": 0, "id": "call_1", "type": "function",
+                 "function": {"name": "ask_user", "arguments": args}}]}}]})
+            ev({"choices": [{"index": 0, "delta": {}, "finish_reason": "tool_calls"}]})
+        else:
+            ev({"choices": [{"index": 0, "delta": {"role": "assistant", "content": "done"}}]})
+            ev({"choices": [{"index": 0, "delta": {}, "finish_reason": "stop"}]})
+        self.wfile.write(b"data: [DONE]\n\n")
+        self.wfile.flush()
+
+
+fails = 0
+
+
+def check(ok, msg):
+    global fails
+    print(("PASS " if ok else "FAIL ") + msg)
+    fails += 0 if ok else 1
+
+
+class Sse:
+    def __init__(self, base, last_id=None):
+        self.lines = []
+        req = urllib.request.Request(base + "/api/events")
+        if last_id is not None:
+            req.add_header("Last-Event-ID", str(last_id))
+        self.r = urllib.request.urlopen(req, timeout=60)
+        threading.Thread(target=self.pump, daemon=True).start()
+
+    def pump(self):
+        try:
+            for raw in self.r:
+                self.lines.append(raw.decode())
+        except Exception:
+            pass
+
+    def events(self, name):
+        out = []
+        for i, l in enumerate(self.lines):
+            if l.strip() == "event: " + name and i + 1 < len(self.lines):
+                out.append(json.loads(self.lines[i + 1][6:]))
+        return out
+
+    def ids(self):
+        return [int(l[4:]) for l in self.lines if l.startswith("id: ")]
+
+    def close(self):
+        try:
+            self.r.close()
+        except Exception:
+            pass
+
+
+def wait(fn, t=15):
+    end = time.time() + t
+    while time.time() < end:
+        v = fn()
+        if v:
+            return v
+        time.sleep(0.1)
+    return None
+
+
+def post(base, path, obj):
+    req = urllib.request.Request(base + path, data=json.dumps(obj).encode(),
+                                 headers={"Content-Type": "application/json"})
+    try:
+        with urllib.request.urlopen(req, timeout=5) as r:
+            return r.status
+    except urllib.error.HTTPError as e:
+        return e.code
+
+
+api = ThreadingHTTPServer(("127.0.0.1", 0), H)
+threading.Thread(target=api.serve_forever, daemon=True).start()
+ws = tempfile.mkdtemp(prefix="cst-ask-")
+log = open(ws + "/server.log", "w+")
+p = subprocess.Popen([BIN, "--web", "--web-port", "0", "--workspace", ws, "--new",
+                      "--api-key", "sk-x", "--provider", "deepseek", "--model", "deepseek-flash",
+                      "--api-base", "http://127.0.0.1:%d/v1" % api.server_address[1],
+                      "--protocol", "native-eos", "--approval", "auto"],
+                     stdout=log, stderr=subprocess.STDOUT)
+try:
+    def port():
+        log.seek(0)
+        for l in log.read().splitlines():
+            if l.startswith("webui: http://127.0.0.1:"):
+                return l.split(":")[3].split("/")[0]
+    pt = wait(port)
+    base = "http://127.0.0.1:%s" % pt
+    a = Sse(base)
+    time.sleep(0.3)
+    check(post(base, "/api/turn", {"text": "ask me"}) == 202, "turn accepted")
+    ask = wait(lambda: a.events("ask_user"))
+    check(bool(ask), "ask_user event raised")
+    time.sleep(1.5)
+    check(len(REQS) == 1, "the turn waits for the answer (requests so far: %d)" % len(REQS))
+    # a page opened now must still see the open question
+    b = Sse(base)
+    check(bool(wait(lambda: b.events("ask_user"), 5)), "a late page gets the open question")
+    # reconnect with Last-Event-ID: nothing before that id is resent
+    last = a.ids()[-1]
+    c = Sse(base, last_id=last - 1)
+    got = wait(lambda: c.ids(), 5)
+    check(bool(got) and got[0] == last, "Last-Event-ID resumes after that id (%s, want %d)" % (got[:1] if got else got, last))
+    if ask:
+        check(post(base, "/api/ask_user", {"id": ask[0]["id"], "text": "blue"}) == 200, "answer accepted")
+    end = wait(lambda: a.events("turn_end"), 20)
+    check(bool(end) and end[0].get("rc") == 0, "turn ended rc=0")
+    tool_msgs = [m for m in (REQS[1]["messages"] if len(REQS) > 1 else []) if m.get("role") == "tool"]
+    check(bool(tool_msgs) and "blue" in json.dumps(tool_msgs), "the answer reached the model: %s" % json.dumps(tool_msgs)[:200])
+    for s in (a, b, c):
+        s.close()
+finally:
+    p.terminate()
+    try:
+        p.wait(20)
+    except subprocess.TimeoutExpired:
+        p.kill()
+        check(False, "server hung on SIGTERM")
+    check(p.returncode is not None and p.returncode < 128 and p.returncode >= 0, "clean exit (%s)" % p.returncode)
+    log.close()
+    shutil.rmtree(ws, ignore_errors=True)
+print("FAILED:", fails)
+sys.exit(1 if fails else 0)

scripts/e2e_webui_http.py

@@ -0,0 +1,234 @@
+"""WebUI HTTP hardening checks over raw sockets, against a scripted model.
+
+  python3 scripts/e2e_webui_http.py      (CST_AGENT= picks the binary)
+
+Cross-site requests, request parsing (Content-Length, NUL, long lines,
+keep-alive), auth before the body, changes refused while a turn runs,
+cancel waking a pending question, token mode and shutdown.
+"""
+import json, os, re, shutil, socket, subprocess, sys, tempfile, threading, time
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+
+BIN = os.environ.get("CST_AGENT", os.path.join(os.path.dirname(os.path.abspath(__file__)), "..", "cst-agent"))
+REQS = []
+
+
+class Api(BaseHTTPRequestHandler):
+    """First call: ask_user. Later calls: a plain answer."""
+
+    def log_message(self, *a):
+        pass
+
+    def do_GET(self):
+        body = json.dumps({"data": [{"id": "deepseek-flash"}]}).encode()
+        self.send_response(200)
+        self.send_header("Content-Type", "application/json")
+        self.send_header("Content-Length", str(len(body)))
+        self.end_headers()
+        self.wfile.write(body)
+
+    def do_POST(self):
+        n = int(self.headers.get("Content-Length", 0))
+        REQS.append(json.loads(self.rfile.read(n)))
+        self.send_response(200)
+        self.send_header("Content-Type", "text/event-stream")
+        self.end_headers()
+        base = {"id": "x", "object": "chat.completion.chunk", "model": "deepseek-flash"}
+
+        def ev(o):
+            self.wfile.write(b"data: " + json.dumps({**base, **o}).encode() + b"\n\n")
+            self.wfile.flush()
+        if len(REQS) == 1:
+            args = json.dumps({"question": "Pick?", "options": "a\nb"})
+            ev({"choices": [{"index": 0, "delta": {"role": "assistant", "tool_calls": [
+                {"index": 0, "id": "call_1", "type": "function",
+                 "function": {"name": "ask_user", "arguments": args}}]}}]})
+            ev({"choices": [{"index": 0, "delta": {}, "finish_reason": "tool_calls"}]})
+        else:
+            ev({"choices": [{"index": 0, "delta": {"role": "assistant", "content": "ok"}}]})
+            ev({"choices": [{"index": 0, "delta": {}, "finish_reason": "stop"}]})
+        self.wfile.write(b"data: [DONE]\n\n")
+        self.wfile.flush()
+
+
+fails = 0
+
+
+def check(ok, msg):
+    global fails
+    print(("PASS " if ok else "FAIL ") + msg)
+    fails += 0 if ok else 1
+
+
+def raw(port, data, timeout=5, read_all=True):
+    """Send bytes, return everything read until EOF/timeout."""
+    s = socket.create_connection(("127.0.0.1", port), timeout=timeout)
+    s.sendall(data)
+    out = b""
+    try:
+        while True:
+            b = s.recv(65536)
+            if not b:
+                break
+            out += b
+            if not read_all:
+                break
+    except socket.timeout:
+        pass
+    s.close()
+    return out.decode("latin1")
+
+
+def status(resp):
+    # a pipelined response starts right after the previous body, not on a new line
+    return [int(m) for m in re.findall(r"HTTP/1\.1 (\d{3}) ", resp)]
+
+
+def req(method, path, headers=None, body=b"", host=None):
+    h = {"Host": host or "127.0.0.1:%d" % PORT, "Connection": "close"}
+    h.update(headers or {})
+    if body or method == "POST":
+        h["Content-Length"] = str(len(body))
+    head = "%s %s HTTP/1.1\r\n" % (method, path) + "".join("%s: %s\r\n" % kv for kv in h.items()) + "\r\n"
+    return head.encode() + body
+
+
+def start(ws, *extra):
+    log = open(os.path.join(ws, "server.log"), "w+")
+    p = subprocess.Popen([BIN, "--web", "--web-port", "0", "--workspace", ws, "--new",
+                          "--api-key", "sk-x", "--provider", "deepseek", "--model", "deepseek-flash",
+                          "--api-base", "http://127.0.0.1:%d/v1" % api.server_address[1],
+                          "--protocol", "native-eos", "--approval", "auto", *extra],
+                         stdout=log, stderr=subprocess.STDOUT)
+    end = time.time() + 10
+    while time.time() < end:
+        log.seek(0)
+        for l in log.read().splitlines():
+            if l.startswith("webui: http://127.0.0.1:"):
+                return p, int(l.split(":")[3].split("/")[0]), log
+        time.sleep(0.1)
+    raise SystemExit("server did not start")
+
+
+def stop(p, name):
+    p.terminate()
+    try:
+        p.wait(20)
+    except subprocess.TimeoutExpired:
+        p.kill()
+        check(False, name + " hung on SIGTERM")
+        return
+    check(0 <= p.returncode < 128, "%s exits cleanly (%s)" % (name, p.returncode))
+
+
+api = ThreadingHTTPServer(("127.0.0.1", 0), Api)
+threading.Thread(target=api.serve_forever, daemon=True).start()
+root = tempfile.mkdtemp(prefix="cst-http-")
+try:
+    ws = os.path.join(root, "a")
+    os.makedirs(ws)
+    p, PORT, log = start(ws)
+    me = "http://127.0.0.1:%d" % PORT
+    cfg = b'{"approval":"ask"}'
+    # 1. cross-site requests
+    r = raw(PORT, req("POST", "/api/config", {"Origin": "http://evil.example", "Content-Type": "text/plain"}, cfg))
+    check(status(r)[:1] == [401], "a cross-site POST is refused (%s)" % status(r))
+    r = raw(PORT, req("POST", "/api/config", {"Origin": "null"}, cfg))
+    check(status(r)[:1] == [401], "Origin: null is refused")
+    r = raw(PORT, req("POST", "/api/config", {"Origin": me, "Content-Type": "application/json"}, b'{"approval":"auto"}'))
+    check(status(r)[:1] == [200], "our own page may POST (%s)" % status(r))
+    r = raw(PORT, req("POST", "/api/config", {}, b'{"approval":"auto"}'))
+    check(status(r)[:1] == [200], "a client without Origin (curl) may POST")
+    r = raw(PORT, req("GET", "/api/attach/img-0000.png", host="evil.example"))
+    check(status(r)[:1] == [401], "session images need the Host check too (%s)" % status(r))
+    r = raw(PORT, req("GET", "/app.js"))
+    check(status(r)[:1] == [200], "assets are served")
+    # 2. request parsing
+    r = raw(PORT, req("GET", "/api/state%00x"))
+    check(status(r)[:1] == [400], "an encoded NUL in the path is a 400 (%s)" % status(r))
+    smuggle = b"GET /api/sessions HTTP/1.1\r\nHost: 127.0.0.1\r\n\r\n"
+    r = raw(PORT, ("POST /api/cancel HTTP/1.1\r\nHost: 127.0.0.1:%d\r\nContent-Length: abc\r\n\r\n" % PORT).encode() + smuggle)
+    check(status(r) == [400], "Content-Length: abc is a 400, nothing smuggled (%s)" % status(r))
+    r = raw(PORT, ("POST /api/cancel HTTP/1.1\r\nHost: 127.0.0.1:%d\r\nContent-Length: 0\r\nContent-Length: %d\r\n\r\n"
+                   % (PORT, len(smuggle))).encode() + smuggle)
+    check(status(r) == [400], "two different Content-Lengths are a 400 (%s)" % status(r))
+    r = raw(PORT, ("GET /%s HTTP/1.1\r\nHost: x\r\n\r\n" % ("a" * 9000)).encode())
+    check(status(r)[:1] == [400], "an over-long request line is a 400 (%s)" % status(r))
+    r = raw(PORT, req("GET", "/api/state?x=http://a/b"))
+    check(status(r)[:1] == [200], "'://' in the query leaves the path alone (%s)" % status(r))
+    r = raw(PORT, req("GET", "/api/state", {"Connection": "Close"}), timeout=3)
+    check("Connection: close" in r, "Connection: Close (any case) closes")
+    ka = ("GET /api/state HTTP/1.1\r\nHost: 127.0.0.1:%d\r\n\r\n" % PORT).encode()
+    big = ("POST /api/turn HTTP/1.1\r\nHost: 127.0.0.1:%d\r\nContent-Length: 5000000\r\n\r\n" % PORT).encode()
+    r = raw(PORT, ka + big)
+    check(status(r) == [200, 413], "a 413 after a keep-alive request still arrives (%s)" % status(r))
+    # 3. a pending question: changes are refused, approval may change, cancel wakes it
+    events = []
+
+    def sse():
+        s = socket.create_connection(("127.0.0.1", PORT))
+        s.sendall(("GET /api/events HTTP/1.1\r\nHost: 127.0.0.1:%d\r\n\r\n" % PORT).encode())
+        buf = b""
+        s.settimeout(60)
+        try:
+            while True:
+                b = s.recv(65536)
+                if not b:
+                    break
+                buf += b
+                events[:] = [l[7:] for l in buf.decode("latin1").split("\n") if l.startswith("event: ")]
+        except Exception:
+            pass
+    threading.Thread(target=sse, daemon=True).start()
+    time.sleep(0.4)
+    r = raw(PORT, req("POST", "/api/turn", {"Content-Type": "application/json"}, b'{"text":"go"}'))
+    check(status(r)[:1] == [202], "turn accepted")
+    end = time.time() + 15
+    while time.time() < end and "ask_user" not in events:
+        time.sleep(0.1)
+    check("ask_user" in events, "the question is open")
+    for path, body in (("/api/sessions/new", b""), ("/api/workspace", json.dumps({"path": root}).encode()),
+                       ("/api/compact", b""), ("/api/undo", b""), ("/api/config", b'{"model":"x"}')):
+        r = raw(PORT, req("POST", path, {"Content-Type": "application/json"}, body))
+        check(status(r)[:1] == [409], "%s is refused mid-turn (%s)" % (path, status(r)))
+    r = raw(PORT, req("POST", "/api/config", {"Content-Type": "application/json"}, b'{"approval":"yolo"}'))
+    check(status(r)[:1] == [200], "approval may change mid-turn")
+    r = raw(PORT, req("POST", "/api/cancel", {"Content-Type": "application/json"}, b"{}"))
+    end = time.time() + 10
+    while time.time() < end and "turn_end" not in events:
+        time.sleep(0.1)
+    check("turn_end" in events, "cancel ends a turn parked on a question")
+    r = raw(PORT, req("POST", "/api/sessions/new", {"Content-Type": "application/json"}))
+    check(status(r)[:1] == [200], "after the turn, a new session is fine (%s)" % status(r))
+    stop(p, "the server")
+
+    # 4. token mode: the page loads, the API and images do not without it
+    ws2 = os.path.join(root, "b")
+    os.makedirs(ws2)
+    p, PORT, log = start(ws2, "--web-token", "tok-x")
+    r = raw(PORT, req("GET", "/app.js"))
+    check(status(r)[:1] == [200], "token mode: app.js loads without the token (%s)" % status(r))
+    r = raw(PORT, req("GET", "/api/state"))
+    check(status(r)[:1] == [401], "token mode: the API needs it")
+    r = raw(PORT, req("GET", "/api/attach/img-0000.png"))
+    check(status(r)[:1] == [401], "token mode: images need it")
+    r = raw(PORT, req("GET", "/api/state", {"Cookie": "cst_token=tok-x"}))
+    check(status(r)[:1] == [200], "token mode: the cookie works")
+    t0 = time.time()
+    r = raw(PORT, ("POST /api/attach HTTP/1.1\r\nHost: 127.0.0.1:%d\r\nContent-Length: 33554431\r\n"
+                   "Expect: 100-continue\r\n\r\n" % PORT).encode(), timeout=5)
+    check(status(r) == [401] and time.time() - t0 < 3,
+          "no token, no body read: 401 at once (%s, %.1fs)" % (status(r), time.time() - t0))
+    stop(p, "the token server")
+
+    # 5. option checks
+    rc = subprocess.run([BIN, "--web", "--web-port", "70000", "--workspace", ws2],
+                        capture_output=True, timeout=10).returncode
+    check(rc != 0, "--web-port 70000 is refused")
+    rc = subprocess.run([BIN, "--web", "--web-token", "x" * 80, "--web-port", "0", "--workspace", ws2,
+                         "--api-key", "k"], capture_output=True, timeout=10).returncode
+    check(rc != 0, "an over-long --web-token is refused")
+finally:
+    shutil.rmtree(root, ignore_errors=True)
+print("FAILED:", fails)
+sys.exit(1 if fails else 0)

scripts/e2e_webui_settings.py

@@ -0,0 +1,313 @@
+"""WebUI settings and streaming against fake OpenAI-compatible APIs.
+
+  python3 scripts/e2e_webui_settings.py      (CST_AGENT= picks the binary)
+
+Checks: Chinese reasoning streams as whole characters in few events; the
+provider list, the model list with capabilities, config validation and
+persistence; a wrong key is rejected, a good one switches the provider
+(requests then go to the new base URL) and is saved; and all of it
+survives a restart. HOME and XDG_CONFIG_HOME point at a temp dir, so no
+real credentials or settings are read or written.
+"""
+import json, os, shutil, stat, subprocess, sys, tempfile, threading, time, urllib.request
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+
+BIN = os.environ.get("CST_AGENT", os.path.join(os.path.dirname(os.path.abspath(__file__)), "..", "cst-agent"))
+THINK = ["好的，我先看看工作区里", "有哪些文件，然后用中文", "回答这个问题。😀 结束"]
+ANSWER = "这是中文回答。"
+GOOD_KEYS = {"sk-good", "sk-good2"}
+B_REQS = []
+
+
+def sse_reply(h, model, think, answer):
+    h.send_response(200)
+    h.send_header("Content-Type", "text/event-stream")
+    h.end_headers()
+    base = {"id": "x", "object": "chat.completion.chunk", "model": model}
+
+    def ev(o):
+        h.wfile.write(b"data: " + json.dumps({**base, **o}, ensure_ascii=False).encode() + b"\n\n")
+        h.wfile.flush()
+    for part in think:
+        ev({"choices": [{"index": 0, "delta": {"role": "assistant", "reasoning_content": part}}]})
+    ev({"choices": [{"index": 0, "delta": {"content": answer}}]})
+    ev({"choices": [{"index": 0, "delta": {}, "finish_reason": "stop"}]})
+    h.wfile.write(b"data: [DONE]\n\n")
+    h.wfile.flush()
+
+
+def send_json(h, code, obj):
+    body = json.dumps(obj).encode()
+    h.send_response(code)
+    h.send_header("Content-Type", "application/json")
+    h.send_header("Content-Length", str(len(body)))
+    h.end_headers()
+    h.wfile.write(body)
+
+
+class A(BaseHTTPRequestHandler):
+    """The starting provider: any key, one model, Chinese reasoning."""
+    def log_message(self, *a):
+        pass
+
+    def do_GET(self):
+        send_json(self, 200, {"data": [{"id": "deepseek-flash"}]})
+
+    def do_POST(self):
+        self.rfile.read(int(self.headers.get("Content-Length", 0)))
+        sse_reply(self, "deepseek-flash", THINK, ANSWER)
+
+
+class B(BaseHTTPRequestHandler):
+    """The provider switched to: /models checks the key."""
+    def log_message(self, *a):
+        pass
+
+    def key_ok(self):
+        auth = self.headers.get("Authorization", "")
+        return auth.startswith("Bearer ") and auth[7:] in GOOD_KEYS
+
+    def do_GET(self):
+        if not self.key_ok():
+            return send_json(self, 401, {"error": {"message": "invalid api key"}})
+        send_json(self, 200, {"data": [{"id": "fake-vis"}, {"id": "fake-text"}]})
+
+    def do_POST(self):
+        body = json.loads(self.rfile.read(int(self.headers.get("Content-Length", 0))) or b"{}")
+        B_REQS.append({"auth": self.headers.get("Authorization", ""), "model": body.get("model")})
+        if not self.key_ok():
+            return send_json(self, 401, {"error": {"message": "invalid api key"}})
+        sse_reply(self, body.get("model") or "fake-vis", [], "B 的回答")
+
+
+fails = 0
+
+
+def check(ok, msg):
+    global fails
+    print(("PASS " if ok else "FAIL ") + msg)
+    fails += 0 if ok else 1
+
+
+class Sse:
+    def __init__(self, base):
+        self.lines = []
+        self.r = urllib.request.urlopen(base + "/api/events", timeout=60)
+        threading.Thread(target=self.pump, daemon=True).start()
+
+    def pump(self):
+        try:
+            for raw in self.r:
+                self.lines.append(raw)
+        except Exception:
+            pass
+
+    def events(self, name):
+        out = []
+        for i, l in enumerate(self.lines):
+            if l.strip() == b"event: " + name.encode() and i + 1 < len(self.lines):
+                out.append(self.lines[i + 1][6:])
+        return out
+
+    def close(self):
+        try:
+            self.r.close()
+        except Exception:
+            pass
+
+
+def wait(fn, t=15):
+    end = time.time() + t
+    while time.time() < end:
+        v = fn()
+        if v:
+            return v
+        time.sleep(0.1)
+    return None
+
+
+def call(base, path, obj=None, method=None):
+    data = None if obj is None else json.dumps(obj).encode()
+    req = urllib.request.Request(base + path, data=data, method=method or ("POST" if data else "GET"),
+                                 headers={"Content-Type": "application/json"})
+    try:
+        with urllib.request.urlopen(req, timeout=20) as r:
+            raw = r.read()
+            return r.status, json.loads(raw or b"{}"), raw
+    except urllib.error.HTTPError as e:
+        raw = e.read()
+        try:
+            return e.code, json.loads(raw or b"{}"), raw
+        except ValueError:
+            return e.code, {}, raw
+
+
+def serve(handler):
+    s = ThreadingHTTPServer(("127.0.0.1", 0), handler)
+    threading.Thread(target=s.serve_forever, daemon=True).start()
+    return s
+
+
+api_a, api_b = serve(A), serve(B)
+ws = tempfile.mkdtemp(prefix="cst-set-")
+home = tempfile.mkdtemp(prefix="cst-home-")
+cfgdir = home + "/.config/cst-agent"
+os.makedirs(cfgdir)
+with open(cfgdir + "/models.json", "w") as f:
+    json.dump({"providers": {"fakeb": {"api_base": "http://127.0.0.1:%d/v1" % api_b.server_address[1],
+                                       "key_env": "FAKEB_API_KEY", "default_model": ""}},
+               "models": [{"match": "fake-vis", "vision": True, "context": 200000},
+                          {"match": "fake-text", "vision": False}]}, f)
+env = dict(os.environ, HOME=home, XDG_CONFIG_HOME=home + "/.config",
+           XDG_DATA_HOME=home + "/.local/share", XDG_CACHE_HOME=home + "/.cache")
+for k in ("DEEPSEEK_API_KEY", "FAKEB_API_KEY", "OPENAI_API_KEY", "CST_API_KEY"):
+    env.pop(k, None)
+procs = []
+
+
+def start(args, tag):
+    log = open("%s/%s.log" % (ws, tag), "w+")
+    p = subprocess.Popen([BIN, "--web", "--web-port", "0", "--workspace", ws] + args,
+                         stdout=log, stderr=subprocess.STDOUT, env=env)
+    procs.append((p, log))
+
+    def port():
+        log.seek(0)
+        for l in log.read().splitlines():
+            if l.startswith("webui: http://127.0.0.1:"):
+                return l.split(":")[3].split("/")[0]
+    pt = wait(port)
+    return p, "http://127.0.0.1:%s" % pt
+
+
+def stop(p):
+    p.terminate()
+    try:
+        p.wait(20)
+    except subprocess.TimeoutExpired:
+        p.kill()
+        check(False, "server hung on SIGTERM")
+    check(p.returncode is not None and 0 <= p.returncode < 128, "clean exit (%s)" % p.returncode)
+
+
+def leaked(raw):
+    return any(k.encode() in raw for k in GOOD_KEYS | {"sk-bad", "sk-x"})
+
+
+try:
+    p, base = start(["--new", "--api-key", "sk-x", "--provider", "deepseek", "--model", "deepseek-flash",
+                     "--api-base", "http://127.0.0.1:%d/v1" % api_a.server_address[1],
+                     "--protocol", "native-eos", "--approval", "auto"], "first")
+    a = Sse(base)
+    time.sleep(0.3)
+
+    # 1. Chinese reasoning: whole characters, a handful of events
+    code, _, _ = call(base, "/api/turn", {"text": "用中文想一想"})
+    check(code == 202, "turn accepted")
+    end = wait(lambda: a.events("turn_end"), 20)
+    check(bool(end) and json.loads(end[0]).get("rc") == 0, "turn ended rc=0")
+    raw = a.events("think_delta")
+    ok = True
+    for r in raw:
+        try:
+            r.decode("utf-8")
+        except UnicodeDecodeError:
+            ok = False
+    check(ok, "every think_delta frame is valid UTF-8")
+    text = "".join(json.loads(r.decode("utf-8", "replace"))["text"] for r in raw)
+    want = "".join(THINK)
+    check(text == want and "\ufffd" not in text, "the thinking text round trips: %r" % text[:60])
+    check(0 < len(raw) <= 2 * len(THINK) + 2,
+          "think_delta events: %d for %d bytes" % (len(raw), len(want.encode())))
+    ans = "".join(json.loads(r)["text"] for r in a.events("assistant_delta"))
+    check(ans == ANSWER, "the answer round trips: %r" % ans)
+
+    # 2. providers and models
+    code, d, rawp = call(base, "/api/providers")
+    names = {x["name"]: x for x in d.get("providers", [])}
+    check(code == 200 and "deepseek" in names and "fakeb" in names, "provider list: %s" % sorted(names))
+    check(names.get("deepseek", {}).get("current") and names["deepseek"].get("has_key")
+          and names["deepseek"].get("key_from") == "session", "deepseek: current, key for this run")
+    check(names.get("fakeb", {}).get("has_key") is False and names["fakeb"].get("source") == "user",
+          "fakeb: no key yet, from models.json")
+    check(not leaked(rawp), "no key in /api/providers")
+    code, d, _ = call(base, "/api/models")
+    ids = [m["id"] for m in d.get("models", [])]
+    first = (d.get("models") or [{}])[0]
+    check(code == 200 and "deepseek-flash" in ids and first.get("api") is True
+          and "context" in first and "label" in first, "model list: %s" % ids[:5])
+
+    # 3. config: validation and persistence
+    check(call(base, "/api/config", {"effort": "bogus"})[0] == 400, "a bad effort is a 400")
+    check(call(base, "/api/config", {"provider": "fakeb"})[0] == 400, "provider via /api/config is a 400")
+    check(call(base, "/api/config", {"model": ""})[0] == 400, "an empty model is a 400")
+    code, d, _ = call(base, "/api/config", {"model": "deepseek-v4-pro", "effort": "high"})
+    check(code == 200 and d.get("ok"), "model + effort accepted")
+    code, s, _ = call(base, "/api/state")
+    check(s.get("model") == "deepseek-v4-pro" and s.get("effort") == "high", "the state has them")
+    check(s.get("effort_applies") is True and s.get("key_set") is True and "model_info" in s
+          and s.get("perm_timeout", 0) > 0 and "turn_usage" in s, "the state has the new fields")
+    st = open(cfgdir + "/settings.json").read() if os.path.exists(cfgdir + "/settings.json") else ""
+    check("deepseek-v4-pro" in st and "high" in st, "settings.json remembers model and effort")
+
+    # 4. switching provider: no key, a wrong key, a good key
+    code, d, _ = call(base, "/api/provider", {"name": "fakeb"})
+    check(code == 400 and d.get("need_key"), "fakeb without a key: need_key (%s)" % code)
+    code, d, rawp = call(base, "/api/provider", {"name": "fakeb", "key": "sk-bad"})
+    check(code == 401 and "拒绝" in d.get("error", ""), "a wrong key is rejected (%s %s)" % (code, d.get("error")))
+    check(not leaked(rawp), "no key in the rejection")
+    check(call(base, "/api/state")[1].get("provider") == "deepseek", "still on deepseek")
+    check(call(base, "/api/provider", {"name": "nope"})[0] == 404, "an unknown provider is a 404")
+    code, d, rawp = call(base, "/api/provider", {"name": "fakeb", "key": "sk-good"})
+    check(code == 200 and d.get("provider") == "fakeb" and d.get("model") == "fake-vis",
+          "switched to fakeb, first listed model picked (%s %s)" % (code, d))
+    check(not leaked(rawp), "no key in the switch reply")
+    creds = cfgdir + "/credentials"
+    cr = open(creds).read() if os.path.exists(creds) else ""
+    check("fakeb.api_key=sk-good" in cr, "the key was saved to credentials")
+    if os.name == "posix" and os.path.exists(creds):
+        check(stat.S_IMODE(os.stat(creds).st_mode) == 0o600, "credentials is 0600")
+    code, d, _ = call(base, "/api/models")
+    ms = {m["id"]: m for m in d.get("models", [])}
+    check("fake-vis" in ms and ms["fake-vis"].get("vision") is True and ms["fake-vis"].get("context") == 200000
+          and ms.get("fake-text", {}).get("vision") is False, "fakeb models with capabilities")
+    code, d, _ = call(base, "/api/config", {"effort": "low"})
+    check(code == 200 and d.get("warning"), "effort on a model without it: saved with a warning")
+
+    # 5. a turn now goes to the new base URL with the new key
+    n0 = len(B_REQS)
+    code, _, _ = call(base, "/api/turn", {"text": "hi B"})
+    check(code == 202, "turn on fakeb accepted")
+    wait(lambda: len(a.events("turn_end")) >= 2, 20)
+    check(len(B_REQS) > n0 and B_REQS[-1]["auth"] == "Bearer sk-good" and B_REQS[-1]["model"] == "fake-vis",
+          "the request reached fakeb: %s" % (B_REQS[-1:] or "none"))
+
+    # 6. replacing the key
+    code, d, _ = call(base, "/api/key", {"key": "sk-bad"})
+    check(code == 401, "a wrong replacement key is rejected (%s)" % code)
+    code, d, _ = call(base, "/api/key", {"key": "sk-good2"})
+    check(code == 200 and d.get("ok"), "the replacement key is accepted")
+    check("fakeb.api_key=sk-good2" in open(creds).read(), "the replacement key was saved")
+    check(call(base, "/api/key", {})[0] == 400, "an empty key body is a 400")
+    a.close()
+    stop(p)
+
+    # 7. the next start remembers provider, model and key
+    p, base = start(["--new"], "second")
+    code, s, raws = call(base, "/api/state")
+    check(s.get("provider") == "fakeb" and s.get("model") == "fake-vis" and s.get("key_set") is True
+          and s.get("key_from") == "saved", "restart: %s/%s key %s" % (s.get("provider"), s.get("model"), s.get("key_from")))
+    check(not leaked(raws), "no key in /api/state")
+    stop(p)
+
+    logs = b"".join(open(lg.name, "rb").read() for _, lg in procs)
+    check(not leaked(logs.replace(b"sk-x", b"")) , "no key in the server logs")
+finally:
+    for p, log in procs:
+        if p.poll() is None:
+            p.kill()
+        log.close()
+    shutil.rmtree(ws, ignore_errors=True)
+    shutil.rmtree(home, ignore_errors=True)
+print("FAILED:", fails)
+sys.exit(1 if fails else 0)

scripts/embed_web.sh

@@ -0,0 +1,40 @@
+#!/bin/sh
+# Embed web/ into .gen/web_assets.inc as C byte arrays (the same trick the
+# Makefile uses for src/models_builtin.json). POSIX sh on purpose: the musl
+# and busybox build environments have no bash.  Usage:
+#   sh scripts/embed_web.sh [out.inc] [web-dir]
+set -eu
+ROOT="$(cd "$(dirname "$0")/.." && pwd)"
+OUT="${1:-$ROOT/.gen/web_assets.inc}"
+DIR="${2:-$ROOT/web}"
+
+FILES="index.html app.js style.css favicon.svg"
+
+mkdir -p "$(dirname "$OUT")"
+{
+  echo "/* Generated by scripts/embed_web.sh from web/ - do not edit. */"
+  for f in $FILES; do
+    [ -f "$DIR/$f" ] || continue
+    var="wui_asset_$(echo "$f" | tr -c 'a-zA-Z0-9' '_')"
+    echo "static const unsigned char ${var}[] = {"
+    od -An -v -tx1 "$DIR/$f" | sed 's/\([0-9a-f][0-9a-f]\)/0x\1,/g'
+    echo "0x00};"
+  done
+  echo "static const wui_asset_t wui_assets[] = {"
+  for f in $FILES; do
+    [ -f "$DIR/$f" ] || continue
+    var="wui_asset_$(echo "$f" | tr -c 'a-zA-Z0-9' '_')"
+    case "$f" in
+      *.html) ctype="text/html; charset=utf-8" ;;
+      *.js)   ctype="text/javascript; charset=utf-8" ;;
+      *.css)  ctype="text/css; charset=utf-8" ;;
+      *.svg)  ctype="image/svg+xml" ;;
+      *)      ctype="application/octet-stream" ;;
+    esac
+    echo "  {\"$f\", \"$ctype\", ${var}, sizeof(${var}) - 1},"
+  done
+  echo "};"
+  echo "static const int wui_asset_count = (int)(sizeof(wui_assets) / sizeof(wui_assets[0]));"
+} > "$OUT.tmp"
+mv "$OUT.tmp" "$OUT"
+echo "embedded $(wc -c < "$OUT") bytes of C into $OUT"

scripts/tui_keys/esc.py

@@ -0,0 +1,44 @@
+import os
+import sys, time
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+from harness import Term
+import fakeapi
+from keys import input_text, styled, check, FAILS, base
+
+for term in ("xterm-256color", "screen-256color"):
+    t = Term(args=("-b", base, "-k", "sk-test"), term=term)
+    t.send("some text")
+    t.send("\x1b", 0.4)
+    check(term + " lone esc clears input", input_text(t) == [], t)
+    t.send("alpha beta")
+    t.send("\x1bb")  # alt+b word left
+    t.send("X")
+    check(term + " alt+b word left", "alpha Xbeta" in input_text(t), t)
+    t.send("\x1b\r")  # alt+enter newline
+    t.send("Y")
+    check(term + " alt+enter newline", input_text(t)[-2:] == ["alpha X", "Ybeta"], t)
+    t.send("\x03", 0.2)
+    # picker with CSI form
+    t.send("/model\r", 0.8)
+    s0 = styled(t)
+    t.arrow("down", "csi")
+    s1 = styled(t)
+    t.arrow("up", "csi")
+    s2 = styled(t)
+    check(term + " picker csi down", s0 != s1, t)
+    check(term + " picker csi up", s0 == s2, t)
+    t.send("\x1b", 0.4)
+    # home/end variants
+    t.send("middle")
+    for seq in ("\x1b[H", "\x1bOH", "\x1b[1~", "\x1b[7~"):
+        t.send("\x1b[F" + seq)
+        t.send("<")
+        t.send("\x08" if False else "")
+    check(term + " home variants", input_text(t)[-1:] == ["<<<<middle"], t)
+    for seq in ("\x1b[F", "\x1bOF", "\x1b[4~", "\x1b[8~"):
+        t.send("\x1b[H" + seq)
+        t.send(">")
+    check(term + " end variants", input_text(t)[-1:] == ["<<<<middle>>>>"], t)
+    t.close()
+print("FAILED:", len(FAILS))
+sys.exit(min(len(FAILS), 100))

scripts/tui_keys/fakeapi.py

@@ -0,0 +1,43 @@
+import json, sys, threading, time
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+
+LOG = []
+
+
+class H(BaseHTTPRequestHandler):
+    def log_message(self, *a):
+        pass
+
+    def do_GET(self):
+        body = json.dumps({"data": [{"id": "deepseek-flash"}, {"id": "deepseek-v4-pro"}]}).encode()
+        self.send_response(200)
+        self.send_header("Content-Type", "application/json")
+        self.send_header("Content-Length", str(len(body)))
+        self.end_headers()
+        self.wfile.write(body)
+
+    def do_POST(self):
+        n = int(self.headers.get("Content-Length", 0))
+        raw = self.rfile.read(n)
+        try:
+            LOG.append(json.loads(raw))
+        except Exception:
+            LOG.append({"_raw": raw[:200].decode("latin1")})
+        self.send_response(200)
+        self.send_header("Content-Type", "text/event-stream")
+        self.end_headers()
+        def ev(o):
+            self.wfile.write(b"data: " + json.dumps(o).encode() + b"\n\n")
+            self.wfile.flush()
+        base = {"id": "x", "object": "chat.completion.chunk", "model": "deepseek-flash"}
+        ev({**base, "choices": [{"index": 0, "delta": {"role": "assistant", "content": "reply-%d" % len(LOG)}}]})
+        ev({**base, "choices": [{"index": 0, "delta": {}, "finish_reason": "stop"}],
+            "usage": {"prompt_tokens": 10, "completion_tokens": 2, "total_tokens": 12}})
+        self.wfile.write(b"data: [DONE]\n\n")
+        self.wfile.flush()
+
+
+def start():
+    s = ThreadingHTTPServer(("127.0.0.1", 0), H)
+    threading.Thread(target=s.serve_forever, daemon=True).start()
+    return s, "http://127.0.0.1:%d/v1" % s.server_address[1]

scripts/tui_keys/harness.py

@@ -0,0 +1,99 @@
+"""Drive the cst-agent TUI in a pty and read the screen back (pyte).
+
+Run from the repo root after `make`:
+  nix-shell -p python3Packages.pyte --run "python3 scripts/tui_keys/keys.py"
+  nix-shell -p python3Packages.pyte --run "python3 scripts/tui_keys/esc.py"
+(or `pip install pyte`). A local fake OpenAI-compatible server answers the
+model calls, so no API key is needed. Exit status is the failure count.
+"""
+import os, pty, sys, time, select, fcntl, termios, struct, signal, shutil, tempfile
+import pyte
+
+BIN = os.environ.get("CST_BIN") or os.path.join(
+    os.path.dirname(os.path.abspath(__file__)), "..", "..", "cst-agent")
+
+
+class Term:
+    def __init__(self, cols=120, rows=36, kitty=False, env=None, args=(), home=None,
+                 ws=None, term="xterm-256color"):
+        self.cols, self.rows, self.kitty = cols, rows, kitty
+        self.screen = pyte.Screen(cols, rows)
+        self.stream = pyte.ByteStream(self.screen)
+        self.home = home or tempfile.mkdtemp(prefix="cstk_home_")
+        self.ws = ws or tempfile.mkdtemp(prefix="cstk_ws_")
+        e = dict(os.environ)
+        e.update({"TERM": term, "HOME": self.home,
+                  "XDG_CONFIG_HOME": os.path.join(self.home, ".config"),
+                  "LANG": "C.UTF-8", "LC_ALL": "C.UTF-8",
+                  "CST_PROVIDER": "deepseek", "DEEPSEEK_API_KEY": "sk-dummy-test",
+                  "CST_METRICS": "0"})
+        e.pop("NO_COLOR", None)
+        if env:
+            e.update(env)
+        self.raw = b""
+        pid, fd = pty.fork()
+        if pid == 0:
+            os.chdir(self.ws)
+            os.execve(BIN, [BIN, "-N", *args], e)
+        self.pid, self.fd = pid, fd
+        fcntl.ioctl(fd, termios.TIOCSWINSZ, struct.pack("HHHH", rows, cols, 0, 0))
+        self.pump(1.5)
+
+    def pump(self, t=0.25):
+        end = time.time() + t
+        while True:
+            left = end - time.time()
+            if left <= 0:
+                break
+            r, _, _ = select.select([self.fd], [], [], left)
+            if not r:
+                continue
+            try:
+                d = os.read(self.fd, 65536)
+            except OSError:
+                break
+            if not d:
+                break
+            self.raw += d
+            if self.kitty and b"\x1b[?u" in d:
+                os.write(self.fd, b"\x1b[?0u")
+            self.stream.feed(d)
+
+    def send(self, b, t=0.15):
+        if isinstance(b, str):
+            b = b.encode()
+        os.write(self.fd, b)
+        self.pump(t)
+
+    def appcursor(self):
+        return (1 << 5) in self.screen.mode  # pyte DECCKM
+
+    def arrow(self, d, form=None, mod=None, t=0.15):
+        c = {"up": "A", "down": "B", "right": "C", "left": "D"}[d]
+        if mod:
+            seq = "\x1b[1;%d%s" % (mod, c)
+        else:
+            f = form or ("ss3" if self.appcursor() else "csi")
+            seq = ("\x1bO" if f == "ss3" else "\x1b[") + c
+        self.send(seq, t)
+
+    def lines(self):
+        return [l.rstrip() for l in self.screen.display]
+
+    def find(self, s):
+        return [(i, l) for i, l in enumerate(self.lines()) if s in l]
+
+    def cursor(self):
+        return self.screen.cursor.y, self.screen.cursor.x
+
+    def dump(self, title=""):
+        print("=" * 20, title, "cursor", self.cursor())
+        for i, l in enumerate(self.lines()):
+            print("%2d|%s" % (i, l))
+
+    def close(self):
+        try:
+            os.kill(self.pid, signal.SIGTERM)
+            os.waitpid(self.pid, 0)
+        except Exception:
+            pass

scripts/tui_keys/keys.py

@@ -0,0 +1,273 @@
+import os
+import sys, time
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+from harness import Term
+import fakeapi
+
+srv, base = fakeapi.start()
+FAILS = []
+VERBOSE = "-v" in sys.argv
+
+
+def check(name, cond, t=None):
+    print(("PASS " if cond else "FAIL ") + name)
+    if not cond:
+        FAILS.append(name)
+        if t:
+            t.dump(name)
+
+
+def input_text(t):
+    """Input box rows: the bar rows right above the mode line."""
+    L = t.lines()
+    mode = [i for i, l in enumerate(L) if l.lstrip().startswith("┃ ") and
+            ("Build ·" in l or "Plan ·" in l)]
+    if not mode:
+        return []
+    rows = []
+    i = mode[-1] - 1
+    while i >= 0:
+        s = L[i].lstrip()
+        if not s.startswith("┃"):
+            break
+        body = s[2:].split("     ")[0].rstrip() if s.startswith("┃ ") else ""
+        if body == "" and rows:
+            break
+        if body:
+            rows.append(body)
+        i -= 1
+    rows.reverse()
+    if rows and (rows[0].startswith("Ask anything") or rows[0].startswith("Enter queues")):
+        return []
+    return rows
+
+
+def styled(t):
+    """Screen text plus the reverse/bg attributes, to see highlight moves."""
+    out = []
+    for y in range(t.rows):
+        row = t.screen.buffer[y]
+        out.append("".join("%s%s%s" % (row[x].data, int(row[x].reverse), row[x].bg[:3])
+                           for x in range(t.cols)))
+    return out
+
+
+def clear(t):
+    if [r for r in input_text(t) if r]:
+        t.send("\x03", 0.2)  # ctrl+c clears a non-empty input (twice on empty quits)
+
+
+def new(kitty=False, **kw):
+    t = Term(kitty=kitty, args=("-b", base, "-k", "sk-test", "-m", "deepseek-flash"), **kw)
+    return t
+
+
+def basic(t, form, tag):
+    clear(t)
+    t.send("hello world")
+    for _ in range(5):
+        t.arrow("left", form)
+    t.send("X")
+    check(f"{tag} left x5 insert", "hello Xworld" in input_text(t), t)
+    t.arrow("right", form); t.arrow("right", form)
+    t.send("Y")
+    check(f"{tag} right x2 insert", "hello XwoYrld" in input_text(t), t)
+    # boundaries
+    for _ in range(30):
+        t.arrow("left", form, t=0.02)
+    t.pump(0.2)
+    t.send("<")
+    check(f"{tag} left past start", "<hello XwoYrld" in input_text(t), t)
+    for _ in range(30):
+        t.arrow("right", form, t=0.02)
+    t.pump(0.2)
+    t.send(">")
+    check(f"{tag} right past end", "<hello XwoYrld>" in input_text(t), t)
+    clear(t)
+
+
+def cjk(t):
+    clear(t)
+    t.send("你好世界")
+    y, x_end = t.cursor()
+    t.arrow("left"); t.arrow("left")
+    y2, x2 = t.cursor()
+    check("cjk cursor moves 2 cells per char", x_end - x2 == 4, t)
+    t.send("A")
+    check("cjk insert between wide chars", "你好A世界" in input_text(t), t)
+    t.arrow("right")
+    t.send("B")
+    check("cjk right over wide char", "你好A世B界" in input_text(t), t)
+    clear(t)
+    t.send("a😀b")
+    t.arrow("left"); t.arrow("left")
+    t.send("Z")
+    check("emoji left", "aZ😀b" in input_text(t), t)
+    clear(t)
+
+
+def multiline(t):
+    clear(t)
+    t.send("line one is long")
+    t.send("\x0e")  # ctrl+n newline
+    t.send("l2")
+    t.send("\x0e")
+    t.send("third line here")
+    rows = input_text(t)
+    check("multiline 3 rows", rows[-3:] == ["line one is long", "l2", "third line here"], t)
+    for _ in range(4):
+        t.arrow("left")  # cursor before "here"
+    t.arrow("up")       # col 11 -> clamp to end of "l2"
+    t.send("U")
+    rows = input_text(t)
+    check("up clamps column to short line", rows[-2] == "l2U", t)
+    t.arrow("up")       # col 3 on line one
+    t.send("V")
+    rows = input_text(t)
+    check("up keeps column", rows[-3] == "linVe one is long", t)
+    t.arrow("down"); t.arrow("down")
+    t.send("W")
+    rows = input_text(t)
+    check("down twice to last line", "W" in rows[-1], t)
+    clear(t)
+
+
+def history(t):
+    clear(t)
+    t.send("first prompt\r", 1.5)
+    t.send("second prompt\r", 1.5)
+    t.arrow("up")
+    check("hist up -> newest", input_text(t)[-1:] == ["second prompt"], t)
+    t.arrow("up")
+    check("hist up -> older", input_text(t)[-1:] == ["first prompt"], t)
+    t.arrow("up")
+    check("hist up at oldest stays", input_text(t)[-1:] == ["first prompt"], t)
+    t.arrow("down")
+    check("hist down -> newer", input_text(t)[-1:] == ["second prompt"], t)
+    t.arrow("down")
+    check("hist down past newest -> empty", input_text(t) in ([], [""]), t)
+    # draft survives a trip through history
+    t.send("my draft")
+    t.arrow("up")
+    t.arrow("down")
+    check("draft restored after up/down", input_text(t)[-1:] == ["my draft"], t)
+    clear(t)
+
+
+def wrapped(t):
+    clear(t)
+    long = " ".join("w%02d" % i for i in range(40))  # ~160 chars, wraps
+    t.send(long, 0.4)
+    nrows = len(input_text(t))
+    t.arrow("up")
+    txt = "".join(input_text(t))
+    check("up on wrapped line keeps the text", "w39" in txt and "w00" in txt, t)
+    t.send("Q")
+    rows = input_text(t)
+    check("up on wrapped line moves one visual row", nrows >= 2 and "Q" in "".join(rows[:-1]), t)
+    clear(t)
+
+
+def completion(t):
+    clear(t)
+    t.send("/", 0.3)
+    before = styled(t)
+    t.arrow("down"); t.arrow("down")
+    mid = styled(t)
+    t.arrow("up")
+    after = styled(t)
+    check("completion popup reacts to down", before != mid, t)
+    check("completion up differs from down-down", mid != after, t)
+    check("completion arrows do not touch input", input_text(t)[-1:] == ["/"], t)
+    t.send("\x1b", 0.3)
+    clear(t)
+
+
+def modifiers(t):
+    clear(t)
+    t.send("alpha beta gamma")
+    t.arrow("left", mod=5)   # ctrl+left -> word left
+    t.send("X")
+    check("ctrl+left word", "alpha beta Xgamma" in input_text(t), t)
+    t.arrow("left", mod=3)   # alt+left
+    t.arrow("left", mod=3)
+    t.send("Y")
+    check("alt+left word", "alpha Ybeta Xgamma" in input_text(t), t)
+    t.arrow("right", mod=5)
+    t.send("Z")
+    check("ctrl+right word", "alpha YbetaZ Xgamma" in input_text(t), t)
+    t.arrow("left", mod=2)  # shift+left: should at least not insert junk
+    t.send("S")
+    txt = "".join(input_text(t))
+    check("shift+left no junk", "[" not in txt and ";" not in txt and "1;2" not in txt, t)
+    clear(t)
+
+
+def split_and_burst(t, form):
+    clear(t)
+    t.send("abcdef")
+    ss = "\x1bO" if form == "ss3" else "\x1b["
+    # escape sequence split across writes (slow link)
+    import os
+    os.write(t.fd, ss[:1].encode()); time.sleep(0.008)
+    os.write(t.fd, (ss[1:] + "D").encode()); t.pump(0.2)
+    t.send("1")
+    check(f"split {form} seq", "abcde1f" in input_text(t), t)
+    # key-repeat burst in one write
+    t.send((ss + "D") * 4, 0.3)
+    t.send("2")
+    check(f"burst {form} x4", "ab2cde1f" in input_text(t), t)
+    clear(t)
+
+
+def picker(t):
+    clear(t)
+    t.send("/model\r", 0.8)
+    s0 = styled(t)
+    t.arrow("down")
+    s1 = styled(t)
+    t.arrow("up")
+    s2 = styled(t)
+    check("picker down changes selection", s0 != s1, t)
+    check("picker up returns", s0 == s2, t)
+    t.send("\x1b", 0.4)
+    clear(t)
+
+
+def run(kitty, tag):
+    print("#####", tag)
+    t = new(kitty=kitty)
+    try:
+        basic(t, None, tag + " native")
+        basic(t, "csi" if t.appcursor() else "ss3", tag + " other-form")
+        cjk(t)
+        multiline(t)
+        modifiers(t)
+        split_and_burst(t, "ss3")
+        split_and_burst(t, "csi")
+        completion(t)
+        wrapped(t)
+        history(t)
+        picker(t)
+    finally:
+        t.close()
+
+
+if __name__ == "__main__":
+    run(False, "xterm")
+    run(True, "kitty")
+    for term in ("screen-256color", "tmux-256color", "linux", "vt220", "xterm"):
+        print("#####", term)
+        t = Term(args=("-b", base, "-k", "sk-test"), term=term)
+        try:
+            basic(t, "ss3", term + " ss3")
+            basic(t, "csi", term + " csi")
+            modifiers(t)
+            multiline(t)
+            cjk(t)
+        finally:
+            t.close()
+    print("FAILED:", len(FAILS))
+    for f in FAILS:
+        print("  -", f)
+    sys.exit(min(len(FAILS), 100))

scripts/tui_keys/provider.py

@@ -0,0 +1,65 @@
+"""TUI /provider and /key against a fake API whose /models checks the key.
+
+  nix-shell -p python3Packages.pyte --run "python3 scripts/tui_keys/provider.py"
+
+A wrong key is asked for again, a good one is saved and switches the
+provider; /key replaces it. Same code path as the WebUI (provider_switch.c).
+"""
+import json, os, sys, threading
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+from harness import Term
+from keys import check, base
+
+GOOD = {"sk-good", "sk-good2"}
+
+
+class B(BaseHTTPRequestHandler):
+    def log_message(self, *a):
+        pass
+
+    def do_GET(self):
+        auth = self.headers.get("Authorization", "")
+        ok = auth.startswith("Bearer ") and auth[7:] in GOOD
+        body = json.dumps({"data": [{"id": "fake-one"}]} if ok else {"error": {"message": "bad key"}}).encode()
+        self.send_response(200 if ok else 401)
+        self.send_header("Content-Type", "application/json")
+        self.send_header("Content-Length", str(len(body)))
+        self.end_headers()
+        self.wfile.write(body)
+
+
+srv = ThreadingHTTPServer(("127.0.0.1", 0), B)
+threading.Thread(target=srv.serve_forever, daemon=True).start()
+
+t = Term(args=("-b", base, "-k", "sk-test"))
+cfg = os.path.join(t.home, ".config", "cst-agent")
+os.makedirs(cfg, exist_ok=True)
+t.close()
+with open(os.path.join(cfg, "models.json"), "w") as f:
+    json.dump({"providers": {"fakeb": {"api_base": "http://127.0.0.1:%d/v1" % srv.server_address[1],
+                                       "key_env": "FAKEB_API_KEY", "default_model": ""}}}, f)
+t = Term(args=("-b", base, "-k", "sk-test"), home=t.home, env={"FAKEB_API_KEY": ""})
+creds = os.path.join(cfg, "credentials")
+
+
+def screen():
+    return "\n".join(t.lines())
+
+
+t.send("/provider fakeb\r", 0.8)
+check("a key prompt for fakeb", "API key for fakeb" in screen(), t)
+t.send("sk-bad\r", 1.5)
+check("a wrong key is asked for again", "rejected that key" in screen(), t)
+check("the wrong key is not saved", not os.path.exists(creds) or "sk-bad" not in open(creds).read(), t)
+t.send("sk-good\r", 2.0)
+check("switched to fakeb", "provider → fakeb" in screen(), t)
+check("the good key is saved", os.path.exists(creds) and "fakeb.api_key=sk-good" in open(creds).read(), t)
+check("no key on screen", "sk-good" not in screen() and "sk-bad" not in screen(), t)
+t.send("/key\r", 0.8)
+t.send("sk-good2\r", 2.0)
+check("/key replaces it", "fakeb.api_key=sk-good2" in open(creds).read(), t)
+t.close()
+from keys import FAILS
+print("FAILED:", len(FAILS))
+sys.exit(min(len(FAILS), 100))

src/agent_loop.c

@@ -1282,13 +1282,22 @@ static void on_mm(const ctl_event_t *ev, void *ud) {
 }
 
 static void feed_visible(turn_t *t, const char *data, size_t len) {
-  if (!len || t->stop)
-    return;
-  cst_buf_append(&t->raw, data, len);
-  if (t->ctl)
-    ctl_parser_feed(t->ctl, data, len);
-  else if (t->mm)
-    mm_scan_feed(t->mm, data, len);
+  /* cut after each '\n' and '>': a call only becomes ready there, and
+   * raw_mark must not take in what follows it in the same chunk */
+  while (len && !t->stop) {
+    size_t n = 0;
+    while (n < len && data[n] != '\n' && data[n] != '>')
+      n++;
+    if (n < len)
+      n++;
+    cst_buf_append(&t->raw, data, n);
+    if (t->ctl)
+      ctl_parser_feed(t->ctl, data, n);
+    else if (t->mm)
+      mm_scan_feed(t->mm, data, n);
+    data += n;
+    len -= n;
+  }
 }
 
 static void think_out(turn_t *t, const char *s, size_t n) {
@@ -1306,10 +1315,16 @@ static void on_delta(const char *data, size_t len, void *ud) {
     const char *want = t->in_think ? "</think>" : "<think>";
     size_t wl = strlen(want);
     if (t->taglen == 0 && c != '<') {
+      /* the whole run up to the next '<' in one call: byte by byte, a
+       * front end would get every CJK character in three invalid pieces */
+      size_t j = i + 1;
+      while (j < len && data[j] != '<')
+        j++;
       if (t->in_think)
-        think_out(t, &c, 1);
+        think_out(t, data + i, j - i);
       else
-        feed_visible(t, &c, 1);
+        feed_visible(t, data + i, j - i);
+      i = j - 1;
       continue;
     }
     t->tagbuf[t->taglen++] = c;
@@ -2558,9 +2573,9 @@ static int can_continue(const cst_agent_t *a, const turn_t *t,
   return st->http_status > 0 && st->http_status < 400 && !st->client_stopped;
 }
 
-/* Prompt size at which the session is compacted: CST_CTX_BUDGET, else 60% of
- * the context window capped at 256K tokens. Even a cached prompt is billed
- * on every step, so huge windows are not worth filling. */
+/* Prompt size at which the session is compacted: CST_CTX_BUDGET, else 75% of
+ * the context window, capped at 256K tokens. Even a cached prompt is billed on
+ * every step, so a 1M-token window is not worth filling. */
 static long ctx_budget(const cst_agent_t *a) {
   const char *e = getenv("CST_CTX_BUDGET");
   if (e && atol(e) > 0)
@@ -2568,7 +2583,7 @@ static long ctx_budget(const cst_agent_t *a) {
   long win = cst_model_context_tokens(a->cfg.model);
   if (win <= 0)
     return 0;
-  long b = win / 10 * 6;
+  long b = win / 4 * 3;
   return b > 262144 ? 262144 : b;
 }
 
@@ -2948,6 +2963,42 @@ int cst_agent_new_session(cst_agent_t *a, char *id_out, size_t id_sz) {
   return 0;
 }
 
+int cst_agent_set_workspace(cst_agent_t *a, const char *path, char *err,
+                            size_t err_sz) {
+  char dummy[8];
+  if (!err || !err_sz) {
+    err = dummy;
+    err_sz = sizeof(dummy);
+  }
+  err[0] = 0;
+  if (!a || !path || !path[0]) {
+    snprintf(err, err_sz, "empty path");
+    return -1;
+  }
+  char abs[sizeof(a->cfg.workspace_root)];
+  if (os_realpath(path, abs, sizeof(abs)) != 0) {
+    snprintf(err, err_sz, "no such directory: %s", path);
+    return -1;
+  }
+  /* a file, or a tree we may not write to, fails here instead of halfway
+   * through the next turn */
+  if (cst_agent_store_ensure(abs) != 0) {
+    snprintf(err, err_sz, "cannot use as a workspace: %s", abs);
+    return -1;
+  }
+  snprintf(a->cfg.workspace_root, sizeof(a->cfg.workspace_root), "%s", abs);
+  /* everything that is kept per workspace follows the new root */
+  cst_perms_close(a->perms);
+  a->perms = cst_perms_open(a->cfg.workspace_root);
+  cst_metrics_close(a->metrics);
+  a->metrics = cst_metrics_open(a->cfg.workspace_root,
+                                cst_session_path(a->session));
+  cst_agent_reload_notes(a);
+  ensure_system(a);          /* the prompt names the directory */
+  a->last_prompt_tokens = 0; /* the old budget was measured elsewhere */
+  return 0;
+}
+
 int cst_agent_switch_session(cst_agent_t *a, const char *id) {
   if (!a || !id || !id[0])
     return -1;

src/http_srv.c

@@ -0,0 +1,770 @@
+#ifndef _DEFAULT_SOURCE
+#define _DEFAULT_SOURCE
+#endif
+
+#include "cst_http.h"
+#include "cst_util.h"
+#include "os_compat.h"
+
+#include <errno.h>
+#include <pthread.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#define HTTP_HDR_MAX 32768  /* one header block */
+#define HTTP_LINE_MAX 8192  /* one header line */
+#define HTTP_DEFAULT_BODY (1024 * 1024)
+#define HTTP_DEFAULT_CONN 32
+#define HTTP_IDLE_TIMEOUT 60000
+#define HTTP_READ_TIMEOUT 15000
+
+struct cst_http_server {
+  os_socket_t listener;
+  int port;
+  pthread_t accept_th;
+  int accept_started;
+  pthread_mutex_t mu;
+  pthread_cond_t cv;
+  int stopping;
+  int nconn;
+  cst_http_conn_t *conns_first;
+  cst_http_opts_t o;
+  cst_http_route_t *routes;
+  int nroutes;
+};
+
+struct cst_http_conn {
+  cst_http_server_t *srv;
+  os_socket_t sock;
+  cst_http_req_t req;
+  char rbuf[8192];
+  size_t rlen, rpos;
+  char *body;             /* owned while the request is served */
+  char hdrs[HTTP_HDR_MAX]; /* header values are copied here */
+  size_t hdrs_used;
+  int sent;
+  int streaming;
+  int failed;
+  int nreq;            /* requests served on this connection */
+  int expect_continue; /* the client waits for "100 Continue" */
+  int keep_alive_req;  /* what the client asked for */
+  cst_http_conn_t *prev, *next;
+};
+
+static const char *status_text(int code) {
+  switch (code) {
+  case 200: return "OK";
+  case 201: return "Created";
+  case 202: return "Accepted";
+  case 204: return "No Content";
+  case 302: return "Found";
+  case 400: return "Bad Request";
+  case 401: return "Unauthorized";
+  case 403: return "Forbidden";
+  case 404: return "Not Found";
+  case 405: return "Method Not Allowed";
+  case 409: return "Conflict";
+  case 411: return "Length Required";
+  case 413: return "Payload Too Large";
+  case 415: return "Unsupported Media Type";
+  case 429: return "Too Many Requests";
+  case 500: return "Internal Server Error";
+  case 503: return "Service Unavailable";
+  default: return "Error";
+  }
+}
+
+/* --- reading -------------------------------------------------------------- */
+
+/* deadline: os_now_ms() value; the whole read must be done by then, so a
+ * client trickling one byte at a time cannot hold the thread forever. */
+static int fill(cst_http_conn_t *c, double deadline) {
+  if (c->rpos < c->rlen)
+    return 1;
+  double left = deadline - os_now_ms();
+  if (left <= 0)
+    return -2;
+  long r = os_net_recv(c->sock, c->rbuf, sizeof(c->rbuf), (int)left + 1);
+  if (r <= 0)
+    return (int)r; /* 0 eof, -1 error, -2 timeout */
+  c->rpos = 0;
+  c->rlen = (size_t)r;
+  return 1;
+}
+
+#define RD_TOO_LONG (-3)
+
+static int rd_line(cst_http_conn_t *c, char *out, size_t cap, double deadline) {
+  size_t n = 0;
+  for (;;) {
+    int rc = fill(c, deadline);
+    if (rc <= 0)
+      return rc;
+    while (c->rpos < c->rlen) {
+      char ch = c->rbuf[c->rpos++];
+      if (ch == '\n') {
+        if (n && out[n - 1] == '\r')
+          n--;
+        out[n] = 0;
+        return 1;
+      }
+      if (n + 2 >= cap)
+        return RD_TOO_LONG;
+      out[n++] = ch;
+    }
+  }
+}
+
+/* Exactly n bytes, using what the read buffer already holds. malloc'd. */
+static char *rd_body(cst_http_conn_t *c, size_t n, int timeout_ms) {
+  char *buf = malloc(n + 1);
+  if (!buf)
+    return NULL;
+  size_t got = 0;
+  while (got < n) {
+    if (c->rpos < c->rlen) {
+      size_t take = c->rlen - c->rpos;
+      if (take > n - got)
+        take = n - got;
+      memcpy(buf + got, c->rbuf + c->rpos, take);
+      c->rpos += take;
+      got += take;
+      continue;
+    }
+    long r = os_net_recv(c->sock, c->rbuf, sizeof(c->rbuf), timeout_ms);
+    if (r <= 0) {
+      free(buf);
+      return NULL;
+    }
+    c->rpos = 0;
+    c->rlen = (size_t)r;
+  }
+  buf[n] = 0;
+  return buf;
+}
+
+/* -1 when the input encodes a NUL (out then holds what came before it). */
+static int url_decode(const char *in, char *out, size_t cap, int plus_space) {
+  size_t o = 0;
+  for (const char *p = in; *p && o + 1 < cap; p++) {
+    if (*p == '%' && p[1] && p[2]) {
+      int hi = p[1] >= '0' && p[1] <= '9'   ? p[1] - '0'
+               : p[1] >= 'a' && p[1] <= 'f' ? p[1] - 'a' + 10
+               : p[1] >= 'A' && p[1] <= 'F' ? p[1] - 'A' + 10
+                                            : -1;
+      int lo = p[2] >= '0' && p[2] <= '9'   ? p[2] - '0'
+               : p[2] >= 'a' && p[2] <= 'f' ? p[2] - 'a' + 10
+               : p[2] >= 'A' && p[2] <= 'F' ? p[2] - 'A' + 10
+                                            : -1;
+      if (hi >= 0 && lo >= 0) {
+        char ch = (char)((hi << 4) | lo);
+        if (ch == 0) {
+          out[o] = 0;
+          return -1;
+        }
+        out[o++] = ch;
+        p += 2;
+        continue;
+      }
+    }
+    out[o++] = (*p == '+' && plus_space) ? ' ' : *p;
+  }
+  out[o] = 0;
+  return 0;
+}
+
+/* Comma-separated header tokens, compared without case ("Close"). */
+static int has_token(const char *val, const char *tok) {
+  size_t n = strlen(tok);
+  for (const char *p = val; p && *p;) {
+    while (*p == ' ' || *p == '\t' || *p == ',')
+      p++;
+    size_t i = 0;
+    while (i < n && p[i] && (p[i] | 0x20) == (tok[i] | 0x20))
+      i++;
+    if (i == n && (p[i] == 0 || p[i] == ',' || p[i] == ' ' || p[i] == '\t'))
+      return 1;
+    p = strchr(p, ',');
+  }
+  return 0;
+}
+
+static int key_is(const char *line, const char *key) {
+  size_t n = strlen(key);
+  for (size_t i = 0; i < n; i++) {
+    char a = line[i], b = key[i];
+    if (a >= 'A' && a <= 'Z')
+      a = (char)(a - 'A' + 'a');
+    if (b >= 'A' && b <= 'Z')
+      b = (char)(b - 'A' + 'a');
+    if (a != b || !a)
+      return 0;
+  }
+  return line[n] == ':';
+}
+
+/* Copy a header value into the connection (they must not point into the
+ * line buffer, which is reused for the next header). */
+static const char *store_str(cst_http_conn_t *c, const char *val) {
+  size_t n = strlen(val);
+  if (c->hdrs_used + n + 1 > sizeof(c->hdrs))
+    return NULL;
+  char *dst = c->hdrs + c->hdrs_used;
+  memcpy(dst, val, n + 1);
+  c->hdrs_used += n + 1;
+  return dst;
+}
+
+/* 1 = parsed, 0 = the client went away or waited too long, -1 = malformed. */
+static int read_request(cst_http_conn_t *c, size_t *want_body) {
+  cst_http_server_t *s = c->srv;
+  char line[HTTP_LINE_MAX];
+  cst_http_req_t *r = &c->req;
+  int first = c->nreq == 0; /* a keep-alive pause gets the idle timeout */
+  memset(r, 0, sizeof(*r));
+  c->hdrs_used = 0;
+  c->expect_continue = 0;
+  *want_body = 0;
+  int rc = rd_line(c, line, sizeof(line),
+                   os_now_ms() + (first ? s->o.read_timeout_ms
+                                        : s->o.idle_timeout_ms));
+  if (rc == RD_TOO_LONG)
+    return -1;
+  if (rc <= 0)
+    return 0;
+  char *sp1 = strchr(line, ' ');
+  if (!sp1)
+    return -1;
+  *sp1 = 0;
+  char *target = sp1 + 1;
+  char *version = strchr(target, ' ');
+  if (version) {
+    *version++ = 0;
+  } else {
+    version = (char *)"HTTP/1.0";
+  }
+  if (!line[0] || strlen(line) >= sizeof(r->method))
+    return -1;
+  snprintf(r->method, sizeof(r->method), "%s", line);
+  r->keep_alive = strncmp(version, "HTTP/1.", 7) == 0 &&
+                  version[7] >= '1' && version[7] <= '9';
+  /* absolute-form ("http://host/path") is legal: keep the path */
+  if (target[0] != '/') {
+    char *scheme = strstr(target, "://");
+    target = scheme ? strchr(scheme + 3, '/') : NULL;
+  }
+  if (!target || target[0] != '/')
+    return -1;
+  {
+    char raw[sizeof(r->path)];
+    snprintf(raw, sizeof(raw), "%s", target);
+    char *q = strchr(raw, '?');
+    if (q) {
+      *q = 0;
+      snprintf(r->query, sizeof(r->query), "%s", q + 1);
+    }
+    if (url_decode(raw, r->path, sizeof(r->path), 0) != 0)
+      return -1;
+    size_t plen = strlen(r->path);
+    if (plen == 0 || strstr(r->path, ".."))
+      return -1;
+  }
+  size_t hdr_bytes = 0;
+  size_t content_length = 0;
+  int have_length = 0;
+  double deadline = os_now_ms() + s->o.read_timeout_ms;
+  for (;;) {
+    rc = rd_line(c, line, sizeof(line), deadline);
+    if (rc <= 0)
+      return -1;
+    if (!line[0])
+      break;
+    hdr_bytes += strlen(line) + 2;
+    if (hdr_bytes > HTTP_HDR_MAX)
+      return -1;
+    const char *colon = strchr(line, ':');
+    if (!colon)
+      return -1;
+    const char *val = colon + 1;
+    while (*val == ' ' || *val == '\t')
+      val++;
+    if (key_is(line, "content-length")) {
+      /* digits only, and one value: "5x" or two lengths would let the
+       * rest of the body run as a second request behind a proxy */
+      char *end = NULL;
+      errno = 0;
+      unsigned long long v = strtoull(val, &end, 10);
+      while (end && (*end == ' ' || *end == '\t'))
+        end++;
+      if (*val < '0' || *val > '9' || !end || *end || errno == ERANGE ||
+          v > (unsigned long long)SIZE_MAX ||
+          (have_length && content_length != (size_t)v))
+        return -1;
+      content_length = (size_t)v;
+      have_length = 1;
+    } else if (key_is(line, "transfer-encoding")) {
+      return -1; /* chunked request bodies are not supported */
+    } else if (key_is(line, "connection")) {
+      if (has_token(val, "close"))
+        r->keep_alive = 0;
+      else if (has_token(val, "keep-alive"))
+        r->keep_alive = 1;
+    } else if (key_is(line, "expect")) {
+      /* answered once auth and the size cap have passed */
+      c->expect_continue = has_token(val, "100-continue");
+    } else if (key_is(line, "authorization")) {
+      r->authorization = store_str(c, val);
+    } else if (key_is(line, "cookie")) {
+      r->cookie = store_str(c, val);
+    } else if (key_is(line, "origin")) {
+      r->origin = store_str(c, val);
+    } else if (key_is(line, "host")) {
+      r->host = store_str(c, val);
+    } else if (key_is(line, "content-type")) {
+      r->content_type = store_str(c, val);
+    } else if (key_is(line, "user-agent")) {
+      r->user_agent = store_str(c, val);
+    } else if (key_is(line, "last-event-id")) {
+      r->last_event_id = store_str(c, val);
+    }
+  }
+  if (have_length)
+    *want_body = content_length;
+  else if (strcmp(r->method, "POST") == 0)
+    /* No Content-Length means no body (RFC 9112 6.3): curl -X POST sends
+     * exactly that, and the zero-body endpoints (cancel, undo, quit) would
+     * otherwise be unreachable from a hand-written client. Bytes that
+     * follow are not ours to guess at, so the connection ends here. */
+    r->keep_alive = 0;
+  return 1;
+}
+
+/* --- writing -------------------------------------------------------------- */
+
+static int send_head(cst_http_conn_t *c, int status, const char *ctype,
+                     size_t len, int chunked) {
+  cst_buf_t out = {0};
+  cst_buf_printf(&out, "HTTP/1.1 %d %s\r\nContent-Type: %s\r\n",
+                 status, status_text(status), ctype ? ctype : "text/plain");
+  cst_buf_puts(&out, "Cache-Control: no-store\r\n");
+  cst_buf_puts(&out, c->req.keep_alive ? "Connection: keep-alive\r\n"
+                                       : "Connection: close\r\n");
+  if (chunked) {
+    cst_buf_puts(&out, "Transfer-Encoding: chunked\r\n");
+  } else {
+    cst_buf_printf(&out, "Content-Length: %zu\r\n", len);
+  }
+  cst_buf_puts(&out, "\r\n");
+  int rc = out.buf ? os_net_send_all(c->sock, out.buf, out.len, 0) : -1;
+  cst_buf_free(&out);
+  return rc;
+}
+
+int cst_http_send(cst_http_conn_t *c, int status, const char *ctype,
+                  const void *body, size_t n) {
+  if (!c || c->sent)
+    return -1;
+  c->sent = 1;
+  if (send_head(c, status, ctype, n, 0) != 0) {
+    c->failed = 1;
+    return -1;
+  }
+  if (n && os_net_send_all(c->sock, body, n, 0) != 0) {
+    c->failed = 1;
+    return -1;
+  }
+  return 0;
+}
+
+int cst_http_send_str(cst_http_conn_t *c, int status, const char *ctype,
+                      const char *s) {
+  return cst_http_send(c, status, ctype, s ? s : "", s ? strlen(s) : 0);
+}
+
+int cst_http_send_json(cst_http_conn_t *c, int status, const char *json) {
+  return cst_http_send_str(c, status, "application/json; charset=utf-8",
+                           json ? json : "null");
+}
+
+int cst_http_send_err(cst_http_conn_t *c, int status, const char *msg) {
+  cst_buf_t b = {0};
+  cst_buf_puts(&b, "{\"error\":\"");
+  for (const char *p = msg ? msg : "error"; *p; p++) {
+    if (*p == '"' || *p == '\\')
+      cst_buf_printf(&b, "\\%c", *p);
+    else if ((unsigned char)*p < 0x20)
+      cst_buf_printf(&b, "\\u%04x", (unsigned char)*p);
+    else
+      cst_buf_append(&b, p, 1);
+  }
+  cst_buf_puts(&b, "\"}");
+  int rc = cst_http_send_json(c, status, b.buf);
+  cst_buf_free(&b);
+  return rc;
+}
+
+int cst_http_sent(const cst_http_conn_t *c) { return c ? c->sent : 0; }
+
+int cst_http_stream_begin(cst_http_conn_t *c, const char *ctype) {
+  if (!c || c->sent)
+    return -1;
+  c->sent = 1;
+  c->streaming = 1;
+  if (send_head(c, 200, ctype, 0, 1) != 0) {
+    c->failed = 1;
+    return -1;
+  }
+  return 0;
+}
+
+int cst_http_stream_write(cst_http_conn_t *c, const void *data, size_t n) {
+  if (!c || !c->streaming || c->failed)
+    return -1;
+  if (n == 0)
+    return 0;
+  cst_buf_t b = {0};
+  cst_buf_printf(&b, "%zx\r\n", n);
+  cst_buf_append(&b, data, n);
+  cst_buf_puts(&b, "\r\n");
+  int rc = b.buf ? os_net_send_all(c->sock, b.buf, b.len, 0) : -1;
+  cst_buf_free(&b);
+  if (rc != 0)
+    c->failed = 1;
+  return rc;
+}
+
+int cst_http_stream_end(cst_http_conn_t *c) {
+  if (!c || !c->streaming)
+    return -1;
+  c->streaming = 0;
+  if (c->failed)
+    return -1;
+  const char *end = "0\r\n\r\n";
+  if (os_net_send_all(c->sock, end, strlen(end), 0) != 0) {
+    c->failed = 1;
+    return -1;
+  }
+  return 0;
+}
+
+const cst_http_req_t *cst_http_request(cst_http_conn_t *c) {
+  return c ? &c->req : NULL;
+}
+
+int cst_http_query(const cst_http_req_t *req, const char *key, char *out,
+                   size_t out_sz) {
+  if (!req || !key || !out || out_sz == 0)
+    return -1;
+  size_t klen = strlen(key);
+  const char *p = req->query;
+  while (p && *p) {
+    const char *amp = strchr(p, '&');
+    size_t seg = amp ? (size_t)(amp - p) : strlen(p);
+    const char *eq = memchr(p, '=', seg);
+    size_t name_len = eq ? (size_t)(eq - p) : seg;
+    if (name_len == klen && strncmp(p, key, klen) == 0) {
+      char raw[1024];
+      size_t vlen = eq ? seg - name_len - 1 : 0;
+      if (vlen >= sizeof(raw))
+        vlen = sizeof(raw) - 1;
+      if (eq)
+        memcpy(raw, eq + 1, vlen);
+      raw[vlen] = 0;
+      url_decode(raw, out, out_sz, 1);
+      return 0;
+    }
+    p = amp ? amp + 1 : NULL;
+  }
+  return -1;
+}
+
+/* --- connections ---------------------------------------------------------- */
+
+static const cst_http_route_t *find_route(const cst_http_server_t *s,
+                                          const cst_http_req_t *req) {
+  const cst_http_route_t *fallback = NULL;
+  for (int i = 0; i < s->nroutes; i++) {
+    const cst_http_route_t *r = &s->routes[i];
+    if (strcmp(r->method, "*") != 0 && strcmp(r->method, req->method) != 0)
+      continue;
+    if (strcmp(r->path, "*") == 0) {
+      fallback = r;
+      continue;
+    }
+    if (strcmp(r->path, req->path) == 0)
+      return r;
+  }
+  return fallback;
+}
+
+/* Route, auth and the size cap are settled before the body is read: an
+ * unauthenticated client must not make us buffer 32 MiB. */
+static void serve(cst_http_server_t *s, cst_http_conn_t *c, size_t want) {
+  const cst_http_route_t *route = find_route(s, &c->req);
+  size_t cap = route && route->max_body ? route->max_body : s->o.max_body;
+  if (want > 0)
+    c->req.keep_alive = 0; /* until the body is read, the stream is ours */
+  if (!route) {
+    cst_http_send_err(c, 404, "not found");
+    return;
+  }
+  if (route->auth && s->o.auth && s->o.auth(&c->req, s->o.auth_ud) != 0) {
+    cst_http_send_err(c, 401, "unauthorized");
+    return;
+  }
+  if (want > cap) {
+    cst_http_send_err(c, 413, "request body too large");
+    return;
+  }
+  if (want > 0) {
+    if (c->expect_continue) {
+      const char *cont = "HTTP/1.1 100 Continue\r\n\r\n";
+      os_net_send_all(c->sock, cont, strlen(cont), 5000);
+    }
+    char *body = rd_body(c, want, s->o.read_timeout_ms);
+    if (!body) {
+      c->failed = 1;
+      return;
+    }
+    c->body = body;
+    c->req.body = body;
+    c->req.body_len = want;
+    c->req.keep_alive = c->keep_alive_req;
+  }
+  int rc = route->fn(c, route->ud);
+  if (!c->sent)
+    cst_http_send_err(c, rc == 0 ? 500 : 404,
+                      rc == 0 ? "handler sent no response" : "not found");
+}
+
+static void *conn_main(void *arg) {
+  cst_http_conn_t *c = arg;
+  cst_http_server_t *s = c->srv;
+  for (;;) {
+    size_t want = 0;
+    int rc = read_request(c, &want);
+    if (rc < 0) {
+      /* malformed (bad request line, traversal, unsupported transfer
+       * encoding): say so instead of closing silently */
+      c->sent = 0;
+      c->req.keep_alive = 0;
+      cst_http_send_err(c, 400, "malformed request");
+      break;
+    }
+    if (rc == 0)
+      break;
+    c->nreq++;
+    c->sent = 0;
+    c->streaming = 0;
+    c->keep_alive_req = c->req.keep_alive;
+    serve(s, c, want);
+    if (s->o.on_log) {
+      char line[512];
+      snprintf(line, sizeof(line), "%s %.480s -> %d", c->req.method, c->req.path,
+               c->failed ? 499 : (c->sent ? 200 : 500));
+      s->o.on_log(line, s->o.log_ud);
+    }
+    free(c->body);
+    c->body = NULL;
+    c->req.body = NULL;
+    c->req.body_len = 0;
+    c->req.authorization = c->req.cookie = c->req.origin = NULL;
+    c->req.host = c->req.content_type = c->req.user_agent = NULL;
+    c->req.last_event_id = NULL;
+    if (!c->req.keep_alive || s->stopping || c->failed)
+      break;
+  }
+  /* unlink first: stop() shuts down every listed socket, and a closed fd
+   * number may already belong to someone else */
+  pthread_mutex_lock(&s->mu);
+  if (c->prev)
+    c->prev->next = c->next;
+  else
+    s->conns_first = c->next;
+  if (c->next)
+    c->next->prev = c->prev;
+  pthread_mutex_unlock(&s->mu);
+  os_net_shutdown(c->sock);
+  os_net_close(c->sock);
+  free(c);
+  pthread_mutex_lock(&s->mu);
+  s->nconn--;
+  pthread_cond_broadcast(&s->cv);
+  pthread_mutex_unlock(&s->mu);
+  return NULL;
+}
+
+static void *accept_main(void *arg) {
+  cst_http_server_t *s = arg;
+  while (!s->stopping) {
+    int timed_out = 0;
+    os_socket_t sock = os_net_accept(s->listener, 500, &timed_out);
+    if (sock == OS_SOCKET_BAD) {
+      if (!timed_out) { /* EMFILE and friends: the listener stays readable */
+        struct timespec ts;
+        clock_gettime(CLOCK_REALTIME, &ts);
+        ts.tv_nsec += 100000000;
+        if (ts.tv_nsec >= 1000000000) {
+          ts.tv_sec++;
+          ts.tv_nsec -= 1000000000;
+        }
+        pthread_mutex_lock(&s->mu);
+        if (!s->stopping)
+          pthread_cond_timedwait(&s->cv, &s->mu, &ts);
+        pthread_mutex_unlock(&s->mu);
+      }
+      continue;
+    }
+    if (s->stopping) {
+      os_net_close(sock);
+      break;
+    }
+    pthread_mutex_lock(&s->mu);
+    int over = s->nconn >= s->o.max_conn;
+    if (!over)
+      s->nconn++;
+    pthread_mutex_unlock(&s->mu);
+    if (over) {
+      const char *busy = "HTTP/1.1 503 Service Unavailable\r\n"
+                         "Content-Length: 0\r\nConnection: close\r\n\r\n";
+      os_net_send_all(sock, busy, strlen(busy), 2000);
+      os_net_close(sock);
+      continue;
+    }
+    cst_http_conn_t *c = calloc(1, sizeof(*c));
+    if (!c) {
+      pthread_mutex_lock(&s->mu);
+      s->nconn--;
+      pthread_cond_broadcast(&s->cv);
+      pthread_mutex_unlock(&s->mu);
+      os_net_close(sock);
+      continue;
+    }
+    c->srv = s;
+    c->sock = sock;
+    pthread_mutex_lock(&s->mu);
+    c->next = s->conns_first;
+    if (c->next)
+      c->next->prev = c;
+    s->conns_first = c;
+    pthread_mutex_unlock(&s->mu);
+    pthread_t th;
+    if (pthread_create(&th, NULL, conn_main, c) != 0) {
+      pthread_mutex_lock(&s->mu);
+      s->conns_first = c->next;
+      if (c->next)
+        c->next->prev = NULL;
+      s->nconn--;
+      pthread_cond_broadcast(&s->cv);
+      pthread_mutex_unlock(&s->mu);
+      os_net_close(sock);
+      free(c);
+      continue;
+    }
+    pthread_detach(th);
+  }
+  return NULL;
+}
+
+cst_http_server_t *cst_http_start(const cst_http_opts_t *o, char *err,
+                                  size_t err_sz) {
+  if (!o)
+    return NULL;
+  if (os_net_init() != 0) {
+    if (err)
+      snprintf(err, err_sz, "network init failed");
+    return NULL;
+  }
+  cst_http_server_t *s = calloc(1, sizeof(*s));
+  if (!s)
+    return NULL;
+  s->o = *o;
+  s->o.max_body = o->max_body ? o->max_body : HTTP_DEFAULT_BODY;
+  s->o.max_conn = o->max_conn ? o->max_conn : HTTP_DEFAULT_CONN;
+  if (!s->o.read_timeout_ms)
+    s->o.read_timeout_ms = HTTP_READ_TIMEOUT;
+  if (!s->o.idle_timeout_ms)
+    s->o.idle_timeout_ms = HTTP_IDLE_TIMEOUT;
+  if (o->nroutes > 0) {
+    s->routes = calloc((size_t)o->nroutes, sizeof(*s->routes));
+    if (!s->routes) {
+      free(s);
+      return NULL;
+    }
+    memcpy(s->routes, o->routes, (size_t)o->nroutes * sizeof(*s->routes));
+    s->nroutes = o->nroutes;
+  }
+  pthread_mutex_init(&s->mu, NULL);
+  pthread_cond_init(&s->cv, NULL);
+  s->listener = os_net_listen(o->host, o->port, err, err_sz);
+  if (s->listener == OS_SOCKET_BAD) {
+    free(s->routes);
+    free(s);
+    return NULL;
+  }
+  s->port = os_net_port(s->listener);
+  if (pthread_create(&s->accept_th, NULL, accept_main, s) != 0) {
+    os_net_close(s->listener);
+    free(s->routes);
+    free(s);
+    if (err)
+      snprintf(err, err_sz, "cannot start the accept thread");
+    return NULL;
+  }
+  s->accept_started = 1;
+  return s;
+}
+
+int cst_http_port(const cst_http_server_t *s) { return s ? s->port : 0; }
+
+int cst_http_stopping(const cst_http_server_t *s) {
+  return s ? s->stopping : 1;
+}
+
+int cst_http_stop(cst_http_server_t *s) {
+  if (!s)
+    return 0;
+  pthread_mutex_lock(&s->mu);
+  s->stopping = 1;
+  for (cst_http_conn_t *c = s->conns_first; c; c = c->next)
+    os_net_shutdown(c->sock);
+  pthread_mutex_unlock(&s->mu);
+  os_net_shutdown(s->listener);
+  if (s->accept_started) {
+    pthread_join(s->accept_th, NULL);
+    s->accept_started = 0;
+  }
+  /* Handler threads own their connection until they return; the shutdown
+   * above is what wakes them. Bounded wait so free() stays safe. */
+  pthread_mutex_lock(&s->mu);
+  for (int i = 0; i < 300 && s->nconn > 0; i++) {
+    struct timespec ts;
+    clock_gettime(CLOCK_REALTIME, &ts);
+    ts.tv_nsec += 100000000; /* 100 ms */
+    if (ts.tv_nsec >= 1000000000) {
+      ts.tv_sec++;
+      ts.tv_nsec -= 1000000000;
+    }
+    pthread_cond_timedwait(&s->cv, &s->mu, &ts);
+  }
+  int left = s->nconn;
+  pthread_mutex_unlock(&s->mu);
+  os_net_close(s->listener);
+  return left;
+}
+
+void cst_http_free(cst_http_server_t *s) {
+  if (!s)
+    return;
+  pthread_mutex_lock(&s->mu);
+  int left = s->nconn;
+  pthread_mutex_unlock(&s->mu);
+  if (left > 0)
+    return; /* a handler still runs on it (see stop()): leak, not crash */
+  pthread_mutex_destroy(&s->mu);
+  pthread_cond_destroy(&s->cv);
+  free(s->routes);
+  free(s);
+}

src/llm_openai.c

@@ -162,7 +162,14 @@ static void parts_add_text(cJSON *parts, cJSON *txt) {
     cJSON_Delete(txt);
     return;
   }
-  cJSON_AddItemToArray(parts, txt);
+  if (!txt->valuestring || !txt->valuestring[0]) {
+    cJSON_Delete(txt);
+    return;
+  }
+  cJSON *t = cJSON_CreateObject();
+  cJSON_AddStringToObject(t, "type", "text");
+  cJSON_AddItemToObject(t, "text", txt);
+  cJSON_AddItemToArray(parts, t);
 }
 
 /* One stored image as an image_url part, or NULL with why filled in. */

src/main.c

@@ -3,6 +3,7 @@
 #include "cst_models.h"
 #include "cst_tools.h"
 #include "cst_version.h"
+#include "cst_webui.h"
 #include "os_compat.h"
 
 #include <curl/curl.h>
@@ -41,6 +42,13 @@ static void usage(const char *argv0) {
           "  -T, --tools PROFILE   default (all tools) | minimal (paper set)\n"
           "  -n, --max-steps N     model calls per turn (30)\n"
           "  -H, --headless -p TEXT  run one prompt without the TUI\n"
+          "      --web             serve the UI over HTTP on 127.0.0.1:8787\n"
+          "      --web-port N      port (default 8787, 0 = pick a free one)\n"
+          "      --web-host ADDR   bind address (default 127.0.0.1; a\n"
+          "                        non-loopback address requires a token)\n"
+          "      --web-token T     require ?token=T / Authorization: Bearer T\n"
+          "      --web-dev DIR     serve the UI from DIR (development)\n"
+          "      --web-open        open the browser at startup\n"
           "  -a, --ask-key         enter and save the provider's API key\n"
           "  -k, --api-key KEY     use KEY for this run only (not saved)\n"
           "  -b, --api-base URL    OpenAI-compatible base URL (…/v1)\n"
@@ -247,6 +255,16 @@ static int demo_ctl_impl(const char *workspace) {
   return 0;
 }
 
+/* 0..65535, digits only: atoi turned "70000" into port 4464, "abc" into 0. */
+static int parse_port(const char *s, int *out) {
+  char *end = NULL;
+  long v = strtol(s, &end, 10);
+  if (!s[0] || *end || v < 0 || v > 65535)
+    return -1;
+  *out = (int)v;
+  return 0;
+}
+
 int main(int argc, char **argv) {
   if (os_console_init() != 0)
     fprintf(stderr, "warning: the process code page is not UTF-8 (Windows 10 "
@@ -320,6 +338,19 @@ int main(int argc, char **argv) {
   const char *prompt = NULL;
   int new_session = 0, do_list = 0;
   const char *resume_id = NULL;
+  /* WebUI (--web): a second front end on the same engine. */
+  int web_mode = 0;
+  cst_webui_opts_t web;
+  memset(&web, 0, sizeof(web));
+  web.port = CST_WEBUI_DEFAULT_PORT; /* 0 would mean "pick a free one" */
+  const char *env = getenv("CST_WEB_PORT");
+  if (env && env[0] && parse_port(env, &web.port) != 0) {
+    fprintf(stderr, "CST_WEB_PORT=%s is not a port (0-65535)\n", env);
+    return 1;
+  }
+  web.host = getenv("CST_WEB_HOST");
+  web.token = getenv("CST_WEB_TOKEN");
+  web.root = getenv("CST_WEB_ROOT");
 
   static struct option opts[] = {
       {"workspace", required_argument, 0, 'w'},
@@ -346,6 +377,12 @@ int main(int argc, char **argv) {
       {"help", no_argument, 0, 'h'},
       {"version", no_argument, 0, 'V'},
       {"list-models", optional_argument, 0, 0x100},
+      {"web", no_argument, 0, 0x101},
+      {"web-port", required_argument, 0, 0x102},
+      {"web-host", required_argument, 0, 0x103},
+      {"web-token", required_argument, 0, 0x104},
+      {"web-dev", required_argument, 0, 0x105},
+      {"web-open", no_argument, 0, 0x106},
       {0, 0, 0, 0}};
   int c;
   while ((c = getopt_long(argc, argv, "w:s:b:k:m:P:r:n:T:Hp:dahVNcR:LyA:C:e:", opts, NULL)) !=
@@ -453,6 +490,33 @@ int main(int argc, char **argv) {
       curl_global_cleanup();
       return rc;
     }
+    case 0x101:
+      web_mode = 1;
+      break;
+    case 0x102:
+      if (parse_port(optarg, &web.port) != 0) {
+        fprintf(stderr, "--web-port %s is not a port (0-65535)\n", optarg);
+        curl_global_cleanup();
+        return 1;
+      }
+      web_mode = 1;
+      break;
+    case 0x103:
+      web.host = optarg;
+      web_mode = 1;
+      break;
+    case 0x104:
+      web.token = optarg;
+      web_mode = 1;
+      break;
+    case 0x105:
+      web.root = optarg;
+      web_mode = 1;
+      break;
+    case 0x106:
+      web.open_browser = 1;
+      web_mode = 1;
+      break;
     case 'V':
       printf("cst-agent %s\n", CST_VERSION);
       curl_global_cleanup();
@@ -600,7 +664,7 @@ int main(int argc, char **argv) {
       return 0;
     }
   }
-  if (!cfg.api_key[0] && cfg.headless) {
+  if (!cfg.api_key[0] && (cfg.headless || web_mode)) {
     fprintf(stderr,
             "no API key for %s: start cst-agent once to enter it, or run "
             "cst-agent --ask-key --provider %s\n",
@@ -610,7 +674,7 @@ int main(int argc, char **argv) {
   }
   /* Router presets have no fixed model: take the first one the API lists
    * (the TUI does the same after its /models probe). */
-  if (!cfg.model[0] && cfg.headless) {
+  if (!cfg.model[0] && (cfg.headless || web_mode)) {
     static cst_model_meta_t list[512];
     char err[256] = "";
     long age = 0;
@@ -654,7 +718,10 @@ int main(int argc, char **argv) {
   }
 
   int rc = 0;
-  if (cfg.headless) {
+  if (web_mode) {
+    web.prompt = prompt; /* --web -p "...": run one turn at startup */
+    rc = cst_webui_run(agent, &web);
+  } else if (cfg.headless) {
     if (!prompt) {
       fprintf(stderr, "--headless requires --prompt\n");
       rc = 1;

src/models.c

@@ -1,6 +1,11 @@
+#ifndef _DEFAULT_SOURCE
+#define _DEFAULT_SOURCE
+#endif
+
 #include "cst_agent.h"
 
 #include <string.h>
+#include <time.h>
 
 /* Provider and model data live in model_registry.c (cst_models.h). */
 

src/models_builtin.json

@@ -31,6 +31,9 @@
   "models": [
     {"match": "deepseek", "context": 1000000, "vision": false,
      "price": {"currency": "¥", "miss": 2, "hit": 0.04, "out": 8, "offpeak_half": true}},
+    {"match": "deepseek-flash", "vision": true},
+    {"match": "deepseek-chat", "vision": true},
+    {"match": "deepseek-reasoner", "vision": true},
     {"match": "deepseek-v4-pro",
      "price": {"currency": "¥", "miss": 9, "hit": 0.3, "out": 27, "offpeak_half": true}},
     {"match": "deepseek-vl", "vision": true},

src/net.c

@@ -0,0 +1,335 @@
+#ifndef _DEFAULT_SOURCE
+#define _DEFAULT_SOURCE
+#endif
+
+#include "cst_net.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#ifdef _WIN32
+#ifndef WIN32_LEAN_AND_MEAN
+#define WIN32_LEAN_AND_MEAN
+#endif
+#if !defined(_WIN32_WINNT) || _WIN32_WINNT < 0x0600
+#undef _WIN32_WINNT
+#define _WIN32_WINNT 0x0600 /* WSAPoll */
+#endif
+#include <winsock2.h>
+#include <ws2tcpip.h>
+#else
+#include <arpa/inet.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <netdb.h>
+#include <netinet/in.h>
+#include <netinet/tcp.h>
+#include <poll.h>
+#include <signal.h>
+#include <sys/socket.h>
+#include <unistd.h>
+#endif
+
+#ifndef MSG_NOSIGNAL
+#define MSG_NOSIGNAL 0
+#endif
+
+#define NET_SEND_TIMEOUT_MS 20000
+
+static void set_err(char *err, size_t err_sz, const char *what) {
+  if (err && err_sz)
+    snprintf(err, err_sz, "%s", what);
+}
+
+#ifdef _WIN32
+static void set_err_win(char *err, size_t err_sz, const char *what) {
+  if (err && err_sz)
+    snprintf(err, err_sz, "%s (winsock error %d)", what, WSAGetLastError());
+}
+#define SET_ERR(e, n, w) set_err_win(e, n, w)
+#define NET_FD(s) ((SOCKET)(s))
+static void net_close_raw(os_socket_t s) { closesocket((SOCKET)s); }
+static int net_last_would_block(void) {
+  int e = WSAGetLastError();
+  return e == WSAEWOULDBLOCK || e == WSAEINTR;
+}
+static int net_last_conn_reset(void) {
+  int e = WSAGetLastError();
+  return e == WSAECONNRESET || e == WSAECONNABORTED || e == WSAENOTSOCK;
+}
+#else
+#define SET_ERR(e, n, w) set_err(e, n, w)
+#define NET_FD(s) ((int)(s))
+static void net_close_raw(os_socket_t s) { close(s); }
+static int net_last_would_block(void) { return errno == EINTR; }
+static int net_last_conn_reset(void) {
+  return errno == ECONNRESET || errno == EPIPE || errno == EBADF;
+}
+#endif
+
+int os_net_init(void) {
+#ifdef _WIN32
+  WSADATA w;
+  if (WSAStartup(MAKEWORD(2, 2), &w) != 0)
+    return -1;
+#else
+  /* A hung-up client must not kill the process with SIGPIPE. */
+  signal(SIGPIPE, SIG_IGN);
+#endif
+  return 0;
+}
+
+void os_net_cleanup(void) {
+#ifdef _WIN32
+  WSACleanup();
+#endif
+}
+
+/* poll, not select: FD_SET past FD_SETSIZE writes out of bounds, and a
+ * failed connect shows up as an error event, not as writable, on Windows. */
+static int wait_io(os_socket_t s, int want_write, int timeout_ms) {
+#ifdef _WIN32
+  WSAPOLLFD p;
+  p.fd = NET_FD(s);
+  p.events = want_write ? POLLWRNORM : POLLRDNORM;
+  p.revents = 0;
+  int rc = WSAPoll(&p, 1, timeout_ms < 0 ? -1 : timeout_ms);
+#else
+  struct pollfd p;
+  p.fd = NET_FD(s);
+  p.events = want_write ? POLLOUT : POLLIN;
+  p.revents = 0;
+  int rc = poll(&p, 1, timeout_ms < 0 ? -1 : timeout_ms);
+#endif
+  if (rc > 0)
+    return 1;
+  if (rc == 0)
+    return 0; /* timed out */
+  return -1;
+}
+
+/* Host to bind: AF_INET unless the name is an IPv6 literal, so "localhost"
+ * and "127.0.0.1" always mean the IPv4 loopback a browser will reach. */
+static int resolve_bind(const char *host, int port, struct sockaddr_storage *sa,
+                        socklen_t *sa_len) {
+  memset(sa, 0, sizeof(*sa));
+  int family = AF_INET;
+  const char *h = host && host[0] ? host : "127.0.0.1";
+  if (strchr(h, ':'))
+    family = AF_INET6;
+  struct addrinfo hints, *res = NULL;
+  memset(&hints, 0, sizeof(hints));
+  hints.ai_family = family;
+  hints.ai_socktype = SOCK_STREAM;
+  hints.ai_flags = AI_PASSIVE;
+  char portstr[16];
+  snprintf(portstr, sizeof(portstr), "%d", port);
+  if (getaddrinfo(h, portstr, &hints, &res) != 0 || !res)
+    return -1;
+  memcpy(sa, res->ai_addr, res->ai_addrlen);
+  *sa_len = (socklen_t)res->ai_addrlen;
+  freeaddrinfo(res);
+  return 0;
+}
+
+os_socket_t os_net_listen(const char *host, int port, char *err, size_t err_sz) {
+  struct sockaddr_storage sa;
+  socklen_t sa_len;
+  if (resolve_bind(host, port, &sa, &sa_len) != 0) {
+    SET_ERR(err, err_sz, "cannot resolve the bind address");
+    return OS_SOCKET_BAD;
+  }
+  os_socket_t s = (os_socket_t)socket(sa.ss_family, SOCK_STREAM, 0);
+  if (s == OS_SOCKET_BAD) {
+    SET_ERR(err, err_sz, "cannot create a socket");
+    return OS_SOCKET_BAD;
+  }
+  int one = 1;
+#ifdef _WIN32
+  /* SO_REUSEADDR on Windows lets another process bind the same port and
+   * take over our connections */
+  setsockopt(NET_FD(s), SOL_SOCKET, SO_EXCLUSIVEADDRUSE, (const char *)&one,
+             sizeof(one));
+#else
+  setsockopt(NET_FD(s), SOL_SOCKET, SO_REUSEADDR, (const char *)&one,
+             sizeof(one));
+#endif
+#ifdef IPV6_V6ONLY
+  if (sa.ss_family == AF_INET6)
+    setsockopt(NET_FD(s), IPPROTO_IPV6, IPV6_V6ONLY, (const char *)&one,
+               sizeof(one));
+#endif
+  if (bind(NET_FD(s), (const struct sockaddr *)&sa, sa_len) != 0) {
+    SET_ERR(err, err_sz, "the address is already in use");
+    net_close_raw(s);
+    return OS_SOCKET_BAD;
+  }
+  if (listen(NET_FD(s), 32) != 0) {
+    SET_ERR(err, err_sz, "listen failed");
+    net_close_raw(s);
+    return OS_SOCKET_BAD;
+  }
+  return s;
+}
+
+int os_net_port(os_socket_t s) {
+  struct sockaddr_storage sa;
+  socklen_t len = sizeof(sa);
+  if (getsockname(NET_FD(s), (struct sockaddr *)&sa, &len) != 0)
+    return 0;
+  if (sa.ss_family == AF_INET)
+    return ntohs(((struct sockaddr_in *)&sa)->sin_port);
+  if (sa.ss_family == AF_INET6)
+    return ntohs(((struct sockaddr_in6 *)&sa)->sin6_port);
+  return 0;
+}
+
+void os_net_peer(os_socket_t s, char *out, size_t out_sz) {
+  if (!out || out_sz == 0)
+    return;
+  out[0] = 0;
+  struct sockaddr_storage sa;
+  socklen_t len = sizeof(sa);
+  if (getpeername(NET_FD(s), (struct sockaddr *)&sa, &len) != 0)
+    return;
+  char ip[64] = "";
+  int port = 0;
+  if (sa.ss_family == AF_INET) {
+    if (!inet_ntop(AF_INET, &((struct sockaddr_in *)&sa)->sin_addr, ip,
+                   sizeof(ip)))
+      return;
+    port = ntohs(((struct sockaddr_in *)&sa)->sin_port);
+  } else if (sa.ss_family == AF_INET6) {
+    if (!inet_ntop(AF_INET6, &((struct sockaddr_in6 *)&sa)->sin6_addr, ip,
+                   sizeof(ip)))
+      return;
+    port = ntohs(((struct sockaddr_in6 *)&sa)->sin6_port);
+  } else {
+    return;
+  }
+  if (strchr(ip, ':'))
+    snprintf(out, out_sz, "[%s]:%d", ip, port);
+  else
+    snprintf(out, out_sz, "%s:%d", ip, port);
+}
+
+os_socket_t os_net_connect(const char *host, int port, int timeout_ms) {
+  const char *h = host && host[0] ? host : "127.0.0.1";
+  struct addrinfo hints, *res = NULL;
+  memset(&hints, 0, sizeof(hints));
+  hints.ai_family = strchr(h, ':') ? AF_INET6 : AF_INET;
+  hints.ai_socktype = SOCK_STREAM;
+  char portstr[16];
+  snprintf(portstr, sizeof(portstr), "%d", port);
+  if (getaddrinfo(h, portstr, &hints, &res) != 0 || !res)
+    return OS_SOCKET_BAD;
+  os_socket_t s = (os_socket_t)socket(res->ai_family, SOCK_STREAM, 0);
+  if (s == OS_SOCKET_BAD) {
+    freeaddrinfo(res);
+    return OS_SOCKET_BAD;
+  }
+  /* Bounded wait: connect() non-blocking, poll, then read SO_ERROR. A dead
+   * server must not hang the caller. */
+#ifdef _WIN32
+  u_long nb = 1;
+  ioctlsocket(NET_FD(s), FIONBIO, &nb);
+#else
+  int fl = fcntl(s, F_GETFL, 0);
+  fcntl(s, F_SETFL, fl | O_NONBLOCK);
+#endif
+  int rc = connect(NET_FD(s), res->ai_addr, (int)res->ai_addrlen);
+  freeaddrinfo(res);
+  if (rc != 0) {
+#ifdef _WIN32
+    int in_progress = WSAGetLastError() == WSAEWOULDBLOCK;
+#else
+    int in_progress = errno == EINPROGRESS || errno == EINTR;
+#endif
+    if (!in_progress || wait_io(s, 1, timeout_ms) <= 0) {
+      net_close_raw(s);
+      return OS_SOCKET_BAD;
+    }
+    int soerr = 0;
+    socklen_t elen = sizeof(soerr);
+    if (getsockopt(NET_FD(s), SOL_SOCKET, SO_ERROR, (char *)&soerr, &elen) != 0 ||
+        soerr != 0) {
+      net_close_raw(s);
+      return OS_SOCKET_BAD;
+    }
+  }
+#ifdef _WIN32
+  nb = 0;
+  ioctlsocket(NET_FD(s), FIONBIO, &nb);
+#else
+  fcntl(s, F_SETFL, fl);
+#endif
+  return s;
+}
+
+os_socket_t os_net_accept(os_socket_t s, int timeout_ms, int *timed_out) {
+  if (timed_out)
+    *timed_out = 0;
+  int w = wait_io(s, 0, timeout_ms);
+  if (w == 0) {
+    if (timed_out)
+      *timed_out = 1;
+    return OS_SOCKET_BAD;
+  }
+  if (w < 0)
+    return OS_SOCKET_BAD;
+  os_socket_t c = (os_socket_t)accept(NET_FD(s), NULL, NULL);
+  if (c == OS_SOCKET_BAD)
+    return OS_SOCKET_BAD;
+  int one = 1;
+  setsockopt(NET_FD(c), IPPROTO_TCP, TCP_NODELAY, (const char *)&one,
+             sizeof(one));
+  return c;
+}
+
+long os_net_recv(os_socket_t s, void *buf, size_t n, int timeout_ms) {
+  int w = wait_io(s, 0, timeout_ms);
+  if (w == 0)
+    return -2;
+  if (w < 0)
+    return -1;
+  size_t want = n > (size_t)(1 << 20) ? (size_t)(1 << 20) : n;
+  int r = (int)recv(NET_FD(s), (char *)buf, (int)want, 0);
+  if (r == 0)
+    return 0;
+  if (r < 0)
+    return net_last_conn_reset() ? 0 : (net_last_would_block() ? -2 : -1);
+  return r;
+}
+
+int os_net_send_all(os_socket_t s, const void *buf, size_t n, int timeout_ms) {
+  const char *p = buf;
+  size_t off = 0;
+  while (off < n) {
+    int w = wait_io(s, 1, timeout_ms ? timeout_ms : NET_SEND_TIMEOUT_MS);
+    if (w <= 0)
+      return -1;
+    long r = (long)send(NET_FD(s), p + off, (int)(n - off), MSG_NOSIGNAL);
+    if (r <= 0) {
+      if (net_last_would_block())
+        continue;
+      return -1;
+    }
+    off += (size_t)r;
+  }
+  return 0;
+}
+
+void os_net_shutdown(os_socket_t s) {
+#ifdef _WIN32
+  shutdown((SOCKET)s, SD_BOTH);
+#else
+  shutdown(s, SHUT_RDWR);
+#endif
+}
+
+void os_net_close(os_socket_t s) {
+  if (s == OS_SOCKET_BAD)
+    return;
+  net_close_raw(s);
+}

src/os_compat.c

@@ -519,7 +519,7 @@ int os_run_timeout_effective(void) {
   const char *e = getenv("CST_SHELL_TIMEOUT");
   if (e && atoi(e) > 0)
     return atoi(e);
-  return 120;
+  return 0; /* no limit: a build, a test suite or a long clone must finish */
 }
 
 double os_now_ms(void) {
@@ -662,7 +662,8 @@ int os_run_capture_ex(const char *cwd, const char *const argv[],
   int truncated = 0, killed = 0;
   int status = 0, reaped = 0, bg_killed = 0;
   double reaped_at = 0, killed_at = 0;
-  double deadline = os_now_ms() + os_run_timeout_effective() * 1000.0;
+  int timeout_s = os_run_timeout_effective();
+  double deadline = timeout_s > 0 ? os_now_ms() + timeout_s * 1000.0 : 0;
   for (;;) {
     struct pollfd pfd = {pipefd[0], POLLIN, 0};
     int pr = poll(&pfd, 1, 100);
@@ -692,7 +693,7 @@ int os_run_capture_ex(const char *cwd, const char *const argv[],
     }
     /* Checked on every pass, so a chatty command still stops promptly. */
     if (!reaped && !killed &&
-        ((g_run_cancel && *g_run_cancel) || now > deadline)) {
+        ((g_run_cancel && *g_run_cancel) || (deadline > 0 && now > deadline))) {
       killed = (g_run_cancel && *g_run_cancel) ? 2 : 1;
       killed_at = now;
       kill(-pid, SIGTERM);

src/os_win.c

@@ -601,7 +601,10 @@ int os_run_capture_token(const char *cwd, const char *const argv[],
   buf[0] = 0;
   int truncated = 0, killed = 0, reaped = 0, bg_killed = 0;
   double reaped_at = 0, killed_at = 0;
-  double deadline = os_now_ms() + os_run_timeout_effective() * 1000.0;
+  /* no timeout unless one was asked for: a deadline far in the future keeps
+   * every `now > deadline` check working without a second code path */
+  int timeout_s = os_run_timeout_effective();
+  double deadline = timeout_s > 0 ? os_now_ms() + timeout_s * 1000.0 : 1e18;
   int idle = 0; /* empty polls in a row: wait longer once output stalls */
   for (;;) {
     DWORD avail = 0;

src/provider_switch.c

@@ -0,0 +1,193 @@
+#include "cst_provider_switch.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+int cst_ps_fetch(const char *provider, const char *api_base, const char *key,
+                 char ids[][128], int max, char *err, size_t err_sz) {
+  cst_model_meta_t *m = calloc((size_t)(max > 0 ? max : 1), sizeof(*m));
+  if (!m) {
+    snprintf(err, err_sz, "out of memory");
+    return -1;
+  }
+  int r = cst_llm_list_models_meta(api_base, key, m, max, err, err_sz);
+  if (r >= 0)
+    cst_models_cache_save(provider, api_base, m, r);
+  for (int i = 0; i < r; i++)
+    snprintf(ids[i], 128, "%s", m[i].id);
+  free(m);
+  return r;
+}
+
+int cst_ps_cached(const char *provider, char ids[][128], int max, int *fresh) {
+  *fresh = 0;
+  cst_model_meta_t *m = calloc((size_t)(max > 0 ? max : 1), sizeof(*m));
+  if (!m)
+    return -1;
+  long age = 0;
+  int n = cst_models_cache_load(provider, m, max, &age);
+  for (int i = 0; i < n; i++)
+    snprintf(ids[i], 128, "%s", m[i].id);
+  free(m);
+  *fresh = n > 0 && age >= 0 && age < CST_MODELS_CACHE_TTL;
+  return n > 0 ? n : -1;
+}
+
+const char *cst_ps_base(const cst_agent_config_t *cfg, const char *provider) {
+  if (cfg && strcmp(cfg->provider, provider) == 0)
+    return cfg->api_base;
+  const cst_provider_t *p = cst_provider_find(provider);
+  return p ? p->api_base : "";
+}
+
+int cst_ps_models(const cst_agent_config_t *cfg, int refresh,
+                  cst_ps_model_t *out, int max, int *api_rc, char *err,
+                  size_t err_sz) {
+  err[0] = 0;
+  *api_rc = -1;
+  char(*ids)[128] = calloc(CST_PS_MAX_MODELS, sizeof(*ids));
+  if (!ids) {
+    snprintf(err, err_sz, "out of memory");
+    return 0;
+  }
+  int fresh = 0;
+  int na = cst_ps_cached(cfg->provider, ids, CST_PS_MAX_MODELS, &fresh);
+  *api_rc = na;
+  if ((refresh || !fresh) && cfg->api_key[0]) {
+    char(*got)[128] = calloc(CST_PS_MAX_MODELS, sizeof(*got));
+    if (got) {
+      int r = cst_ps_fetch(cfg->provider, cfg->api_base, cfg->api_key, got,
+                           CST_PS_MAX_MODELS, err, err_sz);
+      *api_rc = r;
+      if (r >= 0) {
+        memcpy(ids, got, (size_t)r * sizeof(*ids));
+        na = r;
+      }
+      free(got);
+    }
+  }
+  int n = 0;
+  for (int i = 0; i < na && n < max; i++, n++) {
+    snprintf(out[n].id, sizeof(out[n].id), "%s", ids[i]);
+    out[n].api = 1;
+    cst_model_info(cfg->provider, ids[i], &out[n].info);
+  }
+  char builtin[32][128];
+  int nb = cst_models_list(cfg->model, builtin, 32);
+  for (int i = 0; i < nb && n < max; i++) {
+    int dup = 0;
+    for (int k = 0; k < na && !dup; k++)
+      dup = strcmp(ids[k], builtin[i]) == 0;
+    if (dup)
+      continue;
+    memcpy(out[n].id, builtin[i], sizeof(out[n].id));
+    out[n].id[sizeof(out[n].id) - 1] = 0;
+    out[n].api = 0;
+    cst_model_info(cfg->provider, builtin[i], &out[n].info);
+    n++;
+  }
+  free(ids);
+  return n;
+}
+
+int cst_ps_verify_save(const char *provider, const char *api_base,
+                       const char *key, char ids[][128], int max, int *nids,
+                       char *err, size_t err_sz) {
+  err[0] = 0;
+  char(*own)[128] = NULL;
+  if (!ids) {
+    own = calloc(CST_PS_MAX_MODELS, sizeof(*own));
+    if (!own) {
+      snprintf(err, err_sz, "out of memory");
+      return CST_PS_UNVERIFIED | CST_PS_UNSAVED;
+    }
+    ids = own;
+    max = CST_PS_MAX_MODELS;
+  }
+  int r = cst_ps_fetch(provider, api_base, key, ids, max, err, err_sz);
+  free(own);
+  if (nids)
+    *nids = r;
+  if (r == CST_MODELS_AUTH)
+    return CST_MODELS_AUTH;
+  int flags = r < 0 ? CST_PS_UNVERIFIED : 0;
+  if (r >= 0)
+    err[0] = 0;
+  if (cst_creds_save(provider, key) != 0)
+    flags |= CST_PS_UNSAVED;
+  return flags;
+}
+
+void cst_ps_apply_provider(cst_agent_config_t *cfg, const cst_provider_t *p,
+                           const char *key) {
+  cst_settings_t s;
+  cst_settings_load(&s);
+  const char *model = cst_settings_model(&s, p->name);
+  snprintf(cfg->provider, sizeof(cfg->provider), "%s", p->name);
+  snprintf(cfg->api_base, sizeof(cfg->api_base), "%s", p->api_base);
+  snprintf(cfg->model, sizeof(cfg->model), "%s", model ? model : p->model);
+  snprintf(cfg->api_key, sizeof(cfg->api_key), "%s", key);
+  cst_settings_remember(cfg);
+}
+
+int cst_ps_switch_provider(cst_agent_config_t *cfg, const char *name,
+                           const char *key, char *err, size_t err_sz) {
+  err[0] = 0;
+  const cst_provider_t *p = name ? cst_provider_find(name) : NULL;
+  if (!p) {
+    snprintf(err, err_sz, "unknown provider '%.64s'", name ? name : "");
+    return CST_PS_UNKNOWN;
+  }
+  char k[512];
+  int flags = 0;
+  char(*ids)[128] = calloc(CST_PS_MAX_MODELS, sizeof(*ids));
+  int nids = -1;
+  if (key && key[0]) {
+    snprintf(k, sizeof(k), "%s", key);
+    flags = cst_ps_verify_save(p->name, p->api_base, k, ids,
+                               ids ? CST_PS_MAX_MODELS : 0, &nids, err,
+                               err_sz);
+    if (flags < 0) {
+      memset(k, 0, sizeof(k));
+      free(ids);
+      return flags;
+    }
+  } else if (cst_creds_resolve(p->name, k, sizeof(k), NULL) != 0) {
+    free(ids);
+    snprintf(err, err_sz, "%s needs an API key", p->name);
+    return CST_PS_NEED_KEY;
+  }
+  cst_ps_apply_provider(cfg, p, k);
+  memset(k, 0, sizeof(k));
+  if (!cfg->model[0] && ids) {
+    /* a router preset: the first model the API lists */
+    int fresh = 0;
+    if (nids <= 0)
+      nids = cst_ps_cached(p->name, ids, CST_PS_MAX_MODELS, &fresh);
+    if (nids <= 0) {
+      char e2[256];
+      nids = cst_ps_fetch(p->name, cfg->api_base, cfg->api_key, ids,
+                          CST_PS_MAX_MODELS, e2, sizeof(e2));
+    }
+    if (nids > 0) {
+      snprintf(cfg->model, sizeof(cfg->model), "%s", ids[0]);
+      cst_settings_remember(cfg);
+    }
+  }
+  free(ids);
+  return flags;
+}
+
+int cst_ps_set_key(cst_agent_config_t *cfg, const char *key, char *err,
+                   size_t err_sz) {
+  if (!key || !key[0]) {
+    snprintf(err, err_sz, "empty key");
+    return CST_PS_NEED_KEY;
+  }
+  int r = cst_ps_verify_save(cfg->provider, cfg->api_base, key, NULL, 0, NULL,
+                             err, err_sz);
+  if (r >= 0)
+    snprintf(cfg->api_key, sizeof(cfg->api_key), "%s", key);
+  return r;
+}

src/tui.c

@@ -11,6 +11,7 @@
 #include "cst_mcp.h"
 #include "cst_models.h"
 #include "cst_plugins.h"
+#include "cst_provider_switch.h"
 #include "cst_skills.h"
 #include "cst_subagent.h"
 #include "cst_textsel.h"
@@ -378,6 +379,7 @@ typedef struct {
   char *prompt_hist[PROMPT_HIST_CAP];
   int prompt_hist_n;
   int hist_browse;
+  char *hist_draft; /* the live input while browsing */
 
   /* Busy-turn FIFO */
   char *queue[QUEUE_CAP];
@@ -3163,28 +3165,16 @@ static size_t word_right(const tui_t *t, size_t at) {
   return at;
 }
 
-/* Move to the previous/next logical line keeping the column (in chars).
- * Returns 0 when already on the first/last line. */
+static size_t input_hit(const tui_t *t, int want_row, int want_col);
+
+/* Move to the previous/next visual row (soft-wrapped or not) keeping the
+ * screen column. Returns 0 when already on the first/last row. */
 static int input_line_move(tui_t *t, int dir) {
-  size_t ls = line_start(t, t->cur);
-  int col = 0;
-  for (size_t i = ls; i < t->cur; i += next_len(t, i))
-    col++;
-  size_t target;
-  if (dir < 0) {
-    if (ls == 0)
-      return 0;
-    target = line_start(t, ls - 1);
-  } else {
-    size_t le = line_end(t, t->cur);
-    if (le >= t->in_len)
-      return 0;
-    target = le + 1;
-  }
-  size_t te = line_end(t, target);
-  while (col-- > 0 && target < te)
-    target += next_len(t, target);
-  t->cur = target;
+  int cr, cc;
+  int rows = input_layout(t, NULL, NULL, &cr, &cc);
+  if (cr + dir < 0 || cr + dir >= rows)
+    return 0;
+  t->cur = input_hit(t, cr + dir, cc);
   return 1;
 }
 
@@ -3897,46 +3887,11 @@ static void am_store(const char *provider, int r, char ids[][128],
   pthread_mutex_unlock(&g_am_mu);
 }
 
-/* GET /models; a good answer also refreshes models-cache/<provider>.json. */
-static int fetch_models(const char *provider, const char *base,
-                        const char *key, char ids[][128], int max, char *err,
-                        size_t err_sz) {
-  cst_model_meta_t *m = calloc((size_t)max, sizeof(*m));
-  if (!m) {
-    snprintf(err, err_sz, "out of memory");
-    return -1;
-  }
-  int r = cst_llm_list_models_meta(base, key, m, max, err, err_sz);
-  if (r >= 0)
-    cst_models_cache_save(provider, base, m, r);
-  for (int i = 0; i < r; i++)
-    snprintf(ids[i], 128, "%s", m[i].id);
-  free(m);
-  return r;
-}
-
-/* The cached list for provider into ids: count, or -1; *fresh when it is
- * younger than CST_MODELS_CACHE_TTL. */
-static int cached_models(const char *provider, char ids[][128], int max,
-                         int *fresh) {
-  *fresh = 0;
-  cst_model_meta_t *m = calloc((size_t)max, sizeof(*m));
-  if (!m)
-    return -1;
-  long age = 0;
-  int n = cst_models_cache_load(provider, m, max, &age);
-  for (int i = 0; i < n; i++)
-    snprintf(ids[i], 128, "%s", m[i].id);
-  free(m);
-  *fresh = n > 0 && age >= 0 && age < CST_MODELS_CACHE_TTL;
-  return n > 0 ? n : -1;
-}
-
 static void *probe_thread(void *arg) {
   probe_arg_t *a = arg;
   static char ids[AM_MAX][128]; /* one probe runs per provider switch */
   char err[256];
-  int r = fetch_models(a->provider, a->base, a->key, ids, AM_MAX, err,
+  int r = cst_ps_fetch(a->provider, a->base, a->key, ids, AM_MAX, err,
                        sizeof(err));
   if (r >= 0 || r == CST_MODELS_AUTH || !a->have_cache)
     am_store(a->provider, r, ids, err);
@@ -3948,23 +3903,13 @@ static void *probe_thread(void *arg) {
   return NULL;
 }
 
-/* Base URL for a provider: the live config for the current one (it may be
- * overridden), else the preset. */
-static const char *provider_base(const tui_t *t, const char *provider) {
-  const cst_agent_config_t *cfg = cst_agent_config(t->agent);
-  if (cfg && strcmp(cfg->provider, provider) == 0)
-    return cfg->api_base;
-  const cst_provider_t *p = cst_provider_find(provider);
-  return p ? p->api_base : "";
-}
-
 static void start_probe(tui_t *t) {
   const cst_agent_config_t *cfg = cst_agent_config(t->agent);
   if (!cfg || !cfg->api_key[0] || g_am_threads > 0)
     return;
   static char ids[AM_MAX][128];
   int fresh = 0;
-  int nc = cached_models(cfg->provider, ids, AM_MAX, &fresh);
+  int nc = cst_ps_cached(cfg->provider, ids, AM_MAX, &fresh);
   pthread_mutex_lock(&g_am_mu);
   snprintf(g_am.provider, sizeof(g_am.provider), "%s", cfg->provider);
   pthread_mutex_unlock(&g_am_mu);
@@ -4050,13 +3995,14 @@ static int obtain_key(tui_t *t, const char *provider, char *out, size_t n,
     return 0;
   char dir[1024] = "the config directory";
   os_config_dir(dir, sizeof(dir));
-  const char *base = provider_base(t, provider);
+  const char *base = cst_ps_base(cst_agent_config(t->agent), provider);
   char title[128], hint[1300];
   snprintf(title, sizeof(title), "API key for %s", provider);
   snprintf(hint, sizeof(hint), "%.200s · saved to %.900s/credentials (only you "
                                "can read it)",
            base, dir);
   char note[300] = "";
+  int flags;
   for (;;) {
     if (secret_input_run(t, title, hint, out, n) != 0)
       return -1;
@@ -4064,8 +4010,10 @@ static int obtain_key(tui_t *t, const char *provider, char *out, size_t n,
     draw_status(t);
     static char ids[AM_MAX][128];
     char err[256];
-    int r = fetch_models(provider, base, out, ids, AM_MAX, err, sizeof(err));
-    if (r == CST_MODELS_AUTH) {
+    int r = -1;
+    flags = cst_ps_verify_save(provider, base, out, ids, AM_MAX, &r, err,
+                               sizeof(err));
+    if (flags == CST_MODELS_AUTH) {
       snprintf(hint, sizeof(hint), "%s rejected that key (%.200s) — paste "
                                    "another one",
                base, err);
@@ -4075,13 +4023,13 @@ static int obtain_key(tui_t *t, const char *provider, char *out, size_t n,
     snprintf(g_am.provider, sizeof(g_am.provider), "%s", provider);
     pthread_mutex_unlock(&g_am_mu);
     am_store(provider, r, ids, err);
-    if (r < 0)
+    if (flags & CST_PS_UNVERIFIED)
       snprintf(note, sizeof(note), "could not verify the key (%.200s); "
                                    "saved anyway",
                err);
     break;
   }
-  if (cst_creds_save(provider, out) != 0)
+  if (flags & CST_PS_UNSAVED)
     set_status(t, "warning: could not save the key; it is used for this run "
                   "only");
   else if (note[0])
@@ -4313,7 +4261,7 @@ static void cmd_model(tui_t *t, char *arg, int *quit) {
       set_status(t, "fetching the model list…");
       draw_status(t);
       char err[256];
-      na = fetch_models(cfg->provider, cfg->api_base, cfg->api_key, names,
+      na = cst_ps_fetch(cfg->provider, cfg->api_base, cfg->api_key, names,
                         AM_MAX, err, sizeof(err));
       pthread_mutex_lock(&g_am_mu);
       snprintf(g_am.provider, sizeof(g_am.provider), "%s", cfg->provider);
@@ -4447,15 +4395,8 @@ static void cmd_provider(tui_t *t, char *arg, int *quit) {
     push_system_line(t, "(provider unchanged: %s needs an API key)", p->name);
     return;
   }
-  cst_settings_t s;
-  cst_settings_load(&s);
-  const char *model = cst_settings_model(&s, p->name);
-  snprintf(cfg->provider, sizeof(cfg->provider), "%s", p->name);
-  snprintf(cfg->api_base, sizeof(cfg->api_base), "%s", p->api_base);
-  snprintf(cfg->model, sizeof(cfg->model), "%s", model ? model : p->model);
-  snprintf(cfg->api_key, sizeof(cfg->api_key), "%s", key);
+  cst_ps_apply_provider(cfg, p, key);
   memset(key, 0, sizeof(key));
-  cst_settings_remember(cfg);
   push_system_line(t, "(provider → %s · %s · model %s)", p->name, p->api_base,
                    cfg->model);
   if (am_models(p->name, NULL, 0) < 0)
@@ -5375,6 +5316,151 @@ static void cmd_quit(tui_t *t, char *arg, int *quit) {
   *quit = 1;
 }
 
+/* /web: the WebUI is a second front end for the same engine and is designed to
+ * run as its own process, so this starts one on the same workspace and leaves
+ * the TUI holding the terminal. The child gets a fresh session: two writers on
+ * one session file is the one thing that would corrupt it. */
+#ifndef _WIN32
+#include <fcntl.h>
+#include <sys/wait.h> /* waitpid: collect the WebUI started by the last /web */
+#endif
+#ifndef CST_WEBUI_DEFAULT_PORT
+#define CST_WEBUI_DEFAULT_PORT 8787 /* keep in sync with include/cst_webui.h */
+#endif
+static void cmd_web(tui_t *t, char *arg, int *quit) {
+  (void)arg;
+  (void)quit;
+#ifdef _WIN32
+  push_system_line(t, "open another terminal and run: cst-agent --web");
+#else
+  static pid_t spawned = 0;
+  if (spawned > 0) {          /* collect the one we started last time */
+    waitpid(spawned, NULL, WNOHANG);
+    spawned = 0;
+  }
+  char dir[4096], exe[4096 + 16];
+  if (os_exe_dir(dir, sizeof(dir)) != 0) {
+    push_system_line(t, "cannot locate this binary: run cst-agent --web yourself");
+    return;
+  }
+  snprintf(exe, sizeof(exe), "%s/cst-agent", dir);
+  const cst_agent_config_t *cfg = cst_agent_config(t->agent);
+  const char *ws = (cfg && cfg->workspace_root[0]) ? cfg->workspace_root : ".";
+  char logf[4096 + 32];
+  snprintf(logf, sizeof(logf), "%s/.agent", ws);
+  os_mkdir_p(logf, 0755);
+  snprintf(logf, sizeof(logf), "%s/.agent/webui.log", ws);
+  /* The WebUI talks to what this session talks to (-b, -m, a /provider
+   * switch...). The key goes through the environment, not argv (ps). */
+  const char *argv[20];
+  int na = 0;
+  argv[na++] = "cst-agent";
+  argv[na++] = "--web";
+  argv[na++] = "--workspace";
+  argv[na++] = ws;
+  argv[na++] = "--new";
+  if (cfg && cfg->provider[0])
+    argv[na++] = "--provider", argv[na++] = cfg->provider;
+  if (cfg && cfg->model[0])
+    argv[na++] = "--model", argv[na++] = cfg->model;
+  if (cfg && cfg->api_base[0])
+    argv[na++] = "--api-base", argv[na++] = cfg->api_base;
+  if (cfg && cfg->effort[0])
+    argv[na++] = "--effort", argv[na++] = cfg->effort;
+  if (cfg)
+    argv[na++] = "--approval", argv[na++] = cst_approval_name(cfg->approval);
+  argv[na] = NULL;
+  size_t nenv = 0;
+  while (environ[nenv])
+    nenv++;
+  char **envp = calloc(nenv + 2, sizeof(char *));
+  char keyvar[600] = "";
+  if (!envp) {
+    push_system_line(t, "out of memory");
+    return;
+  }
+  size_t ne = 0;
+  for (size_t i = 0; i < nenv; i++)
+    if (strncmp(environ[i], "CST_API_KEY=", 12) != 0)
+      envp[ne++] = environ[i];
+  if (cfg && cfg->api_key[0]) {
+    snprintf(keyvar, sizeof(keyvar), "CST_API_KEY=%s", cfg->api_key);
+    envp[ne++] = keyvar;
+  }
+  envp[ne] = NULL;
+  pid_t pid = fork();
+  if (pid < 0) {
+    free(envp);
+    push_system_line(t, "cannot start a second process");
+    return;
+  }
+  if (pid == 0) {
+    setsid();
+    /* never inherit the TUI's terminal: its output would land on our screen */
+    int in = open("/dev/null", O_RDONLY);
+    if (in >= 0 && in != STDIN_FILENO) {
+      dup2(in, STDIN_FILENO);
+      close(in);
+    }
+    int fd = open(logf, O_WRONLY | O_CREAT | O_TRUNC, 0600);
+    if (fd < 0)
+      fd = open("/dev/null", O_WRONLY);
+    if (fd >= 0) {
+      dup2(fd, STDOUT_FILENO);
+      dup2(fd, STDERR_FILENO);
+      if (fd > 2)
+        close(fd);
+    }
+    execve(exe, (char *const *)argv, envp);
+    _exit(127);
+  }
+  free(envp);
+  spawned = pid;
+  /* Wait (2 s at most) for the URL it prints once it listens: the port may
+   * come from CST_WEB_PORT, or it may not start at all. */
+  char url[512] = "";
+  int died = 0;
+  for (int i = 0; i < 40 && !url[0] && !died; i++) {
+    usleep(50000);
+    FILE *fp = fopen(logf, "r");
+    char line[512];
+    while (fp && fgets(line, sizeof(line), fp)) {
+      if (strncmp(line, "webui: http", 11) == 0) {
+        line[strcspn(line, "\r\n")] = 0;
+        snprintf(url, sizeof(url), "%s", line + 7);
+        break;
+      }
+    }
+    if (fp)
+      fclose(fp);
+    if (!url[0] && waitpid(pid, NULL, WNOHANG) == pid) {
+      died = 1;
+      spawned = 0;
+    }
+  }
+  /* its --new made the WebUI's session CURRENT; a plain `cst-agent` later
+   * should still resume this one */
+  char sid[128] = "";
+  cst_agent_session_id(t->agent, sid, sizeof(sid));
+  if (sid[0])
+    cst_agent_store_set_current(ws, sid);
+  char msg[4200];
+  if (died)
+    snprintf(msg, sizeof(msg), "the WebUI did not start (port busy?): see %s",
+             logf);
+  else if (url[0])
+    snprintf(msg, sizeof(msg), "WebUI started in the background: %s (new session)",
+             url);
+  else
+    snprintf(msg, sizeof(msg), "WebUI starting in the background "
+                               "(http://127.0.0.1:%d/ unless CST_WEB_PORT says "
+                               "otherwise)", CST_WEBUI_DEFAULT_PORT);
+  push_system_line(t, msg);
+  snprintf(msg, sizeof(msg), "its log: %s", logf);
+  push_system_line(t, msg);
+#endif
+}
+
 static const slash_cmd_t SLASH[] = {
     {"help", "?", "", "list commands and keys", 1, cmd_help},
     {"new", "clear", "", "start a new session", 0, cmd_new},
@@ -5433,6 +5519,8 @@ static const slash_cmd_t SLASH[] = {
      cmd_agents},
     {"skill", "", "<name> [task]", "run a task with a skill loaded", 0,
      cmd_skill},
+    {"web", "", "", "start the WebUI in the background (127.0.0.1:8787)", 1,
+     cmd_web},
     {"quit", "q exit", "", "quit", 1, cmd_quit},
 };
 
@@ -5854,7 +5942,7 @@ static int submit_input(tui_t *t, int busy) {
 static size_t input_hit(const tui_t *t, int want_row, int want_col) {
   int row = 0, col = IN_COL;
   int right = (t->in_w > 0 ? t->in_w : t->cols) - 2;
-  size_t i = 0, best = t->in_len;
+  size_t i = 0, best = t->in_len, last = 0;
   while (i < t->in_len) {
     if (row == want_row && col >= want_col)
       return i;
@@ -5873,12 +5961,14 @@ static size_t input_hit(const tui_t *t, int want_row, int want_col) {
     utf8_decode(t->inbuf + i, t->in_len - i, &wc, &eaten);
     int cw = utf8_width(wc == L'\t' ? L' ' : wc);
     if (col + cw > right && col > IN_COL) {
+      /* offset i would be drawn at the start of the next row */
       if (row == want_row)
-        return i;
+        return last;
       row++;
       col = IN_COL;
     }
     best = i + eaten;
+    last = i;
     col += cw;
     i += eaten;
   }
@@ -6395,29 +6485,65 @@ static int g_k_alt_up, g_k_alt_down, g_k_ctrl_home, g_k_ctrl_end,
     g_k_ctrl_up, g_k_ctrl_down, g_k_ctrl_left, g_k_ctrl_right, g_k_alt_left,
     g_k_alt_right, g_k_ctrl_dc, g_k_alt_dc;
 
-static int ext_key(const char *cap) {
+/* The terminfo key, else the xterm sequence bound to our own code `own`. */
+static int ext_key(const char *cap, const char *xterm, int own) {
   char *s = tigetstr(cap);
-  if (!s || s == (char *)-1)
-    return 0;
-  int k = key_defined(s);
-  return k > 0 ? k : 0;
+  int k = s && s != (char *)-1 ? key_defined(s) : 0;
+  if (k > 0)
+    return k;
+#ifndef _WIN32
+  k = key_defined(xterm);
+  if (k > 0)
+    return k;
+  if (k == 0 && define_key(xterm, own) == OK)
+    return own;
+#else
+  (void)xterm, (void)own;
+#endif
+  return 0;
+}
+#endif
+
+#if defined(NCURSES_VERSION) && !defined(_WIN32)
+/* Terminfo lists one form per key, but terminals disagree: cursor keys come
+ * as CSI or SS3 whatever DECCKM says (tmux, mosh, a wrong TERM), Home/End as
+ * CSI H, SS3 H or CSI 1~/7~. Unknown forms would be read as Esc + text. */
+static void vt_keys_init(void) {
+  static const struct {
+    const char *seq;
+    int key;
+  } K[] = {
+      {"\033[A", KEY_UP},     {"\033OA", KEY_UP},     {"\033[B", KEY_DOWN},
+      {"\033OB", KEY_DOWN},   {"\033[C", KEY_RIGHT},  {"\033OC", KEY_RIGHT},
+      {"\033[D", KEY_LEFT},   {"\033OD", KEY_LEFT},   {"\033[H", KEY_HOME},
+      {"\033OH", KEY_HOME},   {"\033[1~", KEY_HOME},  {"\033[7~", KEY_HOME},
+      {"\033[F", KEY_END},    {"\033OF", KEY_END},    {"\033[4~", KEY_END},
+      {"\033[8~", KEY_END},   {"\033[2~", KEY_IC},    {"\033[3~", KEY_DC},
+      {"\033[5~", KEY_PPAGE}, {"\033[6~", KEY_NPAGE}, {"\033[Z", KEY_BTAB},
+      {"\033[1;2A", KEY_SR},  {"\033[1;2B", KEY_SF},  {"\033[1;2C", KEY_SRIGHT},
+      {"\033[1;2D", KEY_SLEFT},
+  };
+  for (size_t i = 0; i < sizeof(K) / sizeof(K[0]); i++)
+    if (key_defined(K[i].seq) == 0)
+      define_key(K[i].seq, K[i].key);
 }
 #endif
 
 static void keys_init(void) {
 #ifdef NCURSES_VERSION
-  g_k_alt_up = ext_key("kUP3");
-  g_k_alt_down = ext_key("kDN3");
-  g_k_ctrl_home = ext_key("kHOM5");
-  g_k_ctrl_end = ext_key("kEND5");
-  g_k_ctrl_up = ext_key("kUP5");
-  g_k_ctrl_down = ext_key("kDN5");
-  g_k_ctrl_left = ext_key("kLFT5");
-  g_k_ctrl_right = ext_key("kRIT5");
-  g_k_alt_left = ext_key("kLFT3");
-  g_k_alt_right = ext_key("kRIT3");
-  g_k_ctrl_dc = ext_key("kDC5");
-  g_k_alt_dc = ext_key("kDC3");
+  int own = KEY_OWN_BASE + 4; /* up to KEY_KITTY_BASE - 1 */
+  g_k_alt_up = ext_key("kUP3", "\033[1;3A", own++);
+  g_k_alt_down = ext_key("kDN3", "\033[1;3B", own++);
+  g_k_ctrl_home = ext_key("kHOM5", "\033[1;5H", own++);
+  g_k_ctrl_end = ext_key("kEND5", "\033[1;5F", own++);
+  g_k_ctrl_up = ext_key("kUP5", "\033[1;5A", own++);
+  g_k_ctrl_down = ext_key("kDN5", "\033[1;5B", own++);
+  g_k_ctrl_left = ext_key("kLFT5", "\033[1;5D", own++);
+  g_k_ctrl_right = ext_key("kRIT5", "\033[1;5C", own++);
+  g_k_alt_left = ext_key("kLFT3", "\033[1;3D", own++);
+  g_k_alt_right = ext_key("kRIT3", "\033[1;3C", own++);
+  g_k_ctrl_dc = ext_key("kDC5", "\033[3;5~", own++);
+  g_k_alt_dc = ext_key("kDC3", "\033[3;3~", own++);
 #endif
 }
 
@@ -6630,6 +6756,7 @@ static void tui_free(tui_t *t) {
   queue_free_all(t);
   for (int i = 0; i < t->prompt_hist_n; i++)
     free(t->prompt_hist[i]);
+  free(t->hist_draft);
   for (int i = 0; i < t->nfiles; i++)
     free(t->files[i]);
   free(t->files);
@@ -6720,6 +6847,7 @@ int cst_tui_run(cst_agent_t *agent) {
   /* the Windows console driver reads key events, not VT sequences */
   define_key("\033[200~", KEY_PASTE_BEGIN);
   define_key("\033[201~", KEY_PASTE_END);
+  vt_keys_init();
   printf("\033[?2004h");
   fflush(stdout);
   kitty_init();
@@ -7032,15 +7160,22 @@ int cst_tui_run(cst_agent_t *agent) {
           if (t.prompt_hist_n <= 0)
             continue;
           int idx = t.hist_browse;
+          if (idx < 0) {
+            free(t.hist_draft);
+            t.hist_draft = strdup(t.inbuf);
+          }
           idx = idx < 0 ? t.prompt_hist_n - 1 : (idx > 0 ? idx - 1 : 0);
           hist_load_into_input(&t, idx);
         } else {
           if (t.hist_browse < 0)
             continue;
-          if (t.hist_browse + 1 >= t.prompt_hist_n)
-            input_clear(&t);
-          else
+          if (t.hist_browse + 1 >= t.prompt_hist_n) {
+            input_set(&t, t.hist_draft);
+            free(t.hist_draft);
+            t.hist_draft = NULL;
+          } else {
             hist_load_into_input(&t, t.hist_browse + 1);
+          }
         }
         after_edit(&t);
         continue;

src/util.c

@@ -99,6 +99,26 @@ size_t cst_utf8_floor(const char *s, size_t n) {
   return i - 1; /* sequence incomplete: cut before lead byte */
 }
 
+char *cst_utf8_hold_feed(cst_utf8_hold_t *h, const char *s, size_t n,
+                         size_t *out_len) {
+  size_t have = h->n;
+  char *all = malloc(have + n + 1);
+  if (!all)
+    return NULL;
+  memcpy(all, h->b, have);
+  if (n)
+    memcpy(all + have, s, n);
+  size_t len = have + n;
+  size_t cut = cst_utf8_floor(all, len);
+  if (len - cut > sizeof(h->b))
+    cut = len;
+  h->n = len - cut;
+  memcpy(h->b, all + cut, h->n);
+  all[cut] = 0;
+  *out_len = cut;
+  return all;
+}
+
 static size_t utf8_seq_ok(const unsigned char *p, size_t left) {
   unsigned char c = p[0];
   size_t need;

src/web_md.c

@@ -0,0 +1,329 @@
+#ifndef _DEFAULT_SOURCE
+#define _DEFAULT_SOURCE
+#endif
+
+#include "cst_web_md.h"
+
+#include "cst_util.h"
+#include "sashidown.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+struct cst_web_md {
+  sashidown_stream_t *st;
+  cst_web_md_emit_fn emit;
+  void *ud;
+  cst_buf_t in;    /* HTML from sashidown, waiting for a complete tag */
+  int failed;
+};
+
+/* --- sanitizer ------------------------------------------------------------ */
+
+static const char *ALLOWED_TAGS[] = {
+    "p",      "br",     "hr",     "h1",     "h2",      "h3",     "h4",
+    "h5",     "h6",     "strong", "em",     "b",       "i",      "u",
+    "s",      "del",    "code",   "pre",    "blockquote", "ul",  "ol",
+    "li",     "table",  "thead",  "tbody",  "tr",      "th",     "td",
+    "a",      "img",    "span",   "div",    NULL};
+
+static int tag_allowed(const char *name, size_t n) {
+  for (int i = 0; ALLOWED_TAGS[i]; i++) {
+    if (strlen(ALLOWED_TAGS[i]) == n && strncmp(name, ALLOWED_TAGS[i], n) == 0)
+      return 1;
+  }
+  return 0;
+}
+
+/* A URL we are willing to put in href: http(s), mailto, or a same-site
+ * path. Anything else (javascript:, data:, file:, //evil) is dropped. */
+static int href_ok(const char *v, size_t n) {
+  if (n == 0)
+    return 0;
+  if (!strncmp(v, "http://", 7) || !strncmp(v, "https://", 8) ||
+      !strncmp(v, "mailto:", 7))
+    return 1;
+  if (v[0] == '/' && !(n > 1 && (v[1] == '/' || v[1] == '\\')))
+    return 1;
+  if (v[0] == '#')
+    return 1;
+  return 0;
+}
+
+static int src_ok(const char *v, size_t n) {
+  if (n > 11 && !strncmp(v, "data:image/", 11))
+    return 1;
+  if (n > 12 && !strncmp(v, "/api/attach/", 12))
+    return 1;
+  return 0;
+}
+
+static void attr_append(cst_buf_t *b, const char *name, size_t nlen,
+                        const char *val, size_t vlen) {
+  cst_buf_puts(b, " ");
+  cst_buf_append(b, name, nlen);
+  cst_buf_puts(b, "=\"");
+  /* a single-quoted or bare value may hold '"': it must not end ours */
+  for (size_t i = 0; i < vlen; i++) {
+    if (val[i] == '"')
+      cst_buf_puts(b, "&quot;");
+    else if (val[i] == '<')
+      cst_buf_puts(b, "&lt;");
+    else if (val[i] == '>')
+      cst_buf_puts(b, "&gt;");
+    else
+      cst_buf_append(b, val + i, 1);
+  }
+  cst_buf_puts(b, "\"");
+}
+
+/* Rewrite one tag (open, close or self-closing) into out; may emit nothing
+ * when the tag is not allowed. Returns 0, or -1 on OOM. */
+static int sanitize_tag(const char *tag, size_t len, cst_buf_t *out) {
+  /* <name ...> or </name> or <!-- ... --> */
+  size_t i = 0;
+  int closing = 0;
+  if (len < 3 || tag[0] != '<')
+    return 0;
+  i = 1;
+  if (i < len && tag[i] == '/') {
+    closing = 1;
+    i++;
+  }
+  size_t name_start = i;
+  while (i < len && ((tag[i] >= 'a' && tag[i] <= 'z') ||
+                     (tag[i] >= 'A' && tag[i] <= 'Z') ||
+                     (tag[i] >= '0' && tag[i] <= '9')))
+    i++;
+  size_t name_len = i - name_start;
+  if (name_len == 0 || !tag_allowed(tag + name_start, name_len))
+    return 0; /* comments, <script>, <iframe>: dropped entirely */
+  int is_img = name_len == 3 && strncmp(tag + name_start, "img", 3) == 0;
+  int is_a = name_len == 1 && tag[name_start] == 'a';
+  int self_closing = len > 2 && tag[len - 2] == '/';
+  size_t start = out->len;
+  cst_buf_puts(out, "<");
+  if (closing)
+    cst_buf_puts(out, "/");
+  cst_buf_append(out, tag + name_start, name_len);
+  if (!closing) {
+    /* copy the allowed attributes */
+    int keep_target = 0; /* a link is only useful with a usable href */
+    size_t p = name_start + name_len;
+    while (p < len && tag[p] != '>' && !(tag[p] == '/' && p + 1 < len &&
+                                         tag[p + 1] == '>')) {
+      while (p < len && (tag[p] == ' ' || tag[p] == '\t' || tag[p] == '\n'))
+        p++;
+      size_t an = p;
+      while (p < len && tag[p] != '=' && tag[p] != ' ' && tag[p] != '\t' &&
+             tag[p] != '\n' && tag[p] != '>' && tag[p] != '/')
+        p++;
+      size_t an_len = p - an;
+      if (an_len == 0) {
+        p++;
+        continue;
+      }
+      const char *aval = "";
+      size_t av_len = 0;
+      if (p < len && tag[p] == '=') {
+        p++;
+        char quote = 0;
+        if (p < len && (tag[p] == '"' || tag[p] == '\''))
+          quote = tag[p++];
+        size_t as = p;
+        while (p < len &&
+               (quote ? tag[p] != quote : (tag[p] != ' ' && tag[p] != '>')))
+          p++;
+        aval = tag + as;
+        av_len = p - as;
+        if (quote && p < len)
+          p++; /* closing quote */
+      }
+      int keep = 0;
+      if (is_a && an_len == 4 && strncmp(tag + an, "href", 4) == 0) {
+        keep = href_ok(aval, av_len);
+        keep_target = keep;
+      } else if (is_img && an_len == 3 && strncmp(tag + an, "src", 3) == 0) {
+        keep = src_ok(aval, av_len);
+        keep_target = keep;
+      } else if (is_img && an_len == 3 && strncmp(tag + an, "alt", 3) == 0)
+        keep = 1;
+      else if (an_len == 5 && strncmp(tag + an, "class", 5) == 0 &&
+               av_len > 9 && !strncmp(aval, "language-", 9))
+        keep = 1;
+      if (keep)
+        attr_append(out, tag + an, an_len, aval, av_len);
+    }
+    if (is_a && keep_target) {
+      cst_buf_puts(out, " target=\"_blank\" rel=\"noopener noreferrer\"");
+    }
+    if ((is_a && !keep_target) || (is_img && !keep_target)) {
+      /* a link with an unusable href, or an image from nowhere: drop the
+       * whole tag and keep only its text */
+      out->len = start;
+      if (out->buf)
+        out->buf[start] = 0;
+      return 0;
+    }
+    cst_buf_puts(out, self_closing ? " /" : "");
+  }
+  cst_buf_puts(out, ">");
+  return 0;
+}
+
+/* Emit the text before every '<' and every tag through the sanitizer. */
+static void scan_chunk(cst_web_md_t *s, int final) {
+  if (!s->emit || s->failed)
+    return;
+  size_t i = 0, emitted = 0;
+  while (i < s->in.len) {
+    if (s->in.buf[i] != '<') {
+      i++;
+      continue;
+    }
+    size_t close = i;
+    while (close < s->in.len && s->in.buf[close] != '>')
+      close++;
+    if (close >= s->in.len && !final)
+      break; /* the tag may still be growing: keep it buffered */
+    if (i > emitted && s->emit(s->in.buf + emitted, i - emitted, s->ud) != 0)
+      s->failed = 1;
+    size_t tag_len = close >= s->in.len ? s->in.len - i : close - i + 1;
+    cst_buf_t tag = {0};
+    if (sanitize_tag(s->in.buf + i, tag_len, &tag) != 0) {
+      cst_buf_free(&tag);
+      s->failed = 1;
+      return;
+    }
+    if (tag.len && s->emit(tag.buf, tag.len, s->ud) != 0)
+      s->failed = 1;
+    cst_buf_free(&tag);
+    i = i + tag_len;
+    emitted = i;
+    if (s->failed)
+      break;
+  }
+  if (emitted > 0) {
+    memmove(s->in.buf, s->in.buf + emitted, s->in.len - emitted);
+    s->in.len -= emitted;
+    s->in.buf[s->in.len] = 0;
+  }
+}
+
+/* Collector used by the one-shot sanitizer. */
+typedef struct {
+  cst_buf_t *b;
+  int failed;
+} md_collect_t;
+
+static int collect_emit(const char *html, size_t len, void *ud) {
+  md_collect_t *c = ud;
+  if (cst_buf_append(c->b, html, len) != 0) {
+    c->failed = 1;
+    return 1;
+  }
+  return 0;
+}
+
+char *cst_web_md_sanitize(const char *html, size_t len) {
+  cst_web_md_t tmp;
+  memset(&tmp, 0, sizeof(tmp));
+  cst_buf_t out = {0};
+  md_collect_t col = {&out, 0};
+  if (cst_buf_append(&tmp.in, html ? html : "", len) != 0)
+    return NULL;
+  tmp.emit = collect_emit;
+  tmp.ud = &col;
+  scan_chunk(&tmp, 1);
+  cst_buf_free(&tmp.in);
+  if (col.failed) {
+    cst_buf_free(&out);
+    return NULL;
+  }
+  return cst_buf_steal(&out);
+}
+
+/* --- sashidown glue ------------------------------------------------------- */
+
+static int on_sashidown_html(const char *html, size_t len, void *user) {
+  cst_web_md_t *s = user;
+  if (cst_buf_append(&s->in, html, len) != 0) {
+    s->failed = 1;
+    return 1;
+  }
+  scan_chunk(s, 0);
+  return s->failed ? 1 : 0;
+}
+
+cst_web_md_t *cst_web_md_new(cst_web_md_emit_fn emit, void *ud) {
+  cst_web_md_t *s = calloc(1, sizeof(*s));
+  if (!s)
+    return NULL;
+  sashidown_stream_opts_t opts;
+  memset(&opts, 0, sizeof(opts));
+  opts.on_html = on_sashidown_html;
+  opts.user = s;
+  s->st = sashidown_stream_create(&opts);
+  if (!s->st) {
+    free(s);
+    return NULL;
+  }
+  s->emit = emit;
+  s->ud = ud;
+  return s;
+}
+
+int cst_web_md_feed(cst_web_md_t *s, const char *data, size_t len) {
+  if (!s || !s->st)
+    return -1;
+  if (len && sashidown_stream_feed(s->st, data, len) != 0) {
+    s->failed = 1;
+    return -1;
+  }
+  return s->failed ? -1 : 0;
+}
+
+int cst_web_md_finish(cst_web_md_t *s) {
+  if (!s || !s->st)
+    return -1;
+  if (sashidown_stream_finish(s->st) != 0) {
+    s->failed = 1;
+    return -1;
+  }
+  scan_chunk(s, 1);
+  return s->failed ? -1 : 0;
+}
+
+void cst_web_md_free(cst_web_md_t *s) {
+  if (!s)
+    return;
+  if (s->st)
+    sashidown_stream_destroy(s->st);
+  cst_buf_free(&s->in);
+  free(s);
+}
+
+char *cst_web_md_html(const char *md, size_t len) {
+  if (!md)
+    return NULL;
+  size_t cap = 65536;
+  char *out = NULL;
+  for (int attempt = 0; attempt < 8; attempt++) {
+    char *buf = malloc(cap);
+    if (!buf)
+      return NULL;
+    size_t out_len = 0;
+    if (sashidown_convert(md, len, buf, cap, &out_len, SASHIDOWN_FMT_HTML) ==
+        0) {
+      out = cst_web_md_sanitize(buf, out_len);
+      free(buf);
+      return out;
+    }
+    free(buf);
+    if (cap >= (size_t)8 * 1024 * 1024)
+      break;
+    cap *= 2;
+  }
+  return NULL;
+}

src/webui_stub.c

@@ -0,0 +1,15 @@
+#include "cst_agent.h"
+#include "cst_webui.h"
+
+#include <stdio.h>
+
+#ifdef CST_NO_WEBUI
+int cst_webui_run(cst_agent_t *agent, const cst_webui_opts_t *opts) {
+  (void)agent;
+  (void)opts;
+  fprintf(stderr,
+          "the WebUI was not built into this binary (CST_BUILD_WEBUI=OFF).\n"
+          "Use --headless --prompt '...' or build with make CST_WEBUI=1\n");
+  return 1;
+}
+#endif

tests/cst_test_util.h

@@ -19,6 +19,7 @@
 #include <io.h>
 #include <windows.h>
 #else
+#include <poll.h>
 #include <sys/stat.h>
 #include <unistd.h>
 #endif
@@ -177,7 +178,7 @@ static CST_TEST_UNUSED void test_sleep_ms(int ms) {
 #ifdef _WIN32
   Sleep((DWORD)ms);
 #else
-  usleep((useconds_t)ms * 1000);
+  poll(NULL, 0, ms); /* usleep is hidden by -std=c11 on musl */
 #endif
 }
 

tests/test_llm_body.c

@@ -1,5 +1,7 @@
 #include "cJSON.h"
 #include "cst_agent.h"
+#include "cst_attach.h"
+#include "cst_test_util.h"
 
 #include <stdio.h>
 #include <stdlib.h>
@@ -325,8 +327,88 @@ static void test_images_off(void) {
   cJSON_Delete(j);
 }
 
+/* Every element of every content array must be a typed part object. */
+static int parts_typed(const cJSON *root) {
+  const cJSON *m;
+  cJSON_ArrayForEach(m, cJSON_GetObjectItemCaseSensitive(root, "messages")) {
+    const cJSON *c = cJSON_GetObjectItemCaseSensitive(m, "content");
+    if (!cJSON_IsArray(c))
+      continue;
+    const cJSON *p;
+    cJSON_ArrayForEach(p, c) {
+      const cJSON *t = cJSON_GetObjectItemCaseSensitive(p, "type");
+      if (!cJSON_IsObject(p) || !cJSON_IsString(t))
+        return 0;
+    }
+  }
+  return 1;
+}
+
+static void test_images_on(void) {
+  char dir[1024], sess[1100], err[256];
+  if (test_tmpdir(dir, sizeof(dir), "cst-body") != 0) {
+    EXPECT(0, "scratch dir");
+    return;
+  }
+  snprintf(sess, sizeof(sess), "%s/s.jsonl", dir);
+  static const unsigned char png[32] = {0x89, 'P', 'N', 'G', '\r', '\n', 0x1A,
+                                        '\n', 0, 0, 0, 13, 'I', 'H', 'D', 'R',
+                                        0, 0, 0, 2, 0, 0, 0, 2};
+  cst_attachment_t *img = cst_attach_import_bytes(sess, "shot.png",
+                                                  (const char *)png,
+                                                  sizeof(png), err, sizeof(err));
+  EXPECT(img != NULL, "image stored");
+  if (!img)
+    return;
+  cst_attachment_t gone = *img;
+  snprintf(gone.rel, sizeof(gone.rel), "missing.png");
+  cst_message_t m[5];
+  m[0] = msg(CST_ROLE_USER, "what is this");
+  m[0].attach = img;
+  m[0].nattach = 1;
+  m[1] = msg(CST_ROLE_ASSISTANT, "");
+  m[1].tool_calls_json = "[{\"id\":\"c1\",\"type\":\"function\",\"function\":"
+                         "{\"name\":\"view_image\",\"arguments\":\"{}\"}}]";
+  m[2] = msg(CST_ROLE_TOOL, "view_image shot.png");
+  snprintf(m[2].tool_call_id, sizeof(m[2].tool_call_id), "c1");
+  m[2].attach = img;
+  m[2].nattach = 1;
+  m[3] = msg(CST_ROLE_USER, "and now?");
+  m[4] = msg(CST_ROLE_USER, "");
+  m[4].attach = &gone;
+  m[4].nattach = 1;
+  cst_llm_request_t r;
+  memset(&r, 0, sizeof(r));
+  r.model = "deepseek-flash";
+  r.msgs = m;
+  r.nmsgs = 5;
+  r.session_path = sess;
+  cJSON *j = build_req(&r);
+  EXPECT(parts_typed(j), "content arrays hold only typed parts");
+  const cJSON *p0 = cJSON_GetObjectItemCaseSensitive(
+      cJSON_GetArrayItem(cJSON_GetObjectItemCaseSensitive(j, "messages"), 0),
+      "content");
+  const cJSON *t0 = cJSON_GetObjectItemCaseSensitive(
+      cJSON_GetArrayItem(p0, 0), "text");
+  const cJSON *u1 = cJSON_GetObjectItemCaseSensitive(
+      cJSON_GetObjectItemCaseSensitive(cJSON_GetArrayItem(p0, 1), "image_url"),
+      "url");
+  EXPECT(cJSON_IsString(t0) && strcmp(t0->valuestring, "what is this") == 0,
+         "user text is a text part");
+  EXPECT(cJSON_IsString(u1) &&
+             strncmp(u1->valuestring, "data:image/png;base64,", 22) == 0,
+         "image is a data URL part");
+  char *s = cJSON_PrintUnformatted(j);
+  EXPECT(s && strstr(s, "and now?") && strstr(s, "not sent"),
+         "follow-up text and the missing-image note are parts too");
+  free(s);
+  cJSON_Delete(j);
+  cst_attach_free(img);
+}
+
 int main(void) {
   test_images_off();
+  test_images_on();
   test_reasoning_and_prefix();
   test_cache_hints();
   test_tools_snapshot();

tests/test_model_registry.c

@@ -44,6 +44,10 @@ static void test_builtin(void) {
   EXPECT(mi.has_price && mi.price.miss == 9.0 && mi.price.offpeak_half,
          "longer match wins the price");
   EXPECT(strcmp(mi.source, "built-in") == 0, "source built-in");
+  cst_model_info("deepseek", "deepseek-flash", &mi);
+  EXPECT(mi.context == 1000000 && mi.vision == 1, "deepseek-flash takes images");
+  cst_model_info("deepseek", "deepseek-chat", &mi);
+  EXPECT(mi.vision == 1, "deepseek-chat is deepseek-flash");
   cst_model_info(NULL, "gpt-4o-mini-2024", &mi);
   EXPECT(mi.price.miss == 0.15 && mi.context == 128000 && mi.vision == 1,
          "gpt-4o-mini over gpt-4o");
@@ -152,7 +156,7 @@ static void test_cache(void) {
              strcmp(mi.price.currency, "$") == 0,
          "API vision and USD price");
   cst_model_info("deepseek", "deepseek-flash", &mi);
-  EXPECT(mi.vision == 0 && mi.price.miss == 2.0,
+  EXPECT(mi.price.miss == 2.0 && strcmp(mi.price.currency, "¥") == 0,
          "another provider's cache is not used");
   EXPECT(cst_model_context_tokens("anthropic/claude-sonnet-4") == 1000000,
          "no provider: any cache, exact id");

tests/test_prefix_stable.c

@@ -210,7 +210,7 @@ static void test_vision(const char *root) {
   snprintf(cfg.workspace_root, sizeof(cfg.workspace_root), "%s", root);
   snprintf(cfg.session_path, sizeof(cfg.session_path), "%s/blind.jsonl", dir);
   snprintf(cfg.api_base, sizeof(cfg.api_base), "http://replay/v1");
-  snprintf(cfg.model, sizeof(cfg.model), "deepseek-flash");
+  snprintf(cfg.model, sizeof(cfg.model), "deepseek-v4-pro");
   snprintf(cfg.provider, sizeof(cfg.provider), "deepseek");
   cfg.protocol = CST_PROTO_NATIVE_EOS;
   cfg.max_steps = 4;
@@ -220,7 +220,7 @@ static void test_vision(const char *root) {
   EXPECT(a != NULL, "blind agent");
   if (!a)
     return;
-  EXPECT(cst_agent_sees_images(a) == 0, "deepseek-flash is text only");
+  EXPECT(cst_agent_sees_images(a) == 0, "deepseek-v4-pro is text only");
   EXPECT(cst_agent_attach_bytes(a, "shot.png", (const char *)png, pn, err,
                                 sizeof(err)) == 0,
          "attach to a blind model");

tests/test_webui.c

@@ -0,0 +1,320 @@
+/* WebUI: the socket/HTTP layer, the SSE framing, auth and the HTML
+ * sanitizer. The full turn path is covered by scripts/e2e_webui.sh, which
+ * runs a recorded stream through a real server. */
+#ifndef _DEFAULT_SOURCE
+#define _DEFAULT_SOURCE
+#endif
+
+#include "cst_http.h"
+#include "cst_net.h"
+#include "cst_util.h"
+#include "cst_web_md.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "cst_test_util.h"
+
+#define CHECK(c)                                                               \
+  do {                                                                         \
+    if (!(c)) {                                                                \
+      fprintf(stderr, "%s:%d: CHECK(%s) failed\n", __FILE__, __LINE__, #c);   \
+      exit(1);                                                                 \
+    }                                                                          \
+  } while (0)
+
+/* --- a tiny HTTP client --------------------------------------------------- */
+
+static os_socket_t connect_to(int port) {
+  char host[32];
+  snprintf(host, sizeof(host), "127.0.0.1");
+  os_socket_t s = os_net_connect(host, port, 3000);
+  CHECK(s != OS_SOCKET_BAD);
+  return s;
+}
+
+/* Sends data, reads until the peer closes (or 5 s pass). malloc'd. */
+static char *roundtrip(int port, const char *req, size_t *out_len) {
+  os_socket_t s = connect_to(port);
+  CHECK(os_net_send_all(s, req, strlen(req), 3000) == 0);
+  cst_buf_t b = {0};
+  char buf[4096];
+  for (;;) {
+    long n = os_net_recv(s, buf, sizeof(buf), 5000);
+    if (n <= 0)
+      break;
+    CHECK(cst_buf_append(&b, buf, (size_t)n) == 0);
+  }
+  os_net_close(s);
+  *out_len = b.len;
+  return cst_buf_steal(&b);
+}
+
+/* Same, but stops as soon as `needle` shows up. */
+
+/* --- handlers ------------------------------------------------------------- */
+
+static int h_hello(cst_http_conn_t *c, void *ud) {
+  (void)ud;
+  return cst_http_send_str(c, 200, "text/plain", "world");
+}
+
+static int h_secret(cst_http_conn_t *c, void *ud) {
+  (void)ud;
+  return cst_http_send_str(c, 200, "text/plain", "secret");
+}
+
+static int h_echo(cst_http_conn_t *c, void *ud) {
+  (void)ud;
+  const cst_http_req_t *r = cst_http_request(c);
+  char head[64];
+  snprintf(head, sizeof(head), "%zu:", r->body_len);
+  cst_buf_t b = {0};
+  cst_buf_puts(&b, head);
+  cst_buf_append(&b, r->body ? r->body : "", r->body_len);
+  int rc = cst_http_send(c, 200, "text/plain", b.buf, b.len);
+  cst_buf_free(&b);
+  return rc;
+}
+
+static int h_events(cst_http_conn_t *c, void *ud) {
+  (void)ud;
+  CHECK(cst_http_stream_begin(c, "text/event-stream") == 0);
+  CHECK(cst_http_stream_write(c, "event: one\ndata: {\"n\":1}\n\n", 26) == 0);
+  CHECK(cst_http_stream_write(c, "event: two\ndata: {\"n\":2}\n\n", 26) == 0);
+  CHECK(cst_http_stream_end(c) == 0);
+  return 0;
+}
+
+static int auth_bearer(const cst_http_req_t *req, void *ud) {
+  const char *want = ud;
+  if (!req->authorization)
+    return -1;
+  if (strncmp(req->authorization, "Bearer ", 7) != 0)
+    return -1;
+  return strcmp(req->authorization + 7, want) == 0 ? 0 : -1;
+}
+
+static void test_http(void) {
+  static const cst_http_route_t routes[] = {
+      {"GET", "/hello", h_hello, NULL, 0, 0},
+      {"GET", "/secret", h_secret, NULL, 1, 0},
+      {"GET", "/events", h_events, NULL, 0, 0},
+      {"POST", "/echo", h_echo, NULL, 0, 4096},
+  };
+  cst_http_opts_t o;
+  memset(&o, 0, sizeof(o));
+  o.host = "127.0.0.1";
+  o.port = 0;
+  o.max_body = 1024 * 1024;
+  o.routes = routes;
+  o.nroutes = (int)(sizeof(routes) / sizeof(routes[0]));
+  o.auth = auth_bearer;
+  o.auth_ud = (void *)"s3cret";
+  char err[256] = "";
+  cst_http_server_t *s = cst_http_start(&o, err, sizeof(err));
+  CHECK(s != NULL);
+  int port = cst_http_port(s);
+  CHECK(port > 0);
+
+  size_t n = 0;
+  char *resp = roundtrip(port, "GET /hello HTTP/1.1\r\nHost: x\r\n"
+                               "Connection: close\r\n\r\n", &n);
+  CHECK(strstr(resp, "200 OK") && strstr(resp, "world"));
+  free(resp);
+
+  /* keep-alive: two requests, one connection */
+  resp = roundtrip(port, "GET /hello HTTP/1.1\r\nHost: x\r\n\r\n"
+                         "GET /hello HTTP/1.1\r\nHost: x\r\n"
+                         "Connection: close\r\n\r\n", &n);
+  CHECK(strstr(resp, "200 OK"));
+  CHECK(strstr(resp, "200 OK") != strrchr(resp, '2')); /* twice */
+  {
+    int hits = 0;
+    for (const char *p = resp; (p = strstr(p, "200 OK")); p++)
+      hits++;
+    CHECK(hits == 2);
+  }
+  free(resp);
+
+  /* auth: without, then with the token */
+  resp = roundtrip(port, "GET /secret HTTP/1.1\r\nHost: x\r\n"
+                         "Connection: close\r\n\r\n", &n);
+  CHECK(strstr(resp, "401"));
+  free(resp);
+  resp = roundtrip(port, "GET /secret HTTP/1.1\r\nHost: x\r\n"
+                         "Authorization: Bearer s3cret\r\n"
+                         "Connection: close\r\n\r\n", &n);
+  CHECK(strstr(resp, "200") && strstr(resp, "secret"));
+  free(resp);
+
+  /* unknown route */
+  resp = roundtrip(port, "GET /nope HTTP/1.1\r\nHost: x\r\n"
+                         "Connection: close\r\n\r\n", &n);
+  CHECK(strstr(resp, "404"));
+  free(resp);
+  /* traversal is rejected before any handler runs */
+  resp = roundtrip(port, "GET /../etc/passwd HTTP/1.1\r\nHost: x\r\n"
+                         "Connection: close\r\n\r\n", &n);
+  CHECK(strstr(resp, "400") || strstr(resp, "404"));
+  free(resp);
+
+  /* a body round trip, and the per-route cap */
+  resp = roundtrip(port, "POST /echo HTTP/1.1\r\nHost: x\r\n"
+                         "Content-Length: 3\r\nConnection: close\r\n\r\nabc", &n);
+  CHECK(strstr(resp, "200") && strstr(resp, "3:abc"));
+  free(resp);
+  {
+    cst_buf_t b = {0};
+    cst_buf_puts(&b, "POST /echo HTTP/1.1\r\nHost: x\r\nContent-Length: 9999\r\n"
+                     "Connection: close\r\n\r\n");
+    for (int i = 0; i < 9999; i++)
+      cst_buf_puts(&b, "a");
+    resp = roundtrip(port, b.buf, &n);
+    CHECK(strstr(resp, "413"));
+    free(resp);
+    cst_buf_free(&b);
+  }
+
+  /* chunked streaming: the client must see the decoded frames */
+  resp = roundtrip(port, "GET /events HTTP/1.1\r\nHost: x\r\n"
+                         "Connection: close\r\n\r\n", &n);
+  CHECK(strstr(resp, "Transfer-Encoding: chunked"));
+  CHECK(strstr(resp, "event: one") && strstr(resp, "event: two"));
+  CHECK(strstr(resp, "\r\n0\r\n\r\n")); /* terminator */
+  free(resp);
+
+  cst_http_stop(s);
+  cst_http_free(s);
+}
+
+/* --- query parsing -------------------------------------------------------- */
+
+static void test_query(void) {
+  cst_http_req_t r;
+  memset(&r, 0, sizeof(r));
+  snprintf(r.query, sizeof(r.query), "token=a%%20b&since=17&x=");
+  char buf[64];
+  CHECK(cst_http_query(&r, "token", buf, sizeof(buf)) == 0);
+  CHECK(strcmp(buf, "a b") == 0);
+  CHECK(cst_http_query(&r, "since", buf, sizeof(buf)) == 0);
+  CHECK(strcmp(buf, "17") == 0);
+  CHECK(cst_http_query(&r, "nope", buf, sizeof(buf)) != 0);
+}
+
+/* --- markdown + sanitizer ------------------------------------------------- */
+
+static cst_buf_t g_md_out;
+
+static int collect_md(const char *html, size_t len, void *ud) {
+  (void)ud;
+  return cst_buf_append(&g_md_out, html, len) == 0 ? 0 : 1;
+}
+
+static void test_sanitize(void) {
+  char *html = cst_web_md_html("**hi** [x](javascript:alert(1)) "
+                               "![i](/api/attach/img-1.png)",
+                               strlen("**hi** [x](javascript:alert(1)) "
+                                      "![i](/api/attach/img-1.png)"));
+  CHECK(html != NULL);
+  CHECK(strstr(html, "<strong>hi</strong>"));
+  CHECK(strstr(html, "javascript:") == NULL);
+  CHECK(strstr(html, "/api/attach/img-1.png"));
+  free(html);
+
+  /* raw markup that sashidown would pass through gets stripped */
+  const char *evil = "<p>ok</p><script>alert(1)</script>"
+                     "<img src=\"http://evil.example/x.png\" alt=\"a\">"
+                     "<a href=\"//evil.example\">l</a>";
+  html = cst_web_md_sanitize(evil, strlen(evil));
+  CHECK(html != NULL);
+  CHECK(strstr(html, "ok"));
+  CHECK(strstr(html, "<script") == NULL);
+  CHECK(strstr(html, "evil.example/x.png") == NULL);
+  CHECK(strstr(html, "//evil.example") == NULL);
+  CHECK(strstr(html, "<a") == NULL); /* an unusable href drops the tag */
+  free(html);
+
+  /* a quote inside a single-quoted or bare value cannot open a new attribute */
+  const char *quoted = "<a href='http://x\" onmouseover=\"alert(1)'>q</a>"
+                       "<a href=http://y\"onmouseover=alert(2)//>r</a>";
+  html = cst_web_md_sanitize(quoted, strlen(quoted));
+  CHECK(html != NULL);
+  CHECK(strstr(html, "\" onmouseover") == NULL);
+  CHECK(strstr(html, "\"onmouseover") == NULL);
+  CHECK(strstr(html, "&quot;") != NULL);
+  free(html);
+
+  /* streaming: a javascript: link is dropped, a split tag never leaks */
+  {
+    cst_buf_free(&g_md_out);
+    const char *src = "see [x](javascript:alert(1)) and **ok**";
+    cst_web_md_t *md = cst_web_md_new(collect_md, NULL);
+    CHECK(md != NULL);
+    CHECK(cst_web_md_feed(md, src, 4) == 0); /* a deliberately split tag */
+    CHECK(cst_web_md_feed(md, src + 4, strlen(src) - 4) == 0);
+    CHECK(cst_web_md_finish(md) == 0);
+    cst_web_md_free(md);
+    const char *out = g_md_out.buf ? g_md_out.buf : "";
+    CHECK(strstr(out, "javascript:") == NULL);
+    CHECK(strstr(out, "href") == NULL);
+    CHECK(strstr(out, "see"));
+    CHECK(strstr(out, "<strong>ok</strong>") != NULL);
+    cst_buf_free(&g_md_out);
+  }
+  {
+    /* and the same text fed one byte at a time still comes out whole */
+    cst_buf_free(&g_md_out);
+    const char *md_src = "**bold** and [ok](https://example.com)";
+    cst_web_md_t *md = cst_web_md_new(collect_md, NULL);
+    CHECK(md != NULL);
+    for (size_t i = 0; i < strlen(md_src); i++)
+      CHECK(cst_web_md_feed(md, md_src + i, 1) == 0);
+    CHECK(cst_web_md_finish(md) == 0);
+    cst_web_md_free(md);
+    CHECK(strstr(g_md_out.buf, "<strong>bold</strong>"));
+    CHECK(strstr(g_md_out.buf, "https://example.com"));
+    CHECK(strstr(g_md_out.buf, "noopener") != NULL);
+    cst_buf_free(&g_md_out);
+  }
+}
+
+/* Streamed deltas cut anywhere must reach the browser as whole characters. */
+static void test_utf8_hold(void) {
+  const char *src = "思考：先读 main.c，再改 😀 emoji。";
+  size_t len = strlen(src);
+  for (size_t step = 1; step <= 5; step++) {
+    cst_utf8_hold_t h = {0};
+    cst_buf_t out = {0};
+    for (size_t i = 0; i < len; i += step) {
+      size_t n = len - i < step ? len - i : step;
+      size_t got = 0;
+      char *piece = cst_utf8_hold_feed(&h, src + i, n, &got);
+      CHECK(piece != NULL);
+      CHECK(h.n < 4);
+      CHECK(cst_utf8_valid(piece, got));
+      CHECK(cst_buf_append(&out, piece, got) == 0);
+      free(piece);
+    }
+    CHECK(h.n == 0);
+    CHECK(out.len == len && memcmp(out.buf, src, len) == 0);
+    cst_buf_free(&out);
+  }
+  /* a stray continuation byte is not held forever */
+  cst_utf8_hold_t h = {0};
+  size_t got = 0;
+  char *piece = cst_utf8_hold_feed(&h, "\x80\x80\x80\x80", 4, &got);
+  CHECK(piece && got == 4 && h.n == 0);
+  free(piece);
+}
+
+int main(void) {
+  CHECK(os_net_init() == 0);
+  test_utf8_hold();
+  test_query();
+  test_sanitize();
+  test_http();
+  printf("test_webui: OK\n");
+  return 0;
+}

web/index.html

@@ -0,0 +1,231 @@
+<!DOCTYPE html>
+<html lang="zh-CN" data-theme="dark">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1, viewport-fit=cover">
+<meta name="color-scheme" content="dark light">
+<title>CST-Agent</title>
+<link rel="stylesheet" href="style.css">
+</head>
+<body>
+<!-- one icon set, drawn as a sprite so every glyph shares the same 24px grid,
+     stroke weight and round caps: mixed emoji/unicode never looks like a family -->
+<svg id="atlas" aria-hidden="true" hidden>
+  <symbol id="i-menu" viewBox="0 0 24 24"><path d="M4 7h16M4 12h16M4 17h16"/></symbol>
+  <symbol id="i-panel-left" viewBox="0 0 24 24"><rect x="3.5" y="5" width="17" height="14" rx="3"/><path d="M9.5 5v14"/></symbol>
+  <symbol id="i-panel-right" viewBox="0 0 24 24"><rect x="3.5" y="5" width="17" height="14" rx="3"/><path d="M14.5 5v14"/></symbol>
+  <symbol id="i-chevron" viewBox="0 0 24 24"><path d="M6 9.5l6 5 6-5"/></symbol>
+  <symbol id="i-plus" viewBox="0 0 24 24"><path d="M12 5v14M5 12h14"/></symbol>
+  <symbol id="i-refresh" viewBox="0 0 24 24"><path d="M20 12a8 8 0 1 1-2.6-5.9M20 4.5V9h-4.5"/></symbol>
+  <symbol id="i-settings" viewBox="0 0 24 24"><path d="M4 6.5h5M15 6.5h5M4 12h7M17 12h3M4 17.5h5M15 17.5h5"/><circle cx="11.5" cy="6.5" r="2"/><circle cx="15" cy="12" r="2"/><circle cx="11.5" cy="17.5" r="2"/></symbol>
+  <symbol id="i-moon" viewBox="0 0 24 24"><path d="M20 14.6A8.6 8.6 0 0 1 9.4 4 8.6 8.6 0 1 0 20 14.6z"/></symbol>
+  <symbol id="i-sun" viewBox="0 0 24 24"><circle cx="12" cy="12" r="4.2"/><path d="M12 2.5v2.2M12 19.3v2.2M2.5 12h2.2M19.3 12h2.2M5.2 5.2l1.6 1.6M17.2 17.2l1.6 1.6M18.8 5.2l-1.6 1.6M6.8 17.2l-1.6 1.6"/></symbol>
+  <symbol id="i-more" viewBox="0 0 24 24"><circle cx="5.5" cy="12" r="1.6" class="f"/><circle cx="12" cy="12" r="1.6" class="f"/><circle cx="18.5" cy="12" r="1.6" class="f"/></symbol>
+  <symbol id="i-send" viewBox="0 0 24 24"><path d="M12 19.5V5M5.8 11.2 12 5l6.2 6.2"/></symbol>
+  <symbol id="i-stop" viewBox="0 0 24 24"><rect x="7" y="7" width="10" height="10" rx="2.5" class="f"/></symbol>
+  <symbol id="i-image" viewBox="0 0 24 24"><rect x="3.5" y="5" width="17" height="14" rx="3"/><path d="M4 15.5 8.6 11l4 4 2.8-2.8L20.5 16"/><circle cx="9" cy="9.4" r="1.5"/></symbol>
+  <symbol id="i-clipboard" viewBox="0 0 24 24"><rect x="8.5" y="3.5" width="7" height="4" rx="1.6"/><path d="M15.5 5.5h3v15h-13v-15h3M9 12h6M9 15.5h4"/></symbol>
+  <symbol id="i-list" viewBox="0 0 24 24"><path d="M4 6.8h2M4 12h2M4 17.2h2M9 6.8h11M9 12h11M9 17.2h11"/></symbol>
+  <symbol id="i-copy" viewBox="0 0 24 24"><rect x="8.5" y="8.5" width="11" height="11" rx="2.6"/><path d="M15.5 8.5V7a2.5 2.5 0 0 0-2.5-2.5H6.5A2.5 2.5 0 0 0 4 7v6.5A2.5 2.5 0 0 0 6.5 16h1.5"/></symbol>
+  <symbol id="i-compress" viewBox="0 0 24 24"><path d="M8 3.5V9H3M16 20.5V15h5M3 9h4M16 15h5"/></symbol>
+  <symbol id="i-undo" viewBox="0 0 24 24"><path d="M9.5 5.5 4 11l5.5 5.5"/><path d="M4 11h10.2a5.2 5.2 0 0 1 0 10.4H8.5"/></symbol>
+  <symbol id="i-close" viewBox="0 0 24 24"><path d="M6.5 6.5l11 11M17.5 6.5l-11 11"/></symbol>
+  <symbol id="i-check" viewBox="0 0 24 24"><path d="M5 12.6l4.6 4.6L19 7.6"/></symbol>
+  <symbol id="i-shield" viewBox="0 0 24 24"><path d="M12 3.5 5 6.2v5.4c0 4.2 2.9 7.5 7 8.9 4.1-1.4 7-4.7 7-8.9V6.2z"/></symbol>
+  <symbol id="i-bolt" viewBox="0 0 24 24"><path d="M13.4 3 5.5 13.6H11l-.9 7.4L18.5 10H13z"/></symbol>
+  <symbol id="i-alert" viewBox="0 0 24 24"><path d="M12 4 2.8 20h18.4z"/><path d="M12 10v4.4M12 17.2v.2"/></symbol>
+  <symbol id="i-cpu" viewBox="0 0 24 24"><rect x="7" y="7" width="10" height="10" rx="2.6"/><path d="M10 3v4M14 3v4M10 17v4M14 17v4M3 10h4M3 14h4M17 10h4M17 14h4"/></symbol>
+  <symbol id="i-lv1" viewBox="0 0 24 24"><path d="M5 18.5V13h3.4v5.5z" class="f"/></symbol>
+  <symbol id="i-lv2" viewBox="0 0 24 24"><path d="M5 18.5V13h3.4v5.5zM10.3 18.5V9.6h3.4v8.9z" class="f"/></symbol>
+  <symbol id="i-lv3" viewBox="0 0 24 24"><path d="M5 18.5V13h3.4v5.5zM10.3 18.5V9.6h3.4v8.9zM15.6 18.5V6.4h3.4v12.1z" class="f"/></symbol>
+  <symbol id="i-lv4" viewBox="0 0 24 24"><path d="M4 18.5V13h2.6v5.5zM8.7 18.5V10h2.6v8.5zM13.4 18.5V6.8h2.6v11.7zM18.1 18.5V4h2.6v14.5z" class="f"/></symbol>
+  <symbol id="i-file" viewBox="0 0 24 24"><path d="M13.5 3.5H7A2.5 2.5 0 0 0 4.5 6v12A2.5 2.5 0 0 0 7 20.5h10A2.5 2.5 0 0 0 19.5 18V9.5z"/><path d="M13.5 3.5v6h6"/></symbol>
+  <symbol id="i-clock" viewBox="0 0 24 24"><circle cx="12" cy="12" r="8.5"/><path d="M12 7.2V12l3.2 2"/></symbol>
+  <symbol id="i-folder" viewBox="0 0 24 24"><path d="M3.5 6.5A1.5 1.5 0 0 1 5 5h4l2 2.5h8A1.5 1.5 0 0 1 20.5 9v9A1.5 1.5 0 0 1 19 19.5H5A1.5 1.5 0 0 1 3.5 18z"/></symbol>
+</svg>
+
+<div id="app">
+
+  <!-- rail: identity, one primary action, sessions, identity again.
+       Everything else is a popover anchored where it belongs. -->
+  <aside id="side">
+   <div class="side-inner">
+    <div class="side-head">
+      <div class="brand">
+        <span class="brand-name">CST-Agent</span>
+      </div>
+      <button id="btn-collapse" class="icon" title="收起侧栏 (Ctrl+B)">
+        <svg class="i"><use href="#i-panel-left"/></svg>
+      </button>
+    </div>
+
+    <button id="btn-new" class="new-task">
+      <svg class="i"><use href="#i-plus"/></svg><span>新建任务</span>
+    </button>
+
+    <div class="side-sec">
+      <span>最近</span>
+      <button id="btn-refresh-sessions" class="icon tiny" title="刷新">
+        <svg class="i"><use href="#i-refresh"/></svg>
+      </button>
+    </div>
+    <div id="sessions-list" class="sessions"></div>
+
+    <div class="side-foot">
+      <div class="who">
+        <span class="avatar">C</span>
+        <span class="who-text">
+          <b id="foot-ws">workspace</b>
+          <i id="foot-session">-</i>
+        </span>
+      </div>
+      <button id="btn-settings" class="icon" title="设置">
+        <svg class="i"><use href="#i-settings"/></svg>
+      </button>
+    </div>
+   </div>
+  </aside>
+
+  <!-- main column -->
+  <main id="main">
+    <header id="topbar">
+      <button id="btn-side" class="icon" title="展开侧栏 (Ctrl+B)">
+        <svg class="i"><use href="#i-panel-left"/></svg>
+      </button>
+      <button id="btn-drawer" class="icon only-narrow" title="侧栏">
+        <svg class="i"><use href="#i-menu"/></svg>
+      </button>
+      <button id="chip-model" class="chip model" title="模型与推理强度">
+        <svg class="i"><use href="#i-cpu"/></svg><span class="lbl">-</span>
+        <svg class="i chev"><use href="#i-chevron"/></svg>
+      </button>
+      <span class="grow"></span>
+      <div class="seg" id="seg-mode" title="计划模式只读，不改文件">
+        <button data-mode="build">构建</button>
+        <button data-mode="plan">计划</button>
+      </div>
+      <button id="chip-approval" class="chip quiet" title="审批权限">
+        <svg class="i"><use href="#i-shield"/></svg><span class="lbl">auto</span>
+        <svg class="i chev"><use href="#i-chevron"/></svg>
+      </button>
+      <button id="btn-theme" class="icon" title="明暗切换">
+        <svg class="i only-dark"><use href="#i-moon"/></svg>
+        <svg class="i only-light"><use href="#i-sun"/></svg>
+      </button>
+      <button id="btn-panel" class="icon" title="运行详情 (Ctrl+J)">
+        <svg class="i"><use href="#i-panel-right"/></svg>
+      </button>
+      <button id="btn-menu" class="icon" title="更多">
+        <svg class="i"><use href="#i-more"/></svg>
+      </button>
+    </header>
+
+    <div id="scroll">
+      <section id="empty">
+        <p class="ready">准备好开始构建了吗</p>
+      </section>
+      <div id="log"></div>
+    </div>
+
+    <button id="to-bottom" class="round" hidden title="回到最新">
+      <svg class="i"><use href="#i-chevron"/></svg>
+    </button>
+
+    <footer id="composer">
+      <div id="pending" class="pending" hidden></div>
+
+      <!-- permission gate: inline, the turn really is waiting on it -->
+      <div id="gate" class="gate" hidden>
+        <div class="gate-head">
+          <span class="gate-title" id="gate-title">需要确认</span>
+          <span class="grow"></span>
+          <span class="gate-timer" id="gate-timer"></span>
+        </div>
+        <pre id="gate-detail"></pre>
+        <div class="gate-keys" id="gate-keys"></div>
+        <div class="gate-row" id="gate-buttons"></div>
+      </div>
+
+      <div class="box">
+        <textarea id="input" rows="1" placeholder="给 CST-Agent 派个任务…"></textarea>
+        <div class="box-row">
+          <button id="btn-plus" class="icon att" title="添加图片或计划模式">
+            <svg class="i"><use href="#i-plus"/></svg>
+          </button>
+          <button id="btn-undo" class="chip quiet" title="撤销上一次文件写入">
+            <svg class="i"><use href="#i-undo"/></svg><span>撤销</span>
+          </button>
+          <span class="grow"></span>
+          <button id="cost" class="cost" title="用量详情"></button>
+          <button id="btn-stop" class="icon stop" hidden title="停止">
+            <svg class="i"><use href="#i-stop"/></svg>
+          </button>
+          <button id="btn-send" class="round send" title="发送 (Enter)">
+            <svg class="i"><use href="#i-send"/></svg>
+          </button>
+        </div>
+      </div>
+      <div class="hint-line">
+        <span id="status">就绪</span>
+        <span class="grow"></span>
+        <span id="usage" class="quiet"></span>
+      </div>
+      <input type="file" id="file" accept="image/*" multiple hidden>
+    </footer>
+  </main>
+
+  <!-- right panel: the run, not the chat -->
+  <aside id="panel">
+   <div class="panel-inner">
+    <div class="panel-head">
+      <span>运行详情</span>
+      <span class="grow"></span>
+      <button id="btn-panel-close" class="icon tiny" title="关闭">
+        <svg class="i"><use href="#i-close"/></svg>
+      </button>
+    </div>
+
+    <section class="card" id="card-todo">
+      <h4>任务清单</h4>
+      <div id="todo-body" class="muted">模型还没有给出任务清单。</div>
+    </section>
+
+    <section class="card" id="card-files">
+      <h4>文件改动 <span class="badge" id="files-count">0</span></h4>
+      <div id="files-body" class="muted">本轮还没有改动文件。</div>
+    </section>
+
+    <section class="card">
+      <h4>上下文</h4>
+      <div class="kv"><span>占用</span><b id="kv-ctx">0</b></div>
+      <div class="kv"><span>缓存命中</span><b id="kv-cache">-</b></div>
+    </section>
+   </div>
+  </aside>
+
+</div>
+
+<div id="backdrop"></div>
+<div id="pop" class="pop" hidden></div>
+<div id="modal" hidden><div id="modal-box"></div></div>
+<div id="toast" class="toast" role="status" aria-live="polite" hidden></div>
+<script>
+/* Layout is decided here, before app.js runs and before the first paint: the
+ * rail must not flash open and then fold, and a broken script further down can
+ * no longer take the layout down with it. */
+(() => {
+  const q = new URLSearchParams(location.search);
+  const pick = (key) => {
+    const v = q.get(key);
+    if (v === '0' || v === '1') return v === '0';
+    return localStorage.getItem('cst.' + key) === '0';
+  };
+  if (pick('side')) document.body.classList.add('no-side');
+  if (pick('panel')) document.body.classList.add('no-panel');
+  const t = q.get('theme') || localStorage.getItem('cst.theme');
+  if (t) document.documentElement.dataset.theme = t;
+})();
+</script>
+<script src="app.js"></script>
+</body>
+</html>

web/style.css

@@ -0,0 +1,694 @@
+/* CST-Agent WebUI - one stylesheet, no build step.
+ *
+ * Design rules
+ *   - one fact, one home: sessions in the rail, model/approval/plan in the
+ *     top bar, the run in the right panel, attachments in the composer;
+ *   - settings open as popovers anchored to the control that owns them,
+ *     with icon + label + one line of explanation + a check on the current
+ *     value (the pattern all three reference apps use);
+ *   - motion is short and eased: 120ms for feedback, 180ms for entrances,
+ *     nothing bounces, and prefers-reduced-motion turns it all off. */
+
+/* Tokens are named the way the reference desktop app names them: four text
+ * levels, three surfaces per zone (main / sidebar / composer), and borders
+ * mixed out of the foreground rather than hard-coded greys - which is what
+ * keeps the dark and the light sheet on the same visual rhythm. */
+:root {
+  color-scheme: dark;
+
+  --text-primary: #f2f2f5;
+  --text-secondary: #a9a9b4;
+  --text-tertiary: #7d7d89;
+  --text-quaternary: #5e5e69;
+  --text-error: #f0616d;
+  --text-success: #3ecf95;
+
+  --main-surface-primary: #0d0d0f;
+  --main-surface-secondary: #141417;
+  --main-surface-tertiary: #1d1d22;
+
+  --sidebar-surface-primary: #111114;
+  --sidebar-surface-secondary: #18181c;
+  --sidebar-surface-tertiary: #202026;
+
+  --composer-surface-primary: #16161a;
+  --composer-surface-secondary: #202026;
+
+  --bg-code: #08080a;
+
+  /* every hover/active tint is alpha, so it lands correctly on any surface */
+  --hover: color-mix(in srgb, var(--text-primary) 7%, transparent);
+  --hover-strong: color-mix(in srgb, var(--text-primary) 13%, transparent);
+
+  --accent-primary: #3ecf95;
+  --accent-fg: #04241a;
+  --accent-weak: color-mix(in srgb, var(--accent-primary) 14%, transparent);
+  --accent-line: color-mix(in srgb, var(--accent-primary) 55%, transparent);
+  --link: var(--accent-primary);
+  --violet: #8b7bff;
+  --warn: #e0b341;
+  --selection-bg: color-mix(in srgb, var(--accent-primary) 30%, transparent);
+  --selection-fg: var(--text-primary);
+
+  --border-light: color-mix(in srgb, var(--text-primary) 14%, transparent);
+  --border-medium: color-mix(in srgb, var(--text-primary) 24%, transparent);
+  --border-heavy: color-mix(in srgb, var(--text-primary) 34%, transparent);
+
+  /* short aliases: the rest of the sheet keeps reading the same way */
+  --bg: var(--main-surface-primary);
+  --bg-soft: var(--main-surface-secondary);
+  --bg-lift: var(--main-surface-tertiary);
+  --bg-hover: var(--hover);
+  --fg: var(--text-primary);
+  --fg-dim: var(--text-secondary);
+  --fg-faint: var(--text-tertiary);
+  --fg-quat: var(--text-quaternary);
+  --line: var(--border-light);
+  --line-strong: var(--border-medium);
+  --accent: var(--accent-primary);
+  --danger: var(--text-error);
+
+  --rail: 252px;
+  --panel: 296px;
+  --radius: 12px;
+  --radius-lg: 22px;
+  --shadow: 0 16px 40px rgba(0, 0, 0, .45);
+  --shadow-sm: 0 6px 18px rgba(0, 0, 0, .3);
+  --mono: ui-monospace, SFMono-Regular, "JetBrains Mono", Menlo, Consolas, monospace;
+  --sans: system-ui, -apple-system, "Segoe UI", Roboto, "Noto Sans CJK SC", sans-serif;
+  --ease: cubic-bezier(.22, .61, .36, 1);
+  --ease-pop: cubic-bezier(.2, .9, .3, 1.06);   /* enters with a hint of overshoot */
+  --t-fast: 120ms;
+  --t: 180ms;
+}
+[data-theme="light"] {
+  color-scheme: light;
+
+  --text-primary: #14141a;
+  --text-secondary: #55555f;
+  --text-tertiary: #7c7c86;
+  --text-quaternary: #9c9ca6;
+  --text-error: #c8303f;
+  --text-success: #0f8f66;
+
+  --main-surface-primary: #ffffff;
+  --main-surface-secondary: #f7f7f9;
+  --main-surface-tertiary: #ffffff;
+
+  --sidebar-surface-primary: #fafafb;
+  --sidebar-surface-secondary: #f1f1f4;
+  --sidebar-surface-tertiary: #ffffff;
+
+  --composer-surface-primary: #ffffff;
+  --composer-surface-secondary: #f4f4f6;
+
+  --bg-code: #f5f5f7;
+
+  --accent-primary: #0f9d70;
+  --accent-fg: #ffffff;
+
+  --border-light: color-mix(in srgb, var(--text-primary) 12%, transparent);
+  --border-medium: color-mix(in srgb, var(--text-primary) 20%, transparent);
+  --border-heavy: color-mix(in srgb, var(--text-primary) 30%, transparent);
+
+  --warn: #a9761a;
+  --shadow: 0 16px 40px rgba(20, 20, 35, .14);
+  --shadow-sm: 0 6px 18px rgba(20, 20, 35, .1);
+}
+
+* { box-sizing: border-box; }
+html, body { height: 100%; margin: 0; }
+body {
+  background: var(--bg); color: var(--fg);
+  font: 15px/1.62 var(--sans);
+  -webkit-font-smoothing: antialiased;
+  -webkit-text-size-adjust: 100%;
+  overflow: hidden; /* columns scroll, never the page */
+}
+button, textarea, input { font: inherit; color: inherit; }
+button { cursor: pointer; background: none; border: 0; }
+button:disabled { cursor: default; }
+.grow { flex: 1 1 auto; }
+.quiet, .muted { color: var(--fg-dim); }
+.tiny { font-size: 12px; }
+[hidden] { display: none !important; }
+::selection { background: var(--selection-bg); color: var(--selection-fg); }
+
+/* one icon system: everything is drawn on the same 24px grid, stroked with
+ * the current text colour, so header, rail, popovers and composer match */
+svg.i {
+  width: 18px; height: 18px; flex: 0 0 auto; display: block;
+  fill: none; stroke: currentColor; stroke-width: 1.7;
+  stroke-linecap: round; stroke-linejoin: round;
+}
+svg.i .f { fill: currentColor; stroke: none; }
+.icon svg.i { width: 18px; height: 18px; }
+.icon.tiny svg.i { width: 15px; height: 15px; }
+.chip svg.i { width: 15px; height: 15px; }
+.chip .chev { width: 13px; height: 13px; opacity: .6; }
+.new-task svg.i { width: 17px; height: 17px; }
+.send svg.i { width: 19px; height: 19px; stroke-width: 2; }
+svg.i.only-light { display: none; }
+[data-theme="light"] svg.i.only-dark { display: none; }
+[data-theme="light"] svg.i.only-light { display: block; }
+
+/* the changed-file rows in the run panel: path, then +added / -removed */
+.fstat { display: flex; align-items: center; gap: 8px; padding: 5px 0; font-size: 13px; animation: fade-in var(--t) var(--ease) both; }
+.fstat .fimg { width: 15px; height: 15px; color: var(--text-quaternary); }
+.fstat .p {
+  flex: 1 1 auto; min-width: 0; overflow: hidden; text-overflow: ellipsis;
+  white-space: nowrap; font-family: var(--mono); font-size: 12px;
+}
+.fstat .a, .fstat .d { font-variant-numeric: tabular-nums; font-size: 12px; }
+.fstat .a { color: var(--text-success); }
+.fstat .d { color: var(--text-error); }
+
+::-webkit-scrollbar { width: 10px; height: 10px; }
+::-webkit-scrollbar-thumb {
+  background: var(--line-strong); border-radius: 99px;
+  border: 3px solid transparent; background-clip: content-box;
+}
+::-webkit-scrollbar-thumb:hover { background: var(--fg-faint); background-clip: content-box; }
+
+/* --- frame ---------------------------------------------------------------- */
+#app {
+  display: grid;
+  grid-template-columns: var(--rail) minmax(0, 1fr) var(--panel);
+  grid-template-rows: minmax(0, 1fr);
+  height: 100dvh;
+  /* the rail keeps its track and collapses to 0px, so the two states have the
+   * same track count and the column width can actually be interpolated */
+  transition: grid-template-columns 300ms var(--ease);
+}
+/* a zeroed track plus display:none used to slide #main into column one and
+ * crush it to nothing; the rail now stays in flow and is clipped instead */
+body.no-side #app { grid-template-columns: 0px minmax(0, 1fr) var(--panel); }
+body.no-panel #app { grid-template-columns: var(--rail) minmax(0, 1fr) 0px; }
+body.no-side.no-panel #app { grid-template-columns: 0px minmax(0, 1fr) 0px; }
+/* the expand control only exists while the rail is folded away, so exactly one
+ * toggle is ever on screen */
+#btn-side { display: none; }
+body.no-side #btn-side { display: inline-flex; }
+
+.icon {
+  width: 32px; height: 32px; border-radius: 9px; flex: 0 0 auto;
+  display: inline-flex; align-items: center; justify-content: center;
+  color: var(--fg-dim); font-size: 15px;
+  transition: background var(--t-fast), color var(--t-fast), transform 90ms var(--ease);
+}
+.icon:hover { background: var(--bg-hover); color: var(--fg); }
+.icon:active { transform: scale(.94); }
+.icon.tiny { width: 26px; height: 26px; font-size: 13px; }
+.round { border-radius: 50%; }
+.chip {
+  display: inline-flex; align-items: center; gap: 5px; height: 30px;
+  padding: 0 11px; border-radius: 99px; border: 1px solid transparent;
+  background: none; color: var(--fg-dim); font-size: 13px;
+  max-width: 280px; overflow: hidden; text-overflow: ellipsis; white-space: nowrap;
+  transition: background var(--t-fast), color var(--t-fast), border-color var(--t-fast), transform 90ms var(--ease);
+}
+.chip:hover { background: var(--bg-hover); color: var(--fg); }
+.chip:active { transform: scale(.97); }
+.chip.model { color: var(--fg); font-weight: 600; background: var(--bg-lift); border-color: var(--line); }
+.chip.model:hover { border-color: var(--line-strong); }
+.chip.warn { color: var(--warn); }
+.chip .chev { font-size: 11px; opacity: .65; }
+.badge {
+  min-width: 18px; height: 18px; padding: 0 5px; border-radius: 99px;
+  background: var(--accent-weak); color: var(--accent);
+  font-size: 11px; font-weight: 700;
+  display: inline-flex; align-items: center; justify-content: center;
+}
+
+/* --- rail ----------------------------------------------------------------- */
+#side {
+  background: var(--sidebar-surface-primary); border-right: 1px solid var(--border-light);
+  min-width: 0; overflow: hidden;      /* clips the rail while its track collapses */
+}
+/* the contents keep their own width and slide out, so nothing reflows mid-flight
+ * and the exit reads as one movement instead of a squeeze */
+.side-inner {
+  width: var(--rail); height: 100%; min-width: 0;
+  display: flex; flex-direction: column; padding: 12px 10px 10px;
+  transition: transform 300ms var(--ease-pop), opacity 200ms var(--ease);
+}
+body.no-side .side-inner { transform: translateX(-18px); opacity: 0; }
+body.no-side #side { pointer-events: none; }
+.side-head { display: flex; align-items: center; gap: 2px; margin-bottom: 14px; }
+/* the wordmark is a label, not a control: no logo tile, nothing to click */
+.brand { flex: 1 1 auto; display: flex; align-items: center; min-width: 0; padding: 6px 8px; }
+.brand-name {
+  font-weight: 700; letter-spacing: .2px; font-size: 15px;
+  overflow: hidden; text-overflow: ellipsis; white-space: nowrap;
+}
+
+.new-task {
+  display: flex; align-items: center; gap: 9px; height: 40px; padding: 0 12px;
+  margin-bottom: 18px; border-radius: 12px; font-weight: 600;
+  background: var(--bg-lift); border: 1px solid var(--line);
+  transition: border-color var(--t-fast), background var(--t-fast), transform 90ms var(--ease);
+}
+.new-task:hover { border-color: var(--line-strong); background: var(--bg-hover); }
+.new-task:active { transform: scale(.985); }
+.new-task .plus { color: var(--accent); font-size: 16px; line-height: 1; }
+
+.side-sec {
+  display: flex; align-items: center; gap: 6px; padding: 0 8px 6px;
+  color: var(--fg-faint); font-size: 11px; font-weight: 600;
+  letter-spacing: .7px; text-transform: uppercase;
+}
+.sessions { flex: 1 1 auto; overflow-y: auto; min-height: 40px; scrollbar-gutter: stable; }
+.sess-item {
+  display: flex; align-items: baseline; gap: 8px; width: 100%;
+  padding: 7px 9px; border-radius: 9px; text-align: left;
+  color: var(--text-secondary); font-size: 13.5px;
+  animation: fade-in var(--t) var(--ease) both;
+  transition: background var(--t-fast), color var(--t-fast);
+}
+.sess-item:hover { background: var(--hover); color: var(--text-primary); }
+.sess-item.on { background: var(--sidebar-surface-tertiary); color: var(--text-primary); font-weight: 600; }
+.sess-item.on::before {
+  content: ""; width: 5px; height: 5px; border-radius: 50%;
+  background: var(--accent); flex: 0 0 auto; align-self: center;
+}
+.sess-item .t { flex: 1 1 auto; min-width: 0; overflow: hidden; }
+.sess-item .ts { display: block; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }
+/* a title that does not fit fades at the edge and scrolls itself on hover, the
+ * way the reference sidebar handles long names (its data-marquee-text pattern) */
+.sess-item .t.mq { mask-image: linear-gradient(90deg, #000 calc(100% - 18px), transparent); }
+.sess-item:hover .t.mq { mask-image: none; }
+.sess-item:hover .t.mq .ts {
+  width: max-content; text-overflow: clip;
+  animation: marquee 2.6s var(--ease) 160ms infinite alternate;
+}
+@keyframes marquee { from { transform: none; } to { transform: translateX(calc(-1 * var(--mq, 0px))); } }
+.sess-item .n { font-size: 11px; color: var(--text-quaternary); font-weight: 400; white-space: nowrap; }
+.side-foot {
+  display: flex; align-items: center; gap: 6px; padding-top: 10px;
+  margin-top: 6px; border-top: 1px solid var(--line);
+}
+.who { display: flex; align-items: center; gap: 9px; min-width: 0; flex: 1 1 auto; }
+.avatar {
+  width: 28px; height: 28px; border-radius: 50%; flex: 0 0 auto;
+  background: linear-gradient(135deg, var(--violet), var(--accent));
+  color: #fff; font-weight: 800; font-size: 13px;
+  display: flex; align-items: center; justify-content: center;
+}
+.who-text { min-width: 0; display: flex; flex-direction: column; line-height: 1.3; }
+.who-text b { font-size: 12.5px; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }
+.who-text i { font-size: 11px; color: var(--fg-faint); font-style: normal; }
+
+/* --- top bar -------------------------------------------------------------- */
+#main { display: flex; flex-direction: column; min-width: 0; position: relative; }
+#topbar { display: flex; align-items: center; gap: 4px; padding: 9px 12px; border-bottom: 1px solid var(--line); }
+.seg {
+  display: flex; background: var(--bg-soft); border: 1px solid var(--line);
+  border-radius: 99px; padding: 2px; position: relative;
+}
+.seg button {
+  height: 25px; padding: 0 12px; border-radius: 99px; color: var(--fg-dim); font-size: 12.5px;
+  transition: color var(--t-fast), background var(--t);
+}
+.seg button.on { background: var(--bg-lift); color: var(--fg); font-weight: 600; }
+.only-narrow { display: none; }
+
+#scroll {
+  flex: 1 1 auto; overflow-y: auto; padding: 22px 0 10px;
+  scroll-behavior: smooth; scrollbar-gutter: stable;
+}
+#log { max-width: 760px; margin: 0 auto; padding: 0 20px; }
+#log:empty { display: none; }
+
+/* --- empty state ---------------------------------------------------------- */
+/* the first screen is one line: no logo, no suggestion wall. It goes away the
+ * moment the first message lands (see showEmpty). */
+#empty { margin: 22vh auto 0; padding: 0 20px; text-align: center; }
+#empty.gone { display: none; }
+#empty .ready {
+  margin: 0; color: var(--text-tertiary); font-size: 15px; letter-spacing: .3px;
+  animation: fade-in 320ms var(--ease) both;
+}
+
+/* --- transcript ----------------------------------------------------------- */
+.msg { display: flex; gap: 12px; padding: 13px 0 0; animation: fade-up var(--t) var(--ease) both; }
+.msg .gutter { flex: 0 0 24px; padding-top: 2px; color: var(--fg-faint); font-size: 11px; text-align: center; user-select: none; }
+.msg .body { min-width: 0; flex: 1 1 auto; }
+.msg.user { justify-content: flex-end; }
+.msg.user .gutter { display: none; }
+.msg.user .body {
+  background: var(--bg-lift); border-radius: 18px 18px 6px 18px;
+  padding: 9px 15px; max-width: 76%; white-space: pre-wrap; word-break: break-word;
+}
+.msg.error .body { color: var(--danger); }
+.body > :first-child, .live-html > :first-child { margin-top: 0; }
+.body > :last-child, .live-html > :last-child { margin-bottom: 0; }
+.body h1, .body h2, .body h3, .body h4 { line-height: 1.35; margin: 1.1em 0 .5em; }
+.body h1 { font-size: 1.45em; } .body h2 { font-size: 1.25em; } .body h3 { font-size: 1.1em; }
+.body p, .body ul, .body ol { margin: .55em 0; }
+.body ul, .body ol { padding-left: 1.4em; }
+.body li { margin: .22em 0; }
+.body a { color: var(--link); }
+.body code {
+  font-family: var(--mono); font-size: .88em; background: var(--bg-code);
+  border: 1px solid var(--line); border-radius: 6px; padding: .1em .35em;
+}
+.body pre {
+  background: var(--bg-code); border: 1px solid var(--line);
+  border-radius: 12px; padding: 12px 14px; overflow-x: auto; margin: .75em 0;
+}
+.body pre code { background: none; border: 0; padding: 0; font-size: 13px; }
+.body blockquote { margin: .75em 0; padding: .1em 0 .1em 12px; border-left: 3px solid var(--line-strong); color: var(--fg-dim); }
+.body table { border-collapse: collapse; margin: .75em 0; font-size: 13.5px; display: block; overflow-x: auto; }
+.body th, .body td { border: 1px solid var(--line); padding: 5px 9px; text-align: left; }
+.body th { background: var(--bg-soft); }
+.body img { max-width: 100%; border-radius: 10px; }
+.body hr { border: 0; border-top: 1px solid var(--line); margin: 1.2em 0; }
+/* a caret while tokens are still arriving */
+.body.live > .live-html > :last-child::after {
+  content: ""; display: inline-block; width: 7px; height: 14px;
+  margin-left: 3px; vertical-align: -2px; border-radius: 2px;
+  background: var(--accent); animation: blink 1.1s steps(2, start) infinite;
+}
+
+.attach-strip { display: flex; flex-wrap: wrap; gap: 8px; margin: 2px 0 8px; justify-content: flex-end; }
+.attach-strip img {
+  width: 108px; height: 108px; object-fit: cover; border-radius: 12px;
+  border: 1px solid var(--line); animation: rise var(--t) var(--ease) both;
+}
+
+.turn-meta { display: flex; align-items: center; gap: 10px; padding: 10px 0 2px 36px; }
+.turn-meta details > summary { cursor: pointer; list-style: none; color: var(--fg-faint); font-size: 12px; transition: color var(--t-fast); }
+.turn-meta details > summary::-webkit-details-marker { display: none; }
+.turn-meta details > summary:hover { color: var(--fg-dim); }
+.msg-actions { display: flex; gap: 2px; padding: 4px 0 0 36px; opacity: 0; transition: opacity var(--t) var(--ease); }
+.msg:hover .msg-actions, .msg-actions:focus-within { opacity: 1; }
+.msg-actions button {
+  height: 26px; padding: 0 9px; border-radius: 8px; color: var(--fg-faint); font-size: 12px;
+  transition: background var(--t-fast), color var(--t-fast);
+}
+.msg-actions button:hover { background: var(--bg-hover); color: var(--fg); }
+
+.think { margin: .1em 0 .5em; }
+.think summary { cursor: pointer; list-style: none; color: var(--fg-faint); font-size: 12.5px; transition: color var(--t-fast); }
+.think summary::-webkit-details-marker { display: none; }
+.think summary::before { content: "▸ "; }
+.think[open] summary::before { content: "▾ "; }
+.think summary:hover { color: var(--fg-dim); }
+.think > div {
+  color: var(--fg-dim); font-size: 13.5px; white-space: pre-wrap;
+  padding: 6px 0 2px 14px; border-left: 2px solid var(--line);
+  animation: fade-in var(--t) var(--ease) both;
+}
+
+/* tool calls: one line, opens on demand (like "已执行 python · <<PY>") */
+.turn-steps { padding: 6px 0 0 36px; }
+.turn-steps > summary { cursor: pointer; list-style: none; color: var(--fg-faint); font-size: 12.5px; transition: color var(--t-fast); }
+.turn-steps > summary::-webkit-details-marker { display: none; }
+.turn-steps > summary::before { content: "⌘ "; }
+.turn-steps > summary:hover { color: var(--fg-dim); }
+.steps-box { padding: 2px 0 0 12px; margin-top: 4px; border-left: 2px solid var(--line); }
+.step-line { animation: fade-up var(--t) var(--ease) both; }
+.step-line > summary {
+  display: flex; align-items: baseline; gap: 8px; cursor: pointer; list-style: none;
+  padding: 5px 0; font-size: 13px; color: var(--fg-dim);
+  transition: color var(--t-fast);
+}
+.step-line > summary::-webkit-details-marker { display: none; }
+.step-line > summary:hover { color: var(--fg); }
+.step-line .name { font-family: var(--mono); font-size: 12.5px; color: var(--fg); }
+.step-line .args { flex: 1 1 auto; min-width: 0; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; font-family: var(--mono); font-size: 12px; }
+.step-line .phase { font-size: 11.5px; color: var(--fg-faint); white-space: nowrap; display: inline-flex; align-items: center; gap: 6px; }
+.step-line.run .phase { color: var(--warn); }
+.step-line.run .phase::before {
+  content: ""; width: 6px; height: 6px; border-radius: 50%;
+  background: var(--warn); animation: pulse 1.1s var(--ease) infinite;
+}
+.step-line.bad .phase, .step-line.bad .name { color: var(--danger); }
+.step-line.done .phase::before { content: "✓"; }
+.step-body { padding: 0 0 6px; animation: fade-in var(--t) var(--ease) both; }
+.step-out, .step-args-full {
+  font-family: var(--mono); font-size: 12px; background: var(--bg-code);
+  border: 1px solid var(--line); border-radius: 10px; padding: 9px 11px;
+  margin: 4px 0 0; max-height: 300px; overflow: auto; white-space: pre-wrap; word-break: break-word;
+}
+
+#to-bottom {
+  position: absolute; right: 26px; bottom: 172px; z-index: 5;
+  width: 38px; height: 38px; display: flex; align-items: center; justify-content: center;
+  background: var(--bg-lift); border: 1px solid var(--line-strong); color: var(--fg-dim);
+  box-shadow: var(--shadow-sm); font-size: 16px;
+  animation: fade-in var(--t) var(--ease) both;
+  transition: color var(--t-fast), transform 90ms var(--ease), background var(--t-fast);
+}
+#to-bottom:hover { color: var(--fg); background: var(--bg-hover); }
+#to-bottom:active { transform: scale(.94); }
+
+/* --- composer ------------------------------------------------------------- */
+#composer { padding: 8px 20px 12px; }
+#composer > * { max-width: 760px; margin: 0 auto; }
+.pending { display: flex; flex-wrap: wrap; gap: 8px; padding: 0 0 8px; }
+.pending .chip { background: var(--bg-soft); border-color: var(--line); }
+
+.box {
+  border: 1px solid var(--border-medium); border-radius: var(--radius-lg);
+  background: var(--composer-surface-primary); padding: 11px 13px 9px;
+  transition: border-color var(--t), box-shadow var(--t), background var(--t);
+}
+.box:hover { border-color: var(--border-heavy); }
+.box:focus-within {
+  border-color: var(--accent-line);
+  box-shadow: 0 0 0 3px var(--accent-weak);
+}
+.box textarea {
+  width: 100%; border: 0; background: none; resize: none; outline: none;
+  max-height: 240px; min-height: 26px; padding: 2px 2px 8px;
+  caret-color: var(--accent);
+}
+.box textarea::placeholder { color: var(--fg-faint); }
+.box-row { display: flex; align-items: center; gap: 6px; }
+.att { border: 1px solid var(--line); font-size: 17px; line-height: 1; }
+.cost { color: var(--fg-faint); font-size: 12px; padding: 0 4px; white-space: nowrap; }
+.cost:hover { color: var(--fg-dim); }
+.send {
+  width: 36px; height: 36px; flex: 0 0 auto; font-size: 17px; font-weight: 800;
+  background: var(--accent); color: var(--accent-fg);
+  transition: filter var(--t-fast), transform 90ms var(--ease), opacity var(--t-fast);
+}
+.send:hover { filter: brightness(1.08); }
+.send:active { transform: scale(.94); }
+.send:disabled { opacity: .38; }
+.icon.stop { border: 1px solid var(--line); font-size: 12px; }
+.hint-line { display: flex; align-items: center; gap: 10px; padding: 7px 6px 0; font-size: 12px; color: var(--fg-faint); }
+
+/* permission gate: inline, the turn really is waiting on it */
+.gate {
+  border: 1px solid color-mix(in srgb, var(--warn) 45%, transparent);
+  border-radius: var(--radius); padding: 11px 13px; margin-bottom: 10px;
+  background: color-mix(in srgb, var(--warn) 7%, var(--bg-soft));
+  animation: rise var(--t) var(--ease) both;
+}
+.gate-head { display: flex; align-items: center; gap: 8px; font-weight: 650; font-size: 14px; }
+.gate-timer { color: var(--warn); font-size: 12.5px; font-variant-numeric: tabular-nums; }
+.gate pre {
+  margin: 8px 0 0; background: var(--bg-code); border: 1px solid var(--line);
+  border-radius: 10px; padding: 9px 11px; max-height: 180px; overflow: auto;
+  font-family: var(--mono); font-size: 12.5px; white-space: pre-wrap; word-break: break-word;
+}
+.gate-keys { padding: 8px 0 2px; color: var(--fg-dim); font-size: 12.5px; }
+.gate-row { display: flex; flex-wrap: wrap; gap: 8px; margin-top: 6px; }
+.gate-row button {
+  height: 34px; padding: 0 14px; border-radius: 10px; font-weight: 600;
+  border: 1px solid var(--line-strong); background: var(--bg-lift);
+  transition: filter var(--t-fast), transform 90ms var(--ease), background var(--t-fast);
+}
+.gate-row button.ok { background: var(--accent); color: var(--accent-fg); border-color: transparent; }
+.gate-row button.danger { color: var(--danger); }
+.gate-row button:hover { filter: brightness(1.12); }
+.gate-row button:active { transform: scale(.97); }
+
+/* --- popovers: anchored, animated, and the only place settings live ------- */
+.pop {
+  position: fixed; z-index: 45; min-width: 236px; max-width: 320px;
+  background: var(--main-surface-tertiary); border: 1px solid var(--border-medium);
+  border-radius: 14px; padding: 6px; box-shadow: var(--shadow);
+  animation: pop-in 160ms var(--ease) both;
+}
+.pop-item {
+  display: flex; align-items: flex-start; gap: 10px; width: 100%;
+  padding: 9px 10px; border-radius: 10px; text-align: left;
+  transition: background var(--t-fast);
+}
+.pop-item:hover { background: var(--bg-hover); }
+.pop-item .pi { width: 18px; flex: 0 0 auto; text-align: center; color: var(--fg-dim); font-size: 14px; line-height: 1.4; }
+/* title over description: two lines, never one run-on line */
+.pop-item .pm { min-width: 0; flex: 1 1 auto; display: flex; flex-direction: column; gap: 2px; }
+.pop-item .pt { font-size: 13.5px; font-weight: 600; line-height: 1.35; }
+.pop-item .pd { font-size: 11.5px; color: var(--fg-faint); line-height: 1.4; }
+.pop-item .pk { flex: 0 0 auto; color: var(--accent); font-size: 13px; }
+.pop-item.on { background: var(--accent-weak); }
+.pop-item.on .pi, .pop-item.on .pt { color: var(--accent); }
+.pop-item.danger .pi, .pop-item.danger .pt { color: var(--danger); }
+.pop-sep { height: 1px; background: var(--line); margin: 5px 4px; }
+.pop-title { padding: 8px 10px 4px; color: var(--fg-faint); font-size: 11.5px; font-weight: 600; letter-spacing: .4px; }
+.pop-input { display: flex; gap: 6px; padding: 4px 6px 6px; }
+.pop-input input {
+  flex: 1 1 auto; min-width: 0; background: var(--bg-soft); border: 1px solid var(--line);
+  border-radius: 9px; padding: 7px 9px; outline: none; font-size: 13px;
+  caret-color: var(--accent);
+  transition: border-color var(--t-fast);
+}
+.pop-input input:focus { border-color: var(--accent-line); }
+.pop-input button {
+  height: 32px; padding: 0 12px; border-radius: 9px; font-weight: 600; font-size: 13px;
+  background: var(--accent); color: var(--accent-fg);
+  transition: filter var(--t-fast), transform 90ms var(--ease);
+}
+.pop-input button:active { transform: scale(.96); }
+.kv { display: flex; align-items: center; justify-content: space-between; gap: 10px; padding: 5px 10px; font-size: 13px; }
+.kv span { color: var(--fg-dim); }
+.kv b { font-variant-numeric: tabular-nums; font-weight: 600; min-width: 0; overflow-wrap: anywhere; text-align: right; }
+.pop { max-height: calc(100vh - 16px); overflow-y: auto; }
+.pop:has(.pop-list) { min-width: min(360px, 94vw); max-width: min(440px, 94vw); }
+.pop-head { display: flex; align-items: center; justify-content: space-between; gap: 8px; }
+.pop-list { max-height: min(46vh, 420px); overflow-y: auto; }
+.pop-list .pop-item .pt { overflow-wrap: anywhere; font-family: var(--mono); font-size: 12.5px; }
+.pop-empty { padding: 10px; color: var(--fg-faint); font-size: 12.5px; }
+.pop-note { padding: 2px 10px 6px; color: var(--fg-faint); font-size: 11.5px; line-height: 1.45; }
+.pop-note:empty { display: none; }
+.pop-note.bad { color: var(--danger); }
+.pop-note.warn { color: var(--warn); }
+.pop-item.dim { opacity: .55; }
+button.link { color: var(--link); padding: 0 4px; font-size: inherit; text-decoration: underline; }
+.ok-text { color: var(--text-success); }
+.warn-text { color: var(--warn); }
+
+/* providers and the key dialog */
+#modal-box .prov {
+  display: flex; align-items: center; gap: 10px; padding: 9px 8px;
+  border-bottom: 1px solid var(--line); border-radius: 8px;
+}
+#modal-box .prov.on { background: var(--accent-weak); }
+#modal-box .prov .pm { flex: 1 1 auto; min-width: 0; display: flex; flex-direction: column; gap: 2px; }
+#modal-box .prov .pm span { overflow-wrap: anywhere; }
+#modal-box .prov-acts button {
+  height: 30px; padding: 0 12px; border-radius: 9px; font-weight: 600; font-size: 13px;
+  border: 1px solid var(--line-strong); background: var(--bg-lift); white-space: nowrap;
+}
+#modal-box .prov-acts button.ok { background: var(--accent); color: var(--accent-fg); border-color: transparent; }
+#modal-box .key-input {
+  width: 100%; margin-top: 12px; background: var(--bg-lift); border: 1px solid var(--line);
+  border-radius: 10px; padding: 9px 11px; outline: none; font-family: var(--mono);
+}
+#modal-box .key-input:focus { border-color: var(--accent-line); }
+#modal-box .key-err { min-height: 18px; margin-top: 6px; color: var(--danger); font-size: 12.5px; }
+
+/* one line of feedback for anything that failed or needs a look */
+.toast {
+  position: fixed; left: 50%; bottom: 24px; transform: translateX(-50%); z-index: 60;
+  max-width: min(560px, 92vw); padding: 10px 14px; border-radius: 12px;
+  background: var(--main-surface-tertiary); border: 1px solid var(--border-medium);
+  box-shadow: var(--shadow); font-size: 13px; line-height: 1.45; overflow-wrap: anywhere;
+  animation: fade-up 160ms var(--ease) both;
+}
+.toast.err { border-color: color-mix(in srgb, var(--danger) 55%, transparent); color: var(--danger); }
+.toast.warn { border-color: color-mix(in srgb, var(--warn) 55%, transparent); color: var(--warn); }
+.toast.ok { border-color: var(--accent-line); color: var(--accent); }
+.toast[hidden] { display: none; }
+
+/* --- right panel ---------------------------------------------------------- */
+/* the right panel mirrors the rail: fixed-width contents, clipped by the track,
+ * so folding it in and out is one eased movement on each side */
+#panel { background: var(--main-surface-secondary); border-left: 1px solid var(--border-light); min-width: 0; overflow: hidden; }
+.panel-inner {
+  width: var(--panel); height: 100%; overflow-y: auto; padding: 12px;
+  transition: transform 300ms var(--ease-pop), opacity 200ms var(--ease);
+}
+body.no-panel .panel-inner { transform: translateX(18px); opacity: 0; }
+body.no-panel #panel { pointer-events: none; }
+.panel-head {
+  display: flex; align-items: center; gap: 6px; padding: 2px 4px 10px;
+  color: var(--fg-faint); font-size: 11px; font-weight: 600;
+  letter-spacing: .7px; text-transform: uppercase;
+}
+.card { border-top: 1px solid var(--line); padding: 12px 2px; }
+.card:first-of-type { border-top: 0; }
+.card h4 { margin: 0 0 8px; font-size: 12.5px; font-weight: 650; display: flex; align-items: center; gap: 6px; }
+.card .kv { padding: 3px 0; }
+.todo-list { margin: 0; padding: 0; list-style: none; font-size: 13.5px; }
+.todo-list li { display: flex; gap: 8px; padding: 3px 0; animation: fade-in var(--t) var(--ease) both; transition: color var(--t-fast); }
+.todo-list .boxmark { color: var(--fg-faint); }
+.todo-list li.done { color: var(--fg-faint); text-decoration: line-through; }
+.todo-list li.doing { color: var(--fg); font-weight: 600; }
+#steps-body .step-line > summary { font-size: 12.5px; }
+#steps-body .step-line .phase { font-size: 11px; }
+
+/* --- modal sheet (session list) ------------------------------------------- */
+#modal {
+  position: fixed; inset: 0; background: rgba(0, 0, 0, .5); z-index: 40;
+  display: flex; align-items: center; justify-content: center; padding: 16px;
+  animation: fade-in var(--t) var(--ease) both;
+}
+#modal-box {
+  width: min(560px, 100%); max-height: 82vh; overflow-y: auto;
+  background: var(--main-surface-tertiary); border: 1px solid var(--border-medium);
+  border-radius: 18px; padding: 16px 18px; box-shadow: var(--shadow);
+  animation: pop-in 180ms var(--ease) both;
+}
+#modal-box h3 { margin: 0 0 12px; font-size: 17px; }
+#modal-box .field { display: flex; align-items: center; gap: 8px; margin: 9px 0; flex-wrap: wrap; }
+#modal-box .field > span { width: 70px; color: var(--fg-dim); font-size: 13px; }
+#modal-box .field input {
+  flex: 1 1 auto; min-width: 150px; background: var(--bg-lift);
+  border: 1px solid var(--line); border-radius: 10px; padding: 8px 10px; outline: none;
+}
+#modal-box .buttons { display: flex; flex-wrap: wrap; gap: 8px; align-items: center; margin-top: 12px; }
+#modal-box .buttons button {
+  height: 34px; padding: 0 14px; border-radius: 10px; font-weight: 600;
+  border: 1px solid var(--line-strong); background: var(--bg-lift);
+  transition: filter var(--t-fast), transform 90ms var(--ease);
+}
+#modal-box .buttons button:active { transform: scale(.97); }
+#modal-box .buttons button.ok { background: var(--accent); color: var(--accent-fg); border-color: transparent; }
+#modal-box .buttons button.danger { color: var(--danger); }
+#modal-box .sess {
+  display: flex; align-items: center; gap: 8px; padding: 8px 6px;
+  border-bottom: 1px solid var(--line); border-radius: 8px;
+  transition: background var(--t-fast);
+}
+#modal-box .sess:hover { background: var(--bg-hover); }
+#modal-box .sess .id { font-family: var(--mono); font-size: 11.5px; color: var(--fg-faint); }
+#modal-box .sess .t { flex: 1 1 auto; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; font-size: 13.5px; }
+
+/* --- narrow screens: rail and panel become drawers ------------------------ */
+#backdrop { position: fixed; inset: 0; background: rgba(0, 0, 0, .45); z-index: 30; display: none; animation: fade-in var(--t) var(--ease) both; }
+body.drawer-open #backdrop { display: block; }
+@media (max-width: 900px) {
+  #app { grid-template-columns: minmax(0, 1fr) !important; transition: none; }
+  #side, #panel {
+    position: fixed; top: 0; bottom: 0; width: min(310px, 86vw); z-index: 31;
+    display: block; transition: transform 260ms var(--ease-pop);
+  }
+  .side-inner { width: auto; transform: none !important; opacity: 1 !important; }
+  #side { overflow: visible; pointer-events: auto; }
+  #side { left: 0; transform: translateX(-102%); }
+  #panel { right: 0; transform: translateX(102%); }
+  body.side-open #side, body.panel-open #panel { transform: none; }
+  .only-narrow { display: inline-flex; }
+  #to-bottom { bottom: 150px; right: 16px; }
+  .hero h1 { font-size: 21px; }
+  .msg.user .body { max-width: 88%; }
+  .turn-meta, .msg-actions, .turn-steps { padding-left: 0; }
+}
+@media (min-width: 901px) { #backdrop { display: none !important; } }
+
+/* --- motion --------------------------------------------------------------- */
+@keyframes fade-in { from { opacity: 0; } to { opacity: 1; } }
+@keyframes fade-up { from { opacity: 0; transform: translateY(6px); } to { opacity: 1; transform: none; } }
+@keyframes rise { from { opacity: 0; transform: translateY(10px) scale(.99); } to { opacity: 1; transform: none; } }
+@keyframes pop-in { from { opacity: 0; transform: translateY(-4px) scale(.97); } to { opacity: 1; transform: none; } }
+@keyframes pulse { 0%, 100% { opacity: .3; transform: scale(.85); } 50% { opacity: 1; transform: scale(1); } }
+@keyframes blink { 0%, 100% { opacity: 1; } 50% { opacity: 0; } }
+@media (prefers-reduced-motion: reduce) {
+  *, *::before, *::after { animation-duration: 1ms !important; transition-duration: 1ms !important; }
+  #scroll { scroll-behavior: auto; }
+}