#ifndef _DEFAULT_SOURCE
#define _DEFAULT_SOURCE
#endif

#include "cJSON.h"
#include "cst_attach.h"
#include "cst_agent.h"
#include "cst_mcp.h"
#include "cst_metrics.h"
#include "cst_perms.h"
#include "cst_sandbox.h"
#include "cst_skills.h"
#include "cst_subagent.h"
#include "cst_tools.h"
#include "cst_undo.h"
#include "cst_util.h"
#include "mm_tool.h"
#include "os_compat.h"

#include <stdint.h>
#include <stdio.h>
#include <time.h>
#ifndef _WIN32
#include <sys/utsname.h>
#endif
#include <stdlib.h>
#include <string.h>

/* --- protocols ----------------------------------------------------------- */

static const char *const PROTO_NAMES[] = {"native-eos", "native-inc", "ctl",
                                          "ctl-heredoc", "xml"};

const char *cst_protocol_name(cst_protocol_t p) {
  if ((int)p < 0 || (size_t)p >= sizeof(PROTO_NAMES) / sizeof(PROTO_NAMES[0]))
    return "native-eos";
  return PROTO_NAMES[p];
}

int cst_protocol_parse(const char *s, cst_protocol_t *out) {
  if (!s || !out)
    return -1;
  if (strcmp(s, "native") == 0) {
    *out = CST_PROTO_NATIVE_EOS;
    return 0;
  }
  for (size_t i = 0; i < sizeof(PROTO_NAMES) / sizeof(PROTO_NAMES[0]); i++) {
    if (strcmp(s, PROTO_NAMES[i]) == 0) {
      *out = (cst_protocol_t)i;
      return 0;
    }
  }
  return -1;
}

int cst_protocol_is_native(cst_protocol_t p) {
  return p == CST_PROTO_NATIVE_EOS || p == CST_PROTO_NATIVE_INC;
}

int cst_protocol_is_ctl(cst_protocol_t p) {
  return p == CST_PROTO_CTL || p == CST_PROTO_CTL_HEREDOC;
}

int cst_protocol_is_text(cst_protocol_t p) { return !cst_protocol_is_native(p); }

int cst_protocol_is_eager(cst_protocol_t p) {
  return p == CST_PROTO_NATIVE_INC || cst_protocol_is_ctl(p);
}

/* --- system prompts ------------------------------------------------------ */

#define PROMPT_BASE                                                            \
  "You are CST-Agent, a coding assistant working inside a local workspace. "   \
  "All tool paths are relative to the workspace root and shell commands run "  \
  "there.\n"                                                                   \
  "Reply in the user's language. Be concise.\n\n"

#define PROMPT_TOOLS                                                           \
  "Tools:\n"                                                                   \
  "- read_file(path): read a text file.\n"                                     \
  "- run_shell(cmd): run a non-interactive shell command (stdin is empty; "    \
  "long-running commands are killed).\n"                                       \
  "- write_file(path, content): create or overwrite a whole file.\n"           \
  "- apply_patch(patch): unified diff with ---/+++ headers and "               \
  "@@ -a,b +c,d @@ hunks (keep 3 context lines). Prefer it for small edits "   \
  "to existing files.\n\n"                                                     \
  "Workflow:\n"                                                                \
  "1) Inspect only what you need (read_file; ls/grep via run_shell).\n"        \
  "2) Make the change.\n"                                                      \
  "3) Verify with the project's build/tests via run_shell; fix and re-run "    \
  "until they pass.\n"                                                         \
  "4) Finish with a short plain-text summary of what changed.\n"               \
  "Never invent other tool names.\n"

#define PROMPT_RESULTS                                                         \
  "- After your last frame, STOP writing and wait. Results arrive in the "     \
  "next user message as:\n"                                                    \
  "    result <id> exit=<code> <nbytes>\n"                                     \
  "    <output>\n"                                                             \
  "    end\n"                                                                  \
  "  Never write result frames yourself.\n"

static const char SEC_NATIVE[] =
    "Call tools through the native function-calling interface.\n\n";

static const char SEC_CTL[] =
    "\nHow to call tools: write CTL frames as plain text in your reply. There "
    "is NO function-calling API here; do not use JSON or XML.\n\n"
    "call <id> <tool>\n"
    "arg <key> <nbytes>\n"
    "<exactly nbytes bytes of the raw value>\n"
    "end\n\n"
    "Rules:\n"
    "- <id> is a short unique token per call (c1, c2, ...). Every frame line "
    "starts at the beginning of a line.\n"
    "- nbytes counts UTF-8 BYTES of the raw value (ASCII = 1 byte, most CJK "
    "characters = 3 bytes), including every newline inside it. The value is "
    "copied verbatim, with no escaping. One newline after the value, before "
    "`end`, is optional and not counted.\n"
    "- Short single-line values may be inline instead: `arg path src/main.c` "
    "(the value is the rest of the line; it must not be only digits).\n"
    "- You may write several frames in one reply.\n" PROMPT_RESULTS
    "\nExample (the content value is \"hello world\\n\" = 12 bytes):\n"
    "call c1 read_file\n"
    "arg path README.md\n"
    "end\n"
    "call c2 write_file\n"
    "arg path hello.txt\n"
    "arg content 12\n"
    "hello world\n"
    "end\n";

static const char SEC_CTL_HEREDOC[] =
    "\nHow to call tools: write CTL frames as plain text in your reply. There "
    "is NO function-calling API here; do not use JSON or XML.\n\n"
    "call <id> <tool>\n"
    "arg <key> <<TAG\n"
    "<value lines, verbatim>\n"
    "TAG\n"
    "end\n\n"
    "Rules:\n"
    "- <id> is a short unique token per call (c1, c2, ...). Every frame line "
    "starts at the beginning of a line.\n"
    "- TAG is a word of letters/digits/underscore that never appears alone on "
    "a line inside the value (e.g. EOF_1). The value is the lines between the "
    "header and the TAG line, each ending with a newline. No escaping.\n"
    "- Short single-line values may be inline instead: `arg path src/main.c` "
    "(the value is the rest of the line; it must not be only digits).\n"
    "- You may write several frames in one reply.\n" PROMPT_RESULTS
    "\nExample:\n"
    "call c1 read_file\n"
    "arg path README.md\n"
    "end\n"
    "call c2 write_file\n"
    "arg path hello.txt\n"
    "arg content <<EOF_1\n"
    "hello world\n"
    "EOF_1\n"
    "end\n";

static const char SEC_XML[] =
    "\nHow to call tools: write tool blocks as plain text in your reply. There "
    "is NO function-calling API here.\n\n"
    "<minimax:tool_call>\n"
    "<invoke name=\"read_file\">\n"
    "<parameter name=\"path\">README.md</parameter>\n"
    "</invoke>\n"
    "</minimax:tool_call>\n\n"
    "- Parameter values are verbatim text (no escaping).\n"
    "- You may write several blocks in one reply.\n" PROMPT_RESULTS;

static const char SEC_CTL_NUMBERS[] =
    "- A purely numeric inline value must be quoted (`arg limit \"50\"`); "
    "bare digits are read as a byte count.\n";

static int tool_on(const cst_agent_config_t *cfg, const char *name) {
  return cst_tool_active(cfg, cst_tool_find(name));
}

/* /mode general (cst_mode_name covers "" and unknown values). */
static int general_mode(const cst_agent_config_t *cfg) {
  return cfg && strcmp(cst_mode_name(cfg->mode), "general") == 0;
}

/* Workflow for /mode general: gather, then do the work, then check it. */
static void put_workflow_general(cst_buf_t *b, const cst_agent_config_t *cfg) {
  cst_buf_puts(b, "\nWorkflow:\n1) Gather what you need");
  const char *explore[] = {"list_dir", "glob", "grep"};
  int n = 0;
  for (int i = 0; i < 3; i++) {
    if (tool_on(cfg, explore[i]))
      cst_buf_printf(b, "%s%s", n++ ? ", " : " (", explore[i]);
  }
  if (n)
    cst_buf_puts(b, ")");
  if (tool_on(cfg, "read_file"))
    cst_buf_puts(b, "; read big files in slices with read_file offset/limit, "
                    "and use the documents the user attached");
  cst_buf_puts(b, ".\n2) Do the work");
  n = 0;
  if (tool_on(cfg, "write_file")) {
    cst_buf_puts(b, ": write_file for new files");
    n++;
  }
  if (tool_on(cfg, "edit_file"))
    cst_buf_printf(b, "%sedit_file for small exact replacements",
                   n++ ? ", " : ": ");
  if (tool_on(cfg, "apply_patch"))
    cst_buf_printf(b, "%sapply_patch for multi-hunk diffs to existing files",
                   n++ ? ", " : ": ");
  if (tool_on(cfg, "run_shell"))
    cst_buf_printf(b, "%srun_shell for commands, computations, data "
                      "processing and other CLIs",
                   n++ ? ", " : ": ");
  cst_buf_puts(b, ".\n");
  if (tool_on(cfg, "run_shell"))
    cst_buf_puts(b, "3) Check the result before saying it works (re-read it, "
                    "run it, or verify the numbers); fix and re-run until it "
                    "holds.\n");
  if (tool_on(cfg, "todo_write"))
    cst_buf_puts(b, "- For work with several steps keep a todo list with "
                    "todo_write and update it as you go.\n");
  if (tool_on(cfg, "ask_user"))
    cst_buf_puts(b, "- Use ask_user only when a decision is truly the user's; "
                    "otherwise choose sensibly and say so.\n");
  cst_buf_puts(b, "Finish with a short plain-text summary of the result.\n"
                  "Never invent other tool names.\n");
}

/* Workflow section for the default tool profile; names only active tools. */
static void put_workflow(cst_buf_t *b, const cst_agent_config_t *cfg) {
  if (general_mode(cfg)) {
    put_workflow_general(b, cfg);
    return;
  }
  cst_buf_puts(b, "\nWorkflow:\n1) Inspect only what you need");
  const char *explore[] = {"list_dir", "glob", "grep"};
  int n = 0;
  for (int i = 0; i < 3; i++) {
    if (tool_on(cfg, explore[i]))
      cst_buf_printf(b, "%s%s", n++ ? ", " : " (", explore[i]);
  }
  if (n)
    cst_buf_puts(b, ")");
  if (tool_on(cfg, "read_file"))
    cst_buf_puts(b, "; read big files in slices with read_file offset/limit");
  cst_buf_puts(b, ".\n2) Make the change");
  n = 0;
  if (tool_on(cfg, "edit_file")) {
    cst_buf_puts(b, ": edit_file for small exact replacements");
    n++;
  }
  if (tool_on(cfg, "apply_patch"))
    cst_buf_printf(b, "%sapply_patch for multi-hunk or multi-file diffs",
                   n++ ? ", " : ": ");
  if (tool_on(cfg, "write_file"))
    cst_buf_printf(b, "%swrite_file for new files or full rewrites",
                   n++ ? ", " : ": ");
  cst_buf_puts(b, ".\n");
  if (tool_on(cfg, "run_shell"))
    cst_buf_puts(b, "3) Verify with the project's build/tests via run_shell; "
                    "fix and re-run until they pass.\n");
  if (tool_on(cfg, "todo_write"))
    cst_buf_puts(b, "- For work with several steps keep a todo list with "
                    "todo_write and update it as you go.\n");
  if (tool_on(cfg, "ask_user"))
    cst_buf_puts(b, "- Use ask_user only when a decision is truly the user's; "
                    "otherwise choose sensibly and say so.\n");
  cst_buf_puts(b, "Finish with a short plain-text summary of what changed.\n"
                  "Never invent other tool names.\n");
}

#define PROJECT_NOTES_MAX (16 * 1024)

/* Default-profile prompt: static sections first (prefix-cache friendly),
 * then the protocol, then per-session facts. */
static const char DEF_INTRO[] =
    "You are CST-Agent, a coding agent working in the user's local "
    "workspace. All tool paths are relative to the workspace root and shell "
    "commands run there.\n\n"
    "Principles:\n"
    "- Reply in the user's language. Be concise: lead with the result, no "
    "filler.\n"
    "- Understand before changing: look at the relevant code instead of "
    "guessing; never invent files, APIs or tool results.\n"
    "- Make the smallest change that fully solves the task, in the style of "
    "the surrounding code; leave unrelated code alone.\n"
    "- Verify (build, tests, or running it) before saying it works; if you "
    "could not verify, say so.\n"
    "- Keep going until the task is done or truly blocked.\n\n";

/* Same role, framed for /mode general: the workspace may hold documents,
 * data or anything else, not just code. */
static const char DEF_INTRO_GENERAL[] =
    "You are CST-Agent, a general-purpose assistant working in the user's "
    "local workspace. All tool paths are relative to the workspace root and "
    "shell commands run there; the material you work on may be code, "
    "documents, data or anything else.\n\n"
    "Principles:\n"
    "- Reply in the user's language. Be concise: lead with the result, no "
    "filler.\n"
    "- Understand before answering: read the files, documents and sources the "
    "task refers to instead of guessing; never invent content, numbers or "
    "tool results.\n"
    "- Do what was asked, at the scope asked: leave unrelated files, data and "
    "material alone.\n"
    "- Check your work when you can (re-read it, run it, cross-check the "
    "source); if you could not verify something, say so.\n"
    "- Keep going until the task is done or truly blocked.\n\n";

static void put_tool_use(cst_buf_t *b, const cst_agent_config_t *cfg) {
  cst_buf_puts(b, "\nTool use:\n");
  if (tool_on(cfg, "run_shell") &&
      (tool_on(cfg, "grep") || tool_on(cfg, "glob")))
    cst_buf_puts(b, "- Use the file tools rather than run_shell to look "
                    "around: read_file (not cat/head/tail), grep (not "
                    "grep/rg), glob and list_dir (not find/ls).\n");
  cst_buf_puts(b, "- Independent read-only calls can go out together in one "
                  "reply; they run in parallel.\n");
  if (tool_on(cfg, "read_file"))
    cst_buf_puts(b, "- Read a file before editing it and again after a failed "
                    "edit. The `N|` prefixes in read_file output are not part "
                    "of the file.\n");
  if (general_mode(cfg) && tool_on(cfg, "web_fetch"))
    cst_buf_puts(b, "- Use web_fetch for pages the user points you to; say so "
                    "when a page could not be fetched instead of guessing its "
                    "content.\n");
  cst_buf_puts(b, "- When a call fails, read the error and change approach "
                  "instead of repeating the same call.\n");
}

static void put_safety(cst_buf_t *b, const cst_agent_config_t *cfg) {
  if (general_mode(cfg))
    cst_buf_puts(b, "\nSafety:\n"
                    "- Do not delete or overwrite files or data you did not "
                    "create unless the task requires it, and do not send the "
                    "user's content anywhere the task did not ask for. Never "
                    "print secrets or API keys.\n");
  else
    cst_buf_puts(b, "\nSafety:\n"
                    "- Do not commit, push or rewrite git history unless asked. "
                    "Do not delete or overwrite files you did not create unless "
                    "the task requires it. Never print secrets or API keys.\n");
  if (tool_on(cfg, "run_shell"))
    cst_buf_puts(b, "- run_shell may run in a sandbox (writes only inside the "
                    "workspace and temp dir; the network works). If a command is "
                    "blocked the user is asked whether to rerun it outside; "
                    "do not try to work around the sandbox.\n");
}

/* Current branch from .git/HEAD of the workspace or a parent directory. */
static void git_branch(const char *ws, char *out, size_t n) {
  out[0] = 0;
  char dir[4096];
  snprintf(dir, sizeof(dir), "%s", ws ? ws : ".");
  for (int depth = 0; depth < 8; depth++) {
    char p[4200], line[512];
    snprintf(p, sizeof(p), "%s/.git", dir);
    FILE *fp = fopen(p, "rb");
    if (fp) {
      /* worktree / submodule: ".git" is a file "gitdir: <path>" */
      char gd[4096] = "";
      if (fgets(line, sizeof(line), fp) && strncmp(line, "gitdir: ", 8) == 0) {
        line[strcspn(line, "\r\n")] = 0;
        if (line[8] == '/')
          snprintf(gd, sizeof(gd), "%s", line + 8);
        else
          snprintf(gd, sizeof(gd), "%.2000s/%.2000s", dir, line + 8);
      }
      fclose(fp);
      if (gd[0])
        snprintf(p, sizeof(p), "%s/HEAD", gd);
      else
        snprintf(p, sizeof(p), "%s/.git/HEAD", dir);
      fp = fopen(p, "rb");
      if (fp) {
        if (fgets(line, sizeof(line), fp)) {
          line[strcspn(line, "\r\n")] = 0;
          if (strncmp(line, "ref: refs/heads/", 16) == 0)
            snprintf(out, n, "%.100s", line + 16);
          else if (strlen(line) >= 12)
            snprintf(out, n, "detached at %.12s", line);
        }
        fclose(fp);
      }
      return;
    }
    char *sl = strrchr(dir, '/');
    if (!sl || sl == dir)
      return;
    *sl = 0;
  }
}

static void os_label(char *out, size_t n) {
#ifdef _WIN32
  char ver[64];
  os_windows_version(ver, sizeof(ver));
  snprintf(out, n,
           "%s; run_shell is busybox sh (POSIX syntax and applets; write paths "
           "as C:/dir or relative, Windows programs like git or python run "
           "from PATH)",
           ver);
#else
  struct utsname u;
  if (uname(&u) == 0)
    snprintf(out, n, "%s %s %s", u.sysname, u.release, u.machine);
  else
    snprintf(out, n, "Unix");
#endif
}

static void today(char *out, size_t n) {
  time_t now = time(NULL);
  struct tm tmv;
  os_localtime(&now, &tmv);
  strftime(out, n, "%Y-%m-%d", &tmv);
}

static const char *approval_desc(const cst_agent_config_t *cfg) {
  return cfg->auto_approve ? "yolo (no confirmations)"
         : cfg->approval == CST_APPROVAL_ASK
             ? "ask (the user confirms writes and commands)"
         : cfg->approval == CST_APPROVAL_AUTO
             ? "auto (edits run; risky commands need confirmation)"
             : "yolo (no confirmations)";
}

static void put_env(cst_buf_t *b, const cst_agent_config_t *cfg) {
  char osl[256], br[128], day[32];
  os_label(osl, sizeof(osl));
  today(day, sizeof(day));
  git_branch(cfg->workspace_root, br, sizeof(br));
  cst_buf_printf(b, "\nEnvironment:\n- OS: %s; date: %s\n- Workspace: %s%s%s%s\n",
                 osl, day, cfg->workspace_root[0] ? cfg->workspace_root : ".",
                 br[0] ? " (git: " : " (not a git repo", br,
                 ")");
  const char *ap = approval_desc(cfg);
  if (tool_on(cfg, "run_shell")) {
    int sb = cst_sandbox_probe();
    cst_buf_printf(b, "- Approval: %s; shell sandbox: %s\n", ap,
                   sb == CST_SANDBOX_NONE ? "none" : cst_sandbox_name(sb));
  } else {
    cst_buf_printf(b, "- Approval: %s\n", ap);
  }
}

char *cst_agent_build_prompt(const cst_agent_config_t *cfg, cst_protocol_t p,
                             const char *project_notes) {
  int minimal = !cfg || cfg->tool_profile == CST_TOOLS_MINIMAL;
  cst_buf_t b = {0};
  cst_buf_puts(&b, minimal ? PROMPT_BASE
                           : general_mode(cfg) ? DEF_INTRO_GENERAL
                                               : DEF_INTRO);
  if (cst_protocol_is_native(p))
    cst_buf_puts(&b, minimal ? SEC_NATIVE
                             : "Call tools through the native "
                               "function-calling interface.\n");
  if (minimal) {
    cst_buf_puts(&b, PROMPT_TOOLS);
  } else {
    /* native requests carry the descriptions in tools[] already */
    if (!cst_protocol_is_native(p)) {
      char *tools = cst_tools_prompt(cfg);
      cst_buf_puts(&b, tools ? tools : "");
      free(tools);
    }
    put_workflow(&b, cfg);
    put_tool_use(&b, cfg);
    put_safety(&b, cfg);
  }
  switch (p) {
  case CST_PROTO_CTL:
    cst_buf_puts(&b, SEC_CTL);
    break;
  case CST_PROTO_CTL_HEREDOC:
    cst_buf_puts(&b, SEC_CTL_HEREDOC);
    break;
  case CST_PROTO_XML:
    cst_buf_puts(&b, SEC_XML);
    break;
  default:
    break;
  }
  if (!minimal && cst_protocol_is_ctl(p))
    cst_buf_puts(&b, SEC_CTL_NUMBERS);
  /* per-workspace, rarely changing */
  if (!minimal && tool_on(cfg, "skill")) {
    char *ix = cst_skills_index(cfg->workspace_root);
    if (ix)
      cst_buf_puts(&b, ix);
    free(ix);
  }
  if (!minimal && tool_on(cfg, "task")) {
    char *ix = cst_subagent_index(cfg->workspace_root);
    if (ix)
      cst_buf_puts(&b, ix);
    free(ix);
  }
  if (project_notes && project_notes[0]) {
    size_t n = strlen(project_notes);
    if (n > PROJECT_NOTES_MAX)
      n = cst_utf8_floor(project_notes, PROJECT_NOTES_MAX);
    cst_buf_puts(&b, "\nProject instructions (from the workspace's "
                     "AGENTS.md; follow them):\n");
    cst_buf_append(&b, project_notes, n);
    cst_buf_puts(&b, "\n");
  }
  /* per-session facts last */
  if (!minimal)
    put_env(&b, cfg);
  if (!minimal && cfg->depth > 0) {
    cst_buf_puts(&b, "\nSUB-AGENT: another agent delegated one task to you "
                     "with the task tool. It sees none of your tool calls or "
                     "outputs, only your final message, and the user is not "
                     "watching this conversation (you cannot ask questions). "
                     "Work autonomously, then end with one self-contained "
                     "report: what you found or changed, with file paths "
                     "and line numbers, and anything left open.\n");
    if (cfg->role_prompt && cfg->role_prompt[0]) {
      cst_buf_puts(&b, "\nRole instructions:\n");
      cst_buf_puts(&b, cfg->role_prompt);
      cst_buf_puts(&b, "\n");
    }
  }
  if (cfg && cfg->plan_mode)
    cst_buf_puts(&b, "\nPLAN MODE: you may only read and research (no file "
                     "changes, no shell). Investigate, then answer with a "
                     "concrete step-by-step plan: files to touch, what to "
                     "change, how to verify. The user switches to Build mode "
                     "to carry it out.\n");
  return cst_buf_steal(&b);
}

/* Legacy prompts: minimal tool set, no project notes. */
const char *cst_agent_system_prompt_for(cst_protocol_t p) {
  static char *cache[5];
  int i = (int)p < 0 || (int)p > 4 ? 0 : (int)p;
  if (!cache[i])
    cache[i] = cst_agent_build_prompt(NULL, (cst_protocol_t)i, NULL);
  return cache[i] ? cache[i] : PROMPT_BASE;
}

const char *cst_agent_system_prompt(void) {
  return cst_agent_system_prompt_for(CST_PROTO_NATIVE_EOS);
}

/* Request-only nudges (never persisted). */
static const char NUDGE_EMPTY[] =
    "[system] Your previous reply was empty. If the task is not finished, "
    "continue with tool calls; otherwise give the final answer now in plain "
    "text, in the user's language.";
static const char NUDGE_FINAL[] =
    "[system] The tool budget for this turn is exhausted. Do NOT call any "
    "tools. Answer now in plain text, in the user's language: summarize what "
    "was done and what remains.";

/* Session keeps tool output up to this size; the wire cap is separate. */
#define TOOL_OUT_CAP (48 * 1024)
#define READ_FP_MAX 256
#define FAIL_FP_MAX 32
#define PAR_MAX 4
#define MAX_CTL_RETRIES 3

/* --- agent --------------------------------------------------------------- */

typedef struct {
  uint64_t hash;
  unsigned epoch;
} read_fp_t;

struct cst_agent {
  cst_agent_config_t cfg;
  cst_session_t *session;
  cst_metrics_t *metrics;
  volatile int cancel;
  unsigned write_epoch;
  read_fp_t reads[READ_FP_MAX];
  int nreads;
  /* this user turn: failing calls (hash -> count) and applied writes */
  struct {
    uint64_t hash;
    int count;
  } fails[FAIL_FP_MAX];
  int nfails;
  read_fp_t writes[FAIL_FP_MAX];
  int nwrites;
  int turn_no;
  char *notes; /* AGENTS.md text, or NULL */
  char *todo;  /* latest todo_write list, or NULL */
  char undo_tag[64];
  long last_prompt_tokens;
  cst_perms_t *perms;
  /* Images the user picked for the next turn (owned; consumed on send). */
  cst_attachment_t *pending;
  int npending, cappending;
  /* Prefix cache: the request is the frozen system prompt + tools snapshot
   * followed by the session, which only grows until a rebase. */
  cst_model_caps_t caps;
  char caps_model[128];
  char *tools_json;  /* frozen tools[] (owned) */
  uint64_t tools_sig; /* tool settings it was built from */
  int win_start;      /* first session message still sent (chunked window) */
  uint64_t *prev_hash; /* per-message hashes of the previous request */
  int nprev, capprev;
  uint64_t prev_tools;
  char prev_model[128];
  int rebased;     /* the session was compacted since the last request */
  int prefix_off;  /* DeepSeek refused prefix completion */
  int hints_off;   /* the provider refused cache hints (HTTP 400) */
  int images_off;  /* the provider refused a request with images (HTTP 400) */
  char images_off_model[128]; /* ...for this model */
  char cache_key[128];
  cst_usage_t usage, turn_usage;
};

typedef enum {
  PC_OK = 0,
  PC_DENIED,
  PC_DUP,
  PC_UNKNOWN,
  PC_INVALID
} pc_status_t;

typedef struct {
  ctl_tool_call_t call; /* owns arg values */
  char *raw_args;       /* native: arguments string as streamed */
  int invalid;
  int done;
  int exit_code;
  pc_status_t status;
  char *output;
  size_t output_len;
  cst_attachment_t *attach; /* images the tool produced (owned) */
  int nattach;
  const char *origin; /* native / ctl / xml */
  double t_open, t_ready, t_exec_start, t_exec_end;
} pcall_t;

typedef struct {
  cst_agent_t *a;
  const cst_agent_hooks_t *hooks;
  cst_protocol_t proto;
  int step;
  int final_pass;
  double t0;
  cst_buf_t text; /* confirmed answer text (no tool syntax) */
  cst_buf_t raw;  /* visible stream with tool syntax, think stripped */
  cst_buf_t think; /* reasoning, kept for providers that want it back */
  size_t raw_mark;
  int in_think;
  int native_opened; /* a native tool call started streaming */
  int cont;          /* prefix continuations done in this step */
  int prefix_break;  /* first message that differs from the last request */
  const char *prefix_reason;
  char tagbuf[16];
  size_t taglen;
  ctl_parser_t *ctl;
  mm_scan_t *mm;
  pcall_t *calls;
  int ncalls;
  int cap;
  int n_native;
  int saw_ready;
  volatile int stop;
  int parse_error;
  char perr[256];
  double native_open_t[64];
  double ctl_open_t;
} turn_t;

static double rel_ms(const turn_t *t) { return os_now_ms() - t->t0; }

static void hook_status(const turn_t *t, const char *msg) {
  if (t->hooks && t->hooks->on_status)
    t->hooks->on_status(msg, t->hooks->userdata);
}

static double ms2(double v) {
  return v < 0 ? v : (double)(long long)(v * 100.0 + 0.5) / 100.0;
}

static void hook_tool(const turn_t *t, const char *phase,
                      const ctl_tool_call_t *c, const char *detail) {
  if (t->hooks && t->hooks->on_tool_event)
    t->hooks->on_tool_event(phase, c, detail, t->hooks->userdata);
}

static int text_is_meaningful(const char *s, size_t n) {
  size_t visible = 0;
  for (size_t i = 0; i < n; i++) {
    if ((unsigned char)s[i] > 0x20)
      visible++;
  }
  return visible >= 2;
}

static const char *trim_left(const char *s) {
  while (*s == ' ' || *s == '\n' || *s == '\r' || *s == '\t')
    s++;
  return s;
}

static void copy_call(ctl_tool_call_t *dst, const ctl_tool_call_t *src) {
  memset(dst, 0, sizeof(*dst));
  memcpy(dst->id, src->id, sizeof(dst->id));
  memcpy(dst->name, src->name, sizeof(dst->name));
  for (int i = 0; i < src->nargs; i++) {
    ctl_arg_t *d = &dst->args[dst->nargs];
    memcpy(d->key, src->args[i].key, sizeof(d->key));
    d->value_len = src->args[i].value_len;
    d->value = malloc(d->value_len + 1);
    if (!d->value)
      continue;
    if (d->value_len)
      memcpy(d->value, src->args[i].value, d->value_len);
    d->value[d->value_len] = 0;
    dst->nargs++;
  }
}

static pcall_t *push_call(turn_t *t) {
  if (t->ncalls == t->cap) {
    int nc = t->cap ? t->cap * 2 : 4;
    pcall_t *nv = realloc(t->calls, (size_t)nc * sizeof(pcall_t));
    if (!nv)
      return NULL;
    t->calls = nv;
    t->cap = nc;
  }
  pcall_t *pc = &t->calls[t->ncalls++];
  memset(pc, 0, sizeof(*pc));
  pc->t_exec_start = pc->t_exec_end = -1;
  return pc;
}

static void turn_free(turn_t *t) {
  for (int i = 0; i < t->ncalls; i++) {
    cst_tool_call_clear(&t->calls[i].call);
    free(t->calls[i].raw_args);
    free(t->calls[i].output);
    cst_attachments_free(t->calls[i].attach, t->calls[i].nattach);
  }
  free(t->calls);
  t->calls = NULL;
  t->ncalls = t->cap = 0;
  cst_buf_free(&t->text);
  cst_buf_free(&t->raw);
  cst_buf_free(&t->think);
  if (t->mm)
    mm_scan_destroy(t->mm);
  if (t->ctl)
    ctl_parser_destroy(t->ctl);
  t->mm = NULL;
  t->ctl = NULL;
}

static uint64_t call_hash(const ctl_tool_call_t *c) {
  uint64_t h = cst_fnv1a64(CST_FNV_INIT, c->name, strlen(c->name));
  for (int i = 0; i < c->nargs; i++) {
    h = cst_fnv1a64(h, "\x1f", 1);
    h = cst_fnv1a64(h, c->args[i].key, strlen(c->args[i].key));
    h = cst_fnv1a64(h, "=", 1);
    if (c->args[i].value)
      h = cst_fnv1a64(h, c->args[i].value, c->args[i].value_len);
  }
  return h;
}

static void args_detail(const ctl_tool_call_t *c, char *out, size_t n) {
  size_t used = 0;
  out[0] = 0;
  for (int i = 0; i < c->nargs && used + 8 < n; i++) {
    const char *v = c->args[i].value ? c->args[i].value : "";
    int k = snprintf(out + used, n - used, "%s%s=%.*s", i ? " " : "",
                     c->args[i].key, (int)(c->args[i].value_len > 160
                                               ? 160
                                               : c->args[i].value_len),
                     v);
    if (k < 0)
      break;
    used += (size_t)k;
    if (used >= n)
      used = n - 1;
  }
}

static void set_output(pcall_t *pc, const char *s, size_t n, int code) {
  free(pc->output);
  size_t keep = n;
  int trunc = 0;
  if (keep > TOOL_OUT_CAP) {
    keep = cst_utf8_floor(s, TOOL_OUT_CAP);
    trunc = 1;
  }
  cst_buf_t b = {0};
  cst_buf_append(&b, s ? s : "", keep);
  if (trunc)
    cst_buf_printf(&b, "\n...(truncated, %zu bytes total)", n);
  pc->output_len = b.len;
  pc->output = cst_buf_steal(&b);
  pc->exit_code = code;
}

static void append_output(pcall_t *pc, const char *note) {
  cst_buf_t b = {0};
  cst_buf_append(&b, pc->output ? pc->output : "", pc->output_len);
  cst_buf_puts(&b, note);
  free(pc->output);
  pc->output_len = b.len;
  pc->output = cst_buf_steal(&b);
}

static void emit_call_metrics(turn_t *t, const pcall_t *pc) {
  cJSON *o = cJSON_CreateObject();
  cJSON_AddStringToObject(o, "type", "call");
  cJSON_AddNumberToObject(o, "turn", t->a->turn_no);
  cJSON_AddNumberToObject(o, "step", t->step);
  cJSON_AddStringToObject(o, "protocol", cst_protocol_name(t->proto));
  cJSON_AddStringToObject(o, "origin", pc->origin ? pc->origin : "");
  cJSON_AddStringToObject(o, "id", pc->call.id);
  cJSON_AddStringToObject(o, "name", pc->call.name);
  cJSON_AddNumberToObject(o, "t_call_open", ms2(pc->t_open));
  cJSON_AddNumberToObject(o, "t_call_ready", ms2(pc->t_ready));
  cJSON_AddNumberToObject(o, "t_exec_start", ms2(pc->t_exec_start));
  cJSON_AddNumberToObject(o, "t_exec_end", ms2(pc->t_exec_end));
  cJSON_AddNumberToObject(o, "exit", pc->exit_code);
  cJSON_AddNumberToObject(o, "bytes_out", (double)pc->output_len);
  static const char *const st[] = {"ok", "denied", "dup", "unknown",
                                   "invalid_args"};
  cJSON_AddStringToObject(o, "status", st[pc->status]);
  cst_metrics_emit(t->a->metrics, o);
}

static void undo_snapshot(cst_agent_t *a, const ctl_tool_call_t *c) {
  char paths[CST_TOOL_MAX_TARGETS][1024];
  int n = cst_tool_targets(c, paths, CST_TOOL_MAX_TARGETS);
  if (n <= 0)
    return;
  char sid[256];
  if (a->cfg.undo_sid[0])
    snprintf(sid, sizeof(sid), "%s", a->cfg.undo_sid);
  else
    cst_agent_session_id(a, sid, sizeof(sid));
  for (int i = 0; i < n; i++)
    cst_undo_snapshot(a->cfg.workspace_root, sid, a->undo_tag, paths[i]);
}

void cst_agent_undo_key(const cst_agent_t *a, char *sid, size_t sid_sz,
                        char *tag, size_t tag_sz) {
  if (a->cfg.undo_sid[0])
    snprintf(sid, sid_sz, "%s", a->cfg.undo_sid);
  else
    cst_agent_session_id(a, sid, sid_sz);
  snprintf(tag, tag_sz, "%s", a->undo_tag);
}

static void agent_set_todo(const char *text, void *ud) {
  turn_t *t = ud;
  free(t->a->todo);
  t->a->todo = text ? strdup(text) : NULL;
  if (t->hooks && t->hooks->on_todo)
    t->hooks->on_todo(text, t->hooks->userdata);
}

static int agent_ask(const char *q, const char *opts, char *ans, size_t n,
                     void *ud) {
  turn_t *t = ud;
  return t->hooks->on_ask_user(q, opts, ans, n, t->hooks->userdata);
}

static const char *tool_cmd(const ctl_tool_call_t *c) {
  for (int i = 0; i < c->nargs; i++) {
    if (strcmp(c->args[i].key, "cmd") == 0 ||
        strcmp(c->args[i].key, "command") == 0)
      return c->args[i].value ? c->args[i].value : "";
  }
  return NULL;
}

static int needs_confirm(const cst_agent_config_t *cfg, cst_tool_perm_t p) {
  switch (p) {
  case CST_TP_SHELL:
    return cfg->confirm_shell;
  case CST_TP_WRITE:
    return cfg->confirm_write;
  case CST_TP_NET:
    return cfg->confirm_net;
  default:
    return 0;
  }
}

/* A sandboxed run_shell looked blocked: ask whether to rerun it outside the
 * sandbox. On yes, *tr is replaced by the unsandboxed result; on no, a note
 * tells the model why it failed. */
static void sandbox_escalate(turn_t *t, cst_tool_ctx_t *ctx,
                             const ctl_tool_call_t *c, const char *cmd,
                             cst_tool_result_t *tr) {
  cst_agent_t *a = t->a;
  int ans = CST_PERM_DENY;
  if (t->hooks && t->hooks->on_permission && !a->cancel) {
    size_t tail = tr->out_len > 1500 ? 1500 : tr->out_len;
    cst_buf_t d = {0};
    cst_buf_puts(&d, CST_PERM_SANDBOX);
    if (tr->out)
      cst_buf_append(&d, tr->out + tr->out_len - tail, tail);
    char *detail = cst_buf_steal(&d);
    hook_status(t, "blocked by the sandbox");
    ans = t->hooks->on_permission(c->name, detail ? detail : CST_PERM_SANDBOX,
                                  cmd, t->hooks->userdata);
    free(detail);
  }
  if (ans == CST_PERM_DENY) {
    const char *note =
        "\n[cst: blocked by the sandbox (writes are limited to the "
        "workspace and temp dir); the user did not allow running it "
        "outside]\n";
    cst_buf_t b = {0};
    if (tr->out)
      cst_buf_append(&b, tr->out, tr->out_len);
    cst_buf_puts(&b, note);
    free(tr->out);
    tr->out_len = b.len;
    tr->out = cst_buf_steal(&b);
    return;
  }
  if (ans == CST_PERM_ALWAYS_PREFIX) {
    char pre[128];
    cst_perms_suggest_prefix(cmd, pre, sizeof(pre));
    if (pre[0])
      cst_perms_allow_prefix(a->perms, pre);
  } else if (ans == CST_PERM_ALWAYS_TOOL) {
    cst_perms_allow_tool(a->perms, c->name);
  }
  hook_status(t, "rerunning outside the sandbox");
  cst_tool_result_free(tr);
  memset(tr, 0, sizeof(*tr));
  ctx->sandbox = 0;
  cst_tool_run(ctx, c, tr);
  cst_buf_t b = {0};
  cst_buf_puts(&b, "[cst: the sandbox blocked this command; the user allowed "
                   "rerunning it outside the sandbox. Output of that rerun:]\n");
  if (tr->out)
    cst_buf_append(&b, tr->out, tr->out_len);
  free(tr->out);
  tr->out_len = b.len;
  tr->out = cst_buf_steal(&b);
}

/* One call through approval, execution and bookkeeping. exec_begin and
 * exec_end run on the agent thread (they call hooks); exec_run may run on a
 * worker thread for read-only batches. */
typedef struct {
  cst_tool_ctx_t ctx;
  const cst_tool_def_t *def;
  const char *cmd;
  uint64_t h;
  int run;
  int mutating;
  cst_tool_result_t tr;
} exec_t;

static int fail_count(cst_agent_t *a, uint64_t h, int bump) {
  for (int i = 0; i < a->nfails; i++) {
    if (a->fails[i].hash == h)
      return bump ? ++a->fails[i].count : a->fails[i].count;
  }
  if (!bump)
    return 0;
  int slot = a->nfails < FAIL_FP_MAX ? a->nfails++ : (int)(h % FAIL_FP_MAX);
  a->fails[slot].hash = h;
  a->fails[slot].count = 1;
  return 1;
}

static void exec_begin(turn_t *t, pcall_t *pc, exec_t *x) {
  memset(x, 0, sizeof(*x));
  pc->done = 1;
  cst_agent_t *a = t->a;
  ctl_tool_call_t *c = &pc->call;
  pc->t_exec_start = rel_ms(t);
  char detail[512];
  args_detail(c, detail, sizeof(detail));
  const cst_tool_def_t *def = cst_tool_find(c->name);
  x->def = def;

  cst_tool_ctx_t *ctx = &x->ctx;
  ctx->workspace_root = a->cfg.workspace_root;
  ctx->compact = a->cfg.tool_profile != CST_TOOLS_MINIMAL;
  ctx->set_todo = agent_set_todo;
  ctx->todo_ud = t;
  ctx->agent = a;
  ctx->hooks = t->hooks;
  if (t->hooks && t->hooks->on_ask_user) {
    ctx->ask = agent_ask;
    ctx->ask_ud = t;
  }

  if (a->cancel) {
    const char *msg = "error: cancelled by the user before it ran";
    set_output(pc, msg, strlen(msg), 130);
    pc->status = PC_DENIED;
    return;
  }
  if (!def || !cst_tool_active(&a->cfg, def)) {
    char *names = cst_tools_names(&a->cfg);
    char msg[768];
    snprintf(msg, sizeof(msg), "error: unknown tool \"%.64s\" (allowed: %s)",
             c->name, names ? names : "");
    free(names);
    set_output(pc, msg, strlen(msg), 1);
    pc->status = PC_UNKNOWN;
    hook_status(t, "unknown tool");
    return;
  }
  /* Plan mode keeps these in tools[] (a stable prefix) and refuses here. */
  if (a->cfg.plan_mode &&
      (def->perm == CST_TP_WRITE || def->perm == CST_TP_SHELL)) {
    char msg[320];
    snprintf(msg, sizeof(msg),
             "error: Plan mode is on (read-only), so %.64s was not run. Keep "
             "investigating with read-only tools and answer with a plan; the "
             "user switches to Build mode to carry it out.",
             c->name);
    set_output(pc, msg, strlen(msg), 1);
    pc->status = PC_DENIED;
    hook_tool(t, "denied", c, detail);
    return;
  }
  if (pc->invalid) {
    const char *msg = "error: tool arguments are not a valid JSON object (or "
                      "exceed 1 MiB); nothing was executed";
    set_output(pc, msg, strlen(msg), 1);
    pc->status = PC_INVALID;
    return;
  }
  /* a sub-agent that may write changes files like a write tool does */
  x->mutating = def->perm == CST_TP_SHELL || def->perm == CST_TP_WRITE ||
                (strcmp(def->name, "task") == 0 &&
                 !(def->parallel && def->parallel(c, def->ud)));
  uint64_t h = call_hash(c);
  x->h = h;
  int dup = 0;
  /* plugin and MCP results can change between identical calls */
  int external = def->owner && strncmp(def->owner, "builtin:", 8) != 0;
  if (def->perm == CST_TP_READ && !external) {
    for (int i = 0; i < a->nreads; i++) {
      if (a->reads[i].hash == h && a->reads[i].epoch == a->write_epoch) {
        dup = 1;
        break;
      }
    }
  } else if (def->perm == CST_TP_WRITE && ctx->compact) {
    /* the same write already applied and nothing changed since */
    for (int i = 0; i < a->nwrites; i++) {
      if (a->writes[i].hash == h && a->writes[i].epoch == a->write_epoch) {
        dup = 2;
        break;
      }
    }
  }
  int allow = 1, ask = 0;
  const char *cmd = def->perm == CST_TP_SHELL ? tool_cmd(c) : NULL;
  x->cmd = cmd;
  if (!dup && needs_confirm(&a->cfg, def->perm) && !a->cfg.auto_approve &&
      !cst_perms_tool_allowed(a->perms, c->name) &&
      !(cmd && cst_perms_shell_allowed(a->perms, cmd))) {
    ask = 1;
    if (a->cfg.approval == CST_APPROVAL_AUTO) {
      if (def->perm == CST_TP_WRITE) {
        ask = 0;
      } else if (def->perm == CST_TP_SHELL) {
        if (cst_sandbox_probe() != CST_SANDBOX_NONE) {
          ctx->sandbox = 1;
          ask = 0;
        } else if (cmd && cst_shell_is_safe(cmd)) {
          ask = 0;
        }
      }
    }
  }
  if (ask) {
    if (t->hooks && t->hooks->on_permission) {
      char *preview = NULL;
      if (def->preview)
        cst_tool_preview(ctx, c, &preview);
      else if (cmd)
        preview = strdup(cmd);
      int ans = t->hooks->on_permission(c->name, detail[0] ? detail : c->name,
                                        preview, t->hooks->userdata);
      free(preview);
      allow = ans != CST_PERM_DENY;
      if (ans == CST_PERM_ALWAYS_TOOL) {
        cst_perms_allow_tool(a->perms, c->name);
      } else if (ans == CST_PERM_ALWAYS_PREFIX && cmd) {
        char pre[128];
        cst_perms_suggest_prefix(cmd, pre, sizeof(pre));
        if (pre[0])
          cst_perms_allow_prefix(a->perms, pre);
      }
    } else {
      allow = 0;
    }
  }
  if (dup == 2) {
    const char *msg = "note: this exact edit was already applied earlier in "
                      "this turn and nothing changed since; it was not run "
                      "again. Re-read the file if unsure of its state.";
    set_output(pc, msg, strlen(msg), 0);
    pc->status = PC_DUP;
    hook_status(t, "skip repeated edit");
  } else if (dup) {
    const char *msg =
        strcmp(c->name, "read_file") == 0
            ? "note: this exact read_file already ran earlier in this turn "
              "and no file changed since; reuse that output."
            : "note: this exact call already ran earlier in this turn and "
              "no file changed since; reuse that output.";
    set_output(pc, msg, strlen(msg), 0);
    pc->status = PC_DUP;
    hook_status(t, "skip duplicate read");
  } else if (!allow) {
    set_output(pc, "denied by user", 14, 1);
    pc->status = PC_DENIED;
    hook_tool(t, "denied", c, detail);
  } else {
    hook_tool(t, "run", c, NULL);
    if (def->perm == CST_TP_WRITE)
      undo_snapshot(a, c);
    x->run = 1;
  }
}

static void exec_run(exec_t *x, const ctl_tool_call_t *c) {
  cst_tool_run(&x->ctx, c, &x->tr);
}

static void exec_end(turn_t *t, pcall_t *pc, exec_t *x) {
  cst_agent_t *a = t->a;
  ctl_tool_call_t *c = &pc->call;
  if (x->run) {
    if (x->ctx.sandbox && x->cmd &&
        cst_sandbox_blocked(x->tr.out, x->tr.out_len, x->tr.exit_code))
      sandbox_escalate(t, &x->ctx, c, x->cmd, &x->tr);
    cst_tool_result_t *tr = &x->tr;
    if (x->ctx.compact) {
      if (tr->exit_code != 0 && fail_count(a, x->h, 1) >= 2) {
        cst_buf_t b = {0};
        cst_buf_append(&b, tr->out ? tr->out : "", tr->out_len);
        cst_buf_printf(&b, "\n[cst: this exact call has now failed %d times "
                           "in this turn. Do not repeat it: read the error, "
                           "inspect the code, and change approach.]",
                       fail_count(a, x->h, 0));
        free(tr->out);
        tr->out_len = b.len;
        tr->out = cst_buf_steal(&b);
      }
    }
    set_output(pc, tr->out ? tr->out : "", tr->out_len, tr->exit_code);
    if (tr->attach && tr->nattach > 0) {
      cst_attachments_free(pc->attach, pc->nattach);
      pc->attach = cst_attachments_dup(tr->attach, tr->nattach);
      pc->nattach = pc->attach ? tr->nattach : 0;
    }
    cst_tool_result_free(tr);
    if (x->mutating)
      a->write_epoch++;
    if (x->def->perm == CST_TP_WRITE && pc->exit_code == 0) {
      int slot = a->nwrites < FAIL_FP_MAX ? a->nwrites++
                                          : (int)(x->h % FAIL_FP_MAX);
      a->writes[slot].hash = x->h;
      a->writes[slot].epoch = a->write_epoch;
    } else if (pc->exit_code == 0 && x->def->perm == CST_TP_READ) {
      int slot = a->nreads < READ_FP_MAX ? a->nreads++ : (int)(x->h % READ_FP_MAX);
      a->reads[slot].hash = x->h;
      a->reads[slot].epoch = a->write_epoch;
    }
  }
  pc->t_exec_end = rel_ms(t);
  if (t->hooks && t->hooks->on_tool_result)
    t->hooks->on_tool_result(c, pc->output, pc->output_len, pc->exit_code,
                             t->hooks->userdata);
  if (t->hooks && t->hooks->on_tool_images && pc->nattach > 0)
    t->hooks->on_tool_images(c, pc->attach, pc->nattach, t->hooks->userdata);
  if (pc->status != PC_DENIED) {
    char d[96];
    snprintf(d, sizeof(d), "%sexit=%d bytes=%zu",
             pc->status == PC_DUP ? "skipped duplicate · " : "", pc->exit_code,
             pc->output_len);
    hook_tool(t, "done", c, d);
  }
  emit_call_metrics(t, pc);
}

/* Execute (or refuse) one confirmed call. Irreversible once started. */
static void exec_call(turn_t *t, pcall_t *pc) {
  if (pc->done)
    return;
  exec_t x;
  exec_begin(t, pc, &x);
  if (x.run)
    exec_run(&x, &pc->call);
  exec_end(t, pc, &x);
}

/* Read-only calls that may share a parallel batch. */
static int parallel_ok(const turn_t *t, const pcall_t *pc) {
  if (pc->done || pc->invalid)
    return 0;
  const cst_tool_def_t *def = cst_tool_find(pc->call.name);
  if (!def || !cst_tool_active(&t->a->cfg, def))
    return 0;
  return def->perm == CST_TP_READ ||
         (def->parallel && def->parallel(&pc->call, def->ud));
}

typedef struct {
  exec_t *x;
  const ctl_tool_call_t *c;
  volatile int *cancel;
} par_job_t;

static void par_main(void *p) {
  par_job_t *j = p;
  /* not reset after: the job may run inline on the agent thread */
  os_set_cancel_flag(j->cancel);
  exec_run(j->x, j->c);
}

static void exec_parallel(turn_t *t, int from, int to) {
  exec_t xs[PAR_MAX];
  par_job_t jobs[PAR_MAX];
  void *args[PAR_MAX];
  int n = 0;
  for (int i = from; i < to; i++)
    exec_begin(t, &t->calls[i], &xs[i - from]);
  for (int i = from; i < to; i++) {
    if (!xs[i - from].run)
      continue;
    jobs[n].x = &xs[i - from];
    jobs[n].c = &t->calls[i].call;
    jobs[n].cancel = &t->a->cancel;
    args[n] = &jobs[n];
    n++;
  }
  if (n > 1)
    hook_status(t, "running reads in parallel");
  os_run_parallel(par_main, args, n);
  for (int i = from; i < to; i++)
    exec_end(t, &t->calls[i], &xs[i - from]);
}

/* Calls left after the stream: consecutive read-only calls run together. */
static void exec_pending(turn_t *t) {
  int par = t->a->cfg.tool_profile != CST_TOOLS_MINIMAL;
  for (int i = 0; i < t->ncalls;) {
    int j = i;
    while (par && j < t->ncalls && j - i < PAR_MAX &&
           parallel_ok(t, &t->calls[j]))
      j++;
    if (j - i >= 2) {
      exec_parallel(t, i, j);
      i = j;
    } else {
      exec_call(t, &t->calls[i]);
      i++;
    }
  }
}

/* --- stream routing ------------------------------------------------------ */

static void on_text(turn_t *t, const char *s, size_t n) {
  if (t->stop || !n)
    return;
  cst_buf_append(&t->text, s, n);
  if (t->hooks && t->hooks->on_assistant_delta)
    t->hooks->on_assistant_delta(s, n, t->hooks->userdata);
}

static int is_blank(const char *s, size_t n) {
  for (size_t i = 0; i < n; i++) {
    if (s[i] != ' ' && s[i] != '\t' && s[i] != '\r' && s[i] != '\n')
      return 0;
  }
  return 1;
}

static void add_text_call(turn_t *t, const ctl_tool_call_t *c,
                          const char *origin) {
  if (t->stop)
    return;
  if (t->final_pass) {
    t->stop = 1; /* no tools allowed now: keep the text, drop the call */
    return;
  }
  pcall_t *pc = push_call(t);
  if (!pc)
    return;
  copy_call(&pc->call, c);
  if (!pc->call.id[0])
    snprintf(pc->call.id, sizeof(pc->call.id), "%c%d", origin[0],
             t->ncalls);
  pc->origin = origin;
  pc->t_open = cst_protocol_is_ctl(t->proto) ? t->ctl_open_t : rel_ms(t);
  pc->t_ready = rel_ms(t);
  hook_tool(t, "ready", &pc->call, NULL);
  t->saw_ready = 1;
  t->raw_mark = t->raw.len;
  if (cst_protocol_is_eager(t->proto))
    exec_call(t, &t->calls[t->ncalls - 1]);
}

static void on_ctl(const ctl_event_t *ev, void *ud) {
  turn_t *t = ud;
  switch (ev->type) {
  case CTL_EV_TEXT:
    if (t->saw_ready && !is_blank(ev->text, ev->text_len)) {
      /* prose after a confirmed call: likely an invented result. Stop. */
      t->stop = 1;
      return;
    }
    on_text(t, ev->text, ev->text_len);
    break;
  case CTL_EV_CALL_OPEN:
    t->ctl_open_t = rel_ms(t);
    if (!t->stop)
      hook_tool(t, "open", ev->call, NULL);
    break;
  case CTL_EV_ARG:
    if (!t->stop)
      hook_tool(t, "arg", ev->call, NULL);
    break;
  case CTL_EV_CALL_READY:
    add_text_call(t, ev->call, "ctl");
    break;
  case CTL_EV_ERROR:
    t->parse_error = 1;
    snprintf(t->perr, sizeof(t->perr), "%s", ev->error ? ev->error : "ctl");
    break;
  default:
    break;
  }
}

static void on_mm(const ctl_event_t *ev, void *ud) {
  turn_t *t = ud;
  switch (ev->type) {
  case CTL_EV_TEXT:
    on_text(t, ev->text, ev->text_len);
    break;
  case CTL_EV_CALL_OPEN:
  case CTL_EV_ARG:
    hook_tool(t, ev->type == CTL_EV_ARG ? "arg" : "open", ev->call, NULL);
    break;
  case CTL_EV_CALL_READY:
    /* native modes: XML is only a fallback when no native call came in */
    if (cst_protocol_is_native(t->proto) && t->n_native > 0)
      return;
    add_text_call(t, ev->call, "xml");
    break;
  case CTL_EV_ERROR:
    t->parse_error = 1;
    snprintf(t->perr, sizeof(t->perr), "%s", ev->error ? ev->error : "xml");
    break;
  default:
    break;
  }
}

static void feed_visible(turn_t *t, const char *data, size_t 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) {
  if (n && t->a->caps.reasoning_passback)
    cst_buf_append(&t->think, s, n);
  if (t->hooks && t->hooks->on_think_delta && n)
    t->hooks->on_think_delta(s, n, t->hooks->userdata);
}

/* Strip <think>...</think> (MiniMax inline / reasoning_content wrapper). */
static void on_delta(const char *data, size_t len, void *ud) {
  turn_t *t = ud;
  for (size_t i = 0; i < len; i++) {
    char c = data[i];
    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, data + i, j - i);
      else
        feed_visible(t, data + i, j - i);
      i = j - 1;
      continue;
    }
    t->tagbuf[t->taglen++] = c;
    if (t->taglen <= wl && strncmp(want, t->tagbuf, t->taglen) == 0) {
      if (t->taglen == wl) {
        t->in_think = !t->in_think;
        t->taglen = 0;
      }
      continue;
    }
    /* not a tag: flush all but a possible new '<' start */
    size_t flush = t->taglen;
    int restart = 0;
    if (c == '<' && t->taglen > 1) {
      flush--;
      restart = 1;
    }
    if (t->in_think)
      think_out(t, t->tagbuf, flush);
    else
      feed_visible(t, t->tagbuf, flush);
    t->taglen = 0;
    if (restart)
      t->tagbuf[t->taglen++] = '<';
  }
}

/* A native call starts: release prose still held by the <think> splitter or
 * the XML scanner so it reaches the UI before the tool card. */
static void flush_held_text(turn_t *t) {
  if (t->taglen && !t->in_think) {
    feed_visible(t, t->tagbuf, t->taglen);
    t->taglen = 0;
  }
  if (t->mm)
    mm_scan_finish(t->mm);
}

static void on_native_open(int index, const char *id, const char *name,
                           void *ud) {
  turn_t *t = ud;
  flush_held_text(t);
  t->native_opened = 1;
  if (index >= 0 && index < 64)
    t->native_open_t[index] = rel_ms(t);
  ctl_tool_call_t tmp;
  memset(&tmp, 0, sizeof(tmp));
  snprintf(tmp.id, sizeof(tmp.id), "%s", id ? id : "");
  snprintf(tmp.name, sizeof(tmp.name), "%s", name ? name : "");
  hook_tool(t, "open", &tmp, NULL);
}

static void on_native_tool(const char *id, const char *name,
                           const char *args_json, void *ud) {
  turn_t *t = ud;
  if (!name || t->final_pass)
    return;
  pcall_t *pc = push_call(t);
  if (!pc)
    return;
  int idx = t->n_native++;
  if (cst_tool_call_from_native(&pc->call, id, name, args_json) != 0)
    pc->invalid = 1;
  if (!pc->call.id[0])
    snprintf(pc->call.id, sizeof(pc->call.id), "call_%d_%d", t->step, idx + 1);
  pc->raw_args = strdup(args_json && !pc->invalid ? args_json : "{}");
  pc->origin = "native";
  pc->t_open = idx < 64 && t->native_open_t[idx] > 0 ? t->native_open_t[idx]
                                                      : rel_ms(t);
  pc->t_ready = rel_ms(t);
  hook_tool(t, "ready", &pc->call, NULL);
  if (t->proto == CST_PROTO_NATIVE_INC)
    exec_call(t, &t->calls[t->ncalls - 1]);
}

/* --- request building ---------------------------------------------------- */

typedef struct {
  cst_message_t *msgs;
  int n;
  int ntail; /* trailing request-only messages (nudge, prefix) */
  char **owned;
  int nowned;
} req_msgs_t;

static void req_msgs_free(req_msgs_t *r) {
  for (int i = 0; i < r->nowned; i++)
    free(r->owned[i]);
  free(r->owned);
  free(r->msgs);
  memset(r, 0, sizeof(*r));
}

static char *own(req_msgs_t *r, char *s) {
  if (s)
    r->owned[r->nowned++] = s;
  return s;
}

static char *result_frame(const char *id, int exit_code, const char *body,
                          size_t n, size_t cap) {
  size_t keep = n;
  cst_buf_t b = {0};
  char note[96];
  note[0] = 0;
  if (cap && n > cap) {
    keep = cst_utf8_floor(body, cap);
    snprintf(note, sizeof(note), "\n…(%zu more bytes not shown)", n - keep);
  }
  cst_buf_printf(&b, "result %s exit=%d %zu\n", id, exit_code,
                 keep + strlen(note));
  cst_buf_append(&b, body, keep);
  cst_buf_puts(&b, note);
  cst_buf_puts(&b, "\nend");
  return cst_buf_steal(&b);
}

static void render_value(cst_buf_t *b, cst_protocol_t p, const char *key,
                         const char *v) {
  size_t n = strlen(v);
  if (p == CST_PROTO_XML) {
    cst_buf_printf(b, "<parameter name=\"%s\">%s</parameter>\n", key, v);
  } else if (p == CST_PROTO_CTL_HEREDOC) {
    cst_buf_printf(b, "arg %s <<EOF_%s\n%s%s", key, key, v,
                   (n && v[n - 1] == '\n') ? "" : "\n");
    cst_buf_printf(b, "EOF_%s\n", key);
  } else {
    cst_buf_printf(b, "arg %s %zu\n", key, n);
    cst_buf_append(b, v, n);
    cst_buf_puts(b, "\n");
  }
}

/* Native assistant tool_calls rendered as text-protocol syntax. */
static char *render_calls_as_text(cst_protocol_t p, const char *content,
                                  const char *tool_calls_json) {
  cst_buf_t b = {0};
  if (content && content[0]) {
    cst_buf_puts(&b, content);
    cst_buf_puts(&b, "\n");
  }
  cJSON *arr = cJSON_Parse(tool_calls_json);
  const cJSON *tc = NULL;
  cJSON_ArrayForEach(tc, arr) {
    const cJSON *id = cJSON_GetObjectItemCaseSensitive(tc, "id");
    const cJSON *fn = cJSON_GetObjectItemCaseSensitive(tc, "function");
    const cJSON *nm = cJSON_GetObjectItemCaseSensitive(fn, "name");
    const cJSON *ar = cJSON_GetObjectItemCaseSensitive(fn, "arguments");
    const char *name = cJSON_IsString(nm) ? nm->valuestring : "tool";
    if (p == CST_PROTO_XML)
      cst_buf_printf(&b, "<minimax:tool_call>\n<invoke name=\"%s\">\n", name);
    else
      cst_buf_printf(&b, "call %s %s\n",
                     cJSON_IsString(id) ? id->valuestring : "c", name);
    cJSON *args = cJSON_IsString(ar) ? cJSON_Parse(ar->valuestring) : NULL;
    const cJSON *it = NULL;
    cJSON_ArrayForEach(it, args) {
      if (cJSON_IsString(it)) {
        render_value(&b, p, it->string, it->valuestring);
      } else {
        char *v = cJSON_PrintUnformatted(it);
        render_value(&b, p, it->string, v ? v : "");
        free(v);
      }
    }
    cJSON_Delete(args);
    cst_buf_puts(&b, p == CST_PROTO_XML ? "</invoke>\n</minimax:tool_call>\n"
                                        : "end\n");
  }
  cJSON_Delete(arr);
  return cst_buf_steal(&b);
}

int cst_message_is_note(const cst_message_t *m) {
  return m && m->role == CST_ROLE_USER && m->meta_json &&
         strstr(m->meta_json, "\"kind\":\"note\"");
}
#define msg_is_note cst_message_is_note

/* Where the request starts in the session. Normally the very first message,
 * so every request extends the previous one. CST_CTX_WINDOW=N keeps the
 * start fixed until more than N messages follow it, then jumps so N/2
 * remain: one cache miss per jump instead of one per step. When the last
 * request filled 85% of the context window the oldest half is dropped. */
static int window_start(cst_agent_t *a, int n) {
  int start = a->win_start > 0 && a->win_start < n ? a->win_start : 0;
  const char *e = getenv("CST_CTX_WINDOW");
  int win = e ? atoi(e) : 0;
  int count = 0;
  for (int i = start; i < n; i++)
    count += cst_session_get(a->session, i)->role != CST_ROLE_SYSTEM;
  int keep = -1;
  if (win > 0 && count > win)
    keep = win / 2 > 0 ? win / 2 : 1;
  long ctx = cst_model_context_tokens(a->cfg.model);
  if (ctx > 0 && a->last_prompt_tokens > ctx / 100 * 85 && count > 2) {
    int half = count / 2;
    keep = keep < 0 || half < keep ? half : keep;
    a->last_prompt_tokens = 0; /* once until the next usage report */
  }
  if (keep >= 0) {
    int seen = 0;
    for (int i = n - 1; i > start; i--) {
      if (cst_session_get(a->session, i)->role == CST_ROLE_SYSTEM)
        continue;
      if (++seen >= keep) {
        start = i;
        break;
      }
    }
  }
  a->win_start = start;
  return start;
}

static int build_request_msgs(cst_agent_t *a, cst_protocol_t proto,
                              const char *nudge, req_msgs_t *r) {
  memset(r, 0, sizeof(*r));
  int n = cst_session_count(a->session);
  r->msgs = calloc((size_t)n + 3, sizeof(cst_message_t));
  r->owned = calloc((size_t)n + 4, sizeof(char *));
  if (!r->msgs || !r->owned) {
    req_msgs_free(r);
    return -1;
  }
  /* the system prompt frozen in the session (see ensure_system) */
  const cst_message_t *sys = NULL;
  for (int i = 0; i < n && !sys; i++) {
    const cst_message_t *m = cst_session_get(a->session, i);
    if (m->role == CST_ROLE_SYSTEM)
      sys = m;
  }
  r->msgs[r->n].role = CST_ROLE_SYSTEM;
  r->msgs[r->n].content =
      sys ? sys->content
          : own(r, cst_agent_build_prompt(&a->cfg, proto, a->notes));
  r->n++;

  int text_proto = cst_protocol_is_text(proto);
  int leading = 1;
  for (int i = window_start(a, n); i < n; i++) {
    const cst_message_t *m = cst_session_get(a->session, i);
    if (m->role == CST_ROLE_SYSTEM)
      continue;
    /* never start the window with orphan results */
    if (leading && (m->role == CST_ROLE_TOOL ||
                    (m->role == CST_ROLE_USER && m->tool_call_id[0])))
      continue;
    leading = 0;
    cst_message_t mm = *m;
    mm.prefix = 0;
    if (text_proto && m->role == CST_ROLE_ASSISTANT && m->tool_calls_json) {
      mm.content = own(r, render_calls_as_text(proto, m->content,
                                               m->tool_calls_json));
      mm.tool_calls_json = NULL;
    } else if (text_proto && m->role == CST_ROLE_TOOL) {
      int code = 0;
      const char *ex = m->content ? strstr(m->content, " exit=") : NULL;
      if (ex)
        code = atoi(ex + 6);
      const char *body = m->content ? m->content : "";
      const char *nl = strchr(body, '\n');
      if (strncmp(body, "[tool ", 6) == 0 && nl)
        body = nl + 1;
      mm.role = CST_ROLE_USER;
      mm.content = own(r, result_frame(m->tool_call_id, code, body,
                                       strlen(body), a->cfg.tool_out_cap));
    }
    r->msgs[r->n++] = mm;
  }
  if (nudge) {
    r->msgs[r->n].role = CST_ROLE_USER;
    r->msgs[r->n].content = (char *)nudge;
    r->n++;
    r->ntail++;
  }
  return 0;
}

/* Request-only assistant prefix the model continues (DeepSeek beta). */
static void push_prefix(req_msgs_t *r, const char *text) {
  cst_message_t *m = &r->msgs[r->n++];
  memset(m, 0, sizeof(*m));
  m->role = CST_ROLE_ASSISTANT;
  m->content = own(r, strdup(text ? text : ""));
  m->prefix = 1;
  r->ntail++;
}

/* --- prefix cache bookkeeping -------------------------------------------- */

static uint64_t hash_str(uint64_t h, const char *s) {
  h = cst_fnv1a64(h, s ? s : "", s ? strlen(s) : 0);
  return cst_fnv1a64(h, "\x1f", 1);
}

static uint64_t msg_hash(const cst_message_t *m, int with_reasoning) {
  uint64_t h = cst_fnv1a64(CST_FNV_INIT, &m->role, sizeof(m->role));
  h = hash_str(h, m->content);
  h = hash_str(h, m->tool_call_id);
  h = hash_str(h, m->tool_calls_json);
  if (with_reasoning)
    h = hash_str(h, m->reasoning);
  for (int i = 0; i < m->nattach; i++)
    h = hash_str(h, m->attach[i].id);
  return h;
}

/* Compare this request with the previous one: t->prefix_break is the first
 * message that differs (-1: the request only appends; -2: no previous
 * request to compare with). */
static void check_prefix(cst_agent_t *a, turn_t *t, const req_msgs_t *r,
                         const char *tools_json) {
  int n = r->n - r->ntail;
  uint64_t th = hash_str(CST_FNV_INIT, tools_json);
  int brk = -1;
  const char *why = NULL;
  if (a->nprev == 0) {
    brk = -2;
  } else if (strcmp(a->prev_model, a->cfg.model) != 0) {
    brk = 0;
    why = "model";
  } else if (th != a->prev_tools) {
    brk = 0;
    why = "tools";
  }
  if (n > a->capprev) {
    uint64_t *nh = realloc(a->prev_hash, (size_t)n * sizeof(*nh));
    if (!nh) {
      a->nprev = 0;
      t->prefix_break = -2;
      return;
    }
    a->prev_hash = nh;
    a->capprev = n;
  }
  for (int i = 0; i < n; i++) {
    uint64_t h = msg_hash(&r->msgs[i], a->caps.reasoning_passback);
    if (brk == -1 && i < a->nprev && h != a->prev_hash[i]) {
      brk = i;
      why = i == 0 ? "system" : "history";
    }
    a->prev_hash[i] = h;
  }
  if (brk == -1 && n < a->nprev) {
    brk = n;
    why = "shrunk";
  }
  if (brk >= 0 && a->rebased)
    why = "rebase";
  a->rebased = 0;
  a->nprev = n;
  a->prev_tools = th;
  snprintf(a->prev_model, sizeof(a->prev_model), "%s", a->cfg.model);
  t->prefix_break = brk;
  t->prefix_reason = why;
}

static int images_blocked(const cst_agent_t *a) {
  return a->caps.vision == 0 ||
         (a->images_off && strcmp(a->images_off_model, a->cfg.model) == 0);
}

static int msgs_have_images(const req_msgs_t *r) {
  for (int i = 0; i < r->n; i++)
    if (r->msgs[i].nattach > 0)
      return 1;
  return 0;
}

static uint64_t tools_sig(const cst_agent_t *a) {
  const cst_agent_config_t *c = &a->cfg;
  uint64_t h = hash_str(CST_FNV_INIT, c->tools_off);
  h = hash_str(h, c->tools_only);
  h = cst_fnv1a64(h, &c->read_only, sizeof(c->read_only));
  h = cst_fnv1a64(h, &c->depth, sizeof(c->depth));
  int blind = images_blocked(a);
  h = cst_fnv1a64(h, &blind, sizeof(blind));
  return cst_fnv1a64(h, &c->tool_profile, sizeof(c->tool_profile));
}

/* tools[] as sent, frozen for the session: rebuilt only when the tool
 * settings change or at a rebase, so MCP servers that come up later do not
 * reshuffle the cached prefix mid-session. */
static const char *agent_tools_json(cst_agent_t *a) {
  uint64_t sig = tools_sig(a);
  if (a->tools_json && sig == a->tools_sig)
    return a->tools_json;
  if (!a->tools_json) {
    const char *e = getenv("CST_MCP_WAIT_MS");
    cst_mcp_wait_ready(e && atoi(e) >= 0 ? atoi(e) : 3000);
  }
  cst_agent_config_t c = a->cfg;
  if (images_blocked(a)) /* a model that cannot see gets no view_image */
    cst_tool_set_disabled(&c, "view_image", 1);
  cJSON *arr = cst_tools_schema_cfg(&c);
  char *s = arr ? cJSON_PrintUnformatted(arr) : NULL;
  cJSON_Delete(arr);
  if (!s)
    return a->tools_json;
  free(a->tools_json);
  a->tools_json = s;
  a->tools_sig = sig;
  return s;
}

static void reset_prefix_state(cst_agent_t *a) {
  free(a->tools_json);
  a->tools_json = NULL;
  a->win_start = 0;
  a->rebased = 1;
}

/* "https://api.deepseek.com/v1" -> "https://api.deepseek.com/beta" */
static void beta_base(const char *base, char *out, size_t n) {
  size_t l = strlen(base);
  while (l && base[l - 1] == '/')
    l--;
  if (l >= 3 && strncmp(base + l - 3, "/v1", 3) == 0)
    l -= 3;
  else if (l >= 5 && strncmp(base + l - 5, "/beta", 5) == 0)
    l -= 5;
  snprintf(out, n, "%.*s/beta", (int)l, base);
}

static void fill_request(cst_agent_t *a, cst_llm_request_t *req,
                         const req_msgs_t *rm, cst_protocol_t proto) {
  memset(req, 0, sizeof(*req));
  cst_model_caps(a->cfg.provider, a->cfg.api_base, a->cfg.model, &a->caps);
  req->api_base = a->cfg.api_base;
  req->api_key = a->cfg.api_key;
  req->model = a->cfg.model;
  req->reasoning_effort =
      cst_effort_wire(a->cfg.provider, a->cfg.model, a->cfg.effort);
  req->msgs = rm->msgs;
  req->nmsgs = rm->n;
  req->session_path = cst_session_path(a->session);
  req->cancel_flag = &a->cancel;
  req->tool_out_cap = a->cfg.tool_out_cap;
  req->temperature = a->cfg.temperature;
  req->top_p = a->cfg.top_p;
  req->max_tokens = a->cfg.max_tokens;
  if (cst_protocol_is_native(proto)) {
    req->enable_tools = 1;
    req->tool_cfg = &a->cfg;
    req->tools_json = agent_tools_json(a);
  }
  req->send_reasoning = a->caps.reasoning_passback;
  req->images_off = images_blocked(a);
  if (!a->hints_off) {
    req->cache_style = a->caps.cache_style;
    cst_agent_session_id(a, a->cache_key, sizeof(a->cache_key));
    req->cache_key = a->cache_key;
    req->cache_ttl = getenv("CST_CACHE_TTL");
  }
}

/* Count one response into the session and turn totals; returns its cost,
 * or -1 when the model has no known price. */
static double account(cst_agent_t *a, const cst_llm_stats_t *st) {
  if (st->prompt_tokens <= 0 && st->completion_tokens <= 0)
    return -1;
  cst_price_t p;
  int priced = cst_model_price(a->cfg.model, &p) == 0;
  double c = priced ? cst_price_cost(&p, st->cache_miss_tokens,
                                     st->cached_tokens, st->cache_write_tokens,
                                     st->completion_tokens, time(NULL))
                    : -1;
  cst_usage_t *us[2] = {&a->usage, &a->turn_usage};
  for (int i = 0; i < 2; i++) {
    cst_usage_t *u = us[i];
    if (u->requests == 0) {
      u->priced = 1;
      snprintf(u->currency, sizeof(u->currency), "%s",
               priced ? p.currency : "");
    }
    u->requests++;
    u->prompt += st->prompt_tokens;
    u->cached += st->cached_tokens;
    u->miss += st->cache_miss_tokens;
    u->write += st->cache_write_tokens;
    u->out += st->completion_tokens;
    if (priced && strcmp(u->currency, p.currency) == 0)
      u->cost += c;
    else
      u->priced = 0;
  }
  return c;
}

void cst_agent_usage(const cst_agent_t *a, cst_usage_t *session,
                     cst_usage_t *turn) {
  if (session)
    memset(session, 0, sizeof(*session));
  if (turn)
    memset(turn, 0, sizeof(*turn));
  if (!a)
    return;
  if (session)
    *session = a->usage;
  if (turn)
    *turn = a->turn_usage;
}

/* --- persistence --------------------------------------------------------- */

static char *meta_json(const char *kind, cst_protocol_t p) {
  cJSON *o = cJSON_CreateObject();
  if (kind)
    cJSON_AddStringToObject(o, "kind", kind);
  cJSON_AddStringToObject(o, "protocol", cst_protocol_name(p));
  char *s = cJSON_PrintUnformatted(o);
  cJSON_Delete(o);
  return s;
}

static void append_msg_full(cst_agent_t *a, cst_role_t role,
                            const char *content, const char *tool_id,
                            const char *tool_name, const char *tool_calls_json,
                            char *meta, const cst_attachment_t *attach,
                            int nattach, const char *reasoning) {
  cst_message_t m;
  memset(&m, 0, sizeof(m));
  m.role = role;
  m.content = (char *)(content ? content : "");
  m.reasoning = (char *)reasoning;
  if (tool_id)
    snprintf(m.tool_call_id, sizeof(m.tool_call_id), "%s", tool_id);
  if (tool_name)
    snprintf(m.tool_name, sizeof(m.tool_name), "%s", tool_name);
  m.tool_calls_json = (char *)tool_calls_json;
  m.meta_json = meta;
  if (attach && nattach > 0) {
    m.attach = cst_attachments_dup(attach, nattach);
    m.nattach = m.attach ? nattach : 0;
  }
  cst_session_append(a->session, &m);
  cst_attachments_free(m.attach, m.nattach);
  free(meta);
}

static void append_msg(cst_agent_t *a, cst_role_t role, const char *content,
                       const char *tool_id, const char *tool_name,
                       const char *tool_calls_json, char *meta) {
  append_msg_full(a, role, content, tool_id, tool_name, tool_calls_json, meta,
                  NULL, 0, NULL);
}

/* An assistant reply, with the turn's reasoning when the provider wants the
 * history to keep it (DeepSeek thinking + tools). */
static void append_reply(turn_t *t, const char *content,
                         const char *tool_calls_json, char *meta) {
  const char *r = t->a->caps.reasoning_passback && t->think.len
                      ? t->think.buf
                      : NULL;
  append_msg_full(t->a, CST_ROLE_ASSISTANT, content, NULL, NULL,
                  tool_calls_json, meta, NULL, 0, r);
}

static char *build_tool_calls_json(const turn_t *t) {
  cJSON *arr = cJSON_CreateArray();
  for (int i = 0; i < t->ncalls; i++) {
    const pcall_t *pc = &t->calls[i];
    cJSON *o = cJSON_CreateObject();
    cJSON_AddStringToObject(o, "id", pc->call.id);
    cJSON_AddStringToObject(o, "type", "function");
    cJSON *fn = cJSON_AddObjectToObject(o, "function");
    cJSON_AddStringToObject(fn, "name", pc->call.name);
    if (pc->raw_args) {
      cJSON_AddStringToObject(fn, "arguments", pc->raw_args);
    } else {
      cJSON *args = cJSON_CreateObject();
      for (int k = 0; k < pc->call.nargs; k++)
        cJSON_AddStringToObject(args, pc->call.args[k].key,
                                pc->call.args[k].value ? pc->call.args[k].value
                                                       : "");
      char *s = cJSON_PrintUnformatted(args);
      cJSON_Delete(args);
      cJSON_AddStringToObject(fn, "arguments", s ? s : "{}");
      free(s);
    }
    cJSON_AddItemToArray(arr, o);
  }
  char *s = cJSON_PrintUnformatted(arr);
  cJSON_Delete(arr);
  return s;
}

static void persist_calls(turn_t *t) {
  cst_agent_t *a = t->a;
  if (cst_protocol_is_text(t->proto)) {
    size_t n = t->stop ? t->raw_mark : t->raw.len;
    char *raw = malloc(n + 1);
    if (raw) {
      memcpy(raw, t->raw.buf ? t->raw.buf : "", n);
      raw[n] = 0;
    }
    append_reply(t, raw ? raw : "", NULL, meta_json(NULL, t->proto));
    free(raw);
    for (int i = 0; i < t->ncalls; i++) {
      pcall_t *pc = &t->calls[i];
      char *frame = result_frame(pc->call.id, pc->exit_code,
                                 pc->output ? pc->output : "", pc->output_len,
                                 a->cfg.tool_out_cap);
      append_msg(a, CST_ROLE_USER, frame, pc->call.id, pc->call.name, NULL,
                 meta_json("tool_result", t->proto));
      free(frame);
    }
    return;
  }
  char *tcj = build_tool_calls_json(t);
  append_reply(t, t->text.buf ? trim_left(t->text.buf) : "", tcj,
               meta_json(NULL, t->proto));
  free(tcj);
  for (int i = 0; i < t->ncalls; i++) {
    pcall_t *pc = &t->calls[i];
    cst_buf_t b = {0};
    cst_buf_printf(&b, "[tool %s id=%s exit=%d]\n", pc->call.name, pc->call.id,
                   pc->exit_code);
    cst_buf_append(&b, pc->output ? pc->output : "", pc->output_len);
    append_msg_full(a, CST_ROLE_TOOL, b.buf, pc->call.id, pc->call.name, NULL,
                    meta_json(NULL, t->proto), pc->attach, pc->nattach, NULL);
    cst_buf_free(&b);
  }
}

/* --- agent lifecycle ----------------------------------------------------- */

void cst_agent_session_id(const cst_agent_t *a, char *out, size_t out_sz) {
  if (!out || !out_sz)
    return;
  out[0] = 0;
  if (!a)
    return;
  const char *p = a->cfg.session_path;
  const char *sl = strrchr(p, '/');
  const char *base = sl ? sl + 1 : p;
  size_t n = strlen(base);
  if (n >= out_sz)
    n = out_sz - 1;
  memcpy(out, base, n);
  out[n] = 0;
  if (n > 6 && strcmp(out + n - 6, ".jsonl") == 0)
    out[n - 6] = 0;
}

const char *cst_agent_session_path(const cst_agent_t *a) {
  return a && a->session ? cst_session_path(a->session) : NULL;
}

void cst_agent_pending_clear(cst_agent_t *a) {
  if (!a)
    return;
  cst_attachments_free(a->pending, a->npending);
  a->pending = NULL;
  a->npending = a->cappending = 0;
}

int cst_agent_pending_count(const cst_agent_t *a) { return a ? a->npending : 0; }

int cst_agent_pending_describe(const cst_agent_t *a, int i, char *out,
                               size_t out_sz) {
  if (!a || !out || i < 0 || i >= a->npending)
    return -1;
  cst_attach_describe(&a->pending[i], out, out_sz);
  return 0;
}

int cst_agent_sees_images(const cst_agent_t *a) {
  if (!a)
    return 0;
  cst_model_caps_t c;
  cst_model_caps(a->cfg.provider, a->cfg.api_base, a->cfg.model, &c);
  if (c.vision == 0 ||
      (a->images_off && strcmp(a->images_off_model, a->cfg.model) == 0))
    return 0;
  return c.vision;
}

/* Room for one more picked image: the session to store it in, or NULL. */
static const char *pending_room(cst_agent_t *a, long long *total, char *err,
                                size_t err_sz) {
  if (a->npending >= CST_ATTACH_MAX_COUNT) {
    snprintf(err, err_sz, "at most %d images per message", CST_ATTACH_MAX_COUNT);
    return NULL;
  }
  *total = 0;
  for (int i = 0; i < a->npending; i++)
    *total += a->pending[i].bytes;
  if (*total >= CST_ATTACH_MAX_TOTAL) {
    snprintf(err, err_sz, "the images already picked fill %d MB",
             (int)(CST_ATTACH_MAX_TOTAL >> 20));
    return NULL;
  }
  const char *session = cst_agent_session_path(a);
  if (!session || !session[0]) {
    snprintf(err, err_sz, "no session to store the image in");
    return NULL;
  }
  return session;
}

static int pending_add(cst_agent_t *a, cst_attachment_t *one, long long total,
                       char *err, size_t err_sz) {
  if (!one)
    return -1;
  int rc = 0;
  if (one->bytes > CST_ATTACH_MAX_TOTAL - total ||
      cst_attachments_push(&a->pending, &a->npending, &a->cappending, one) !=
          0) {
    snprintf(err, err_sz, "the images together exceed %d MB",
             (int)(CST_ATTACH_MAX_TOTAL >> 20));
    rc = -1;
  }
  cst_attach_free(one);
  return rc;
}

int cst_agent_attach_file(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, "no file to attach");
    return -1;
  }
  long long total;
  const char *session = pending_room(a, &total, err, err_sz);
  if (!session)
    return -1;
  char full[OS_PATH_MAX];
  if (os_realpath(path, full, sizeof(full)) != 0)
    snprintf(full, sizeof(full), "%s", path);
  return pending_add(a, cst_attach_import_file(session, full, err, err_sz),
                     total, err, err_sz);
}

int cst_agent_attach_bytes(cst_agent_t *a, const char *name, const char *data,
                           size_t n, char *err, size_t err_sz) {
  char dummy[8];
  if (!err || !err_sz) {
    err = dummy;
    err_sz = sizeof(dummy);
  }
  err[0] = 0;
  if (!a) {
    snprintf(err, err_sz, "no agent");
    return -1;
  }
  long long total;
  const char *session = pending_room(a, &total, err, err_sz);
  if (!session)
    return -1;
  return pending_add(
      a, cst_attach_import_bytes(session, name, data, n, err, err_sz), total,
      err, err_sz);
}

/* The stored name of a pending attachment: the window serves the file from the
 * session's attachment directory while the message is still being written. */
const char *cst_agent_pending_rel(const cst_agent_t *a, int i) {
  if (!a || i < 0 || i >= a->npending)
    return NULL;
  return a->pending[i].rel;
}

int cst_agent_pending_is_image(const cst_agent_t *a, int i) {
  if (!a || i < 0 || i >= a->npending)
    return 0;
  return cst_attach_is_image(&a->pending[i]);
}

const char *cst_agent_todo(const cst_agent_t *a) { return a ? a->todo : NULL; }

const char *cst_agent_notes(const cst_agent_t *a) {
  return a ? a->notes : NULL;
}

long cst_agent_last_prompt_tokens(const cst_agent_t *a) {
  return a ? a->last_prompt_tokens : 0;
}

void cst_agent_reload_notes(cst_agent_t *a) {
  if (!a)
    return;
  free(a->notes);
  a->notes = NULL;
  const char *e = getenv("CST_NO_PROJECT_NOTES");
  if (e && e[0] == '1')
    return;
  static const char *const names[] = {"AGENTS.md", "CST.md"};
  for (size_t i = 0; i < sizeof(names) / sizeof(names[0]); i++) {
    char p[4200];
    snprintf(p, sizeof(p), "%s/%s", a->cfg.workspace_root, names[i]);
    FILE *fp = fopen(p, "rb");
    if (!fp)
      continue;
    cst_buf_t b = {0};
    char tmp[4096];
    size_t n;
    while ((n = fread(tmp, 1, sizeof(tmp), fp)) > 0 &&
           b.len < PROJECT_NOTES_MAX)
      cst_buf_append(&b, tmp, n);
    fclose(fp);
    if (b.len && b.buf) {
      size_t clean_len = 0;
      size_t bom = cst_bom_len(b.buf, b.len);
      a->notes = cst_utf8_sanitize(b.buf + bom, b.len - bom, &clean_len);
    }
    cst_buf_free(&b);
    return;
  }
}

/* Latest todo_write result in the session (tool output lists `[ ] x`). */
/* Rough context size of a reopened session until the next real usage
 * report (about 4 bytes per token). */
static void estimate_tokens(cst_agent_t *a) {
  size_t bytes = 0;
  cst_session_lock(a->session);
  for (int i = 0; i < cst_session_count(a->session); i++) {
    const cst_message_t *m = cst_session_get(a->session, i);
    bytes += (m->content ? strlen(m->content) : 0) +
             (m->tool_calls_json ? strlen(m->tool_calls_json) : 0);
  }
  cst_session_unlock(a->session);
  a->last_prompt_tokens = (long)(bytes / 4);
}

static void restore_todo(cst_agent_t *a) {
  free(a->todo);
  a->todo = NULL;
  cst_session_lock(a->session);
  int n = cst_session_count(a->session);
  for (int i = n - 1; i >= 0; i--) {
    const cst_message_t *m = cst_session_get(a->session, i);
    if (!m || strcmp(m->tool_name, "todo_write") != 0 || !m->content)
      continue;
    cst_buf_t b = {0};
    const char *p = m->content;
    while (*p) {
      const char *e = strchr(p, '\n');
      size_t len = e ? (size_t)(e - p) : strlen(p);
      if (len >= 3 && p[0] == '[' && p[2] == ']' &&
          (p[1] == ' ' || p[1] == 'x' || p[1] == '~')) {
        cst_buf_append(&b, p, len);
        cst_buf_puts(&b, "\n");
      }
      if (!e)
        break;
      p = e + 1;
    }
    if (b.len) {
      a->todo = cst_buf_steal(&b);
      break;
    }
    cst_buf_free(&b);
  }
  cst_session_unlock(a->session);
}

int cst_agent_undo(cst_agent_t *a, char *report, size_t report_sz) {
  if (!a || !report || !report_sz)
    return -1;
  char sid[256];
  cst_agent_session_id(a, sid, sizeof(sid));
  int rc = cst_undo_restore_last(a->cfg.workspace_root, sid, report,
                                 report_sz);
  if (rc == 0)
    a->write_epoch++;
  return rc;
}

cst_perms_t *cst_agent_perms(cst_agent_t *a) { return a ? a->perms : NULL; }

/* Drop the last user prompt and everything after it; returns the prompt. */
char *cst_agent_rewind_last_turn(cst_agent_t *a) {
  if (!a || !a->session)
    return NULL;
  cst_session_t *s = a->session;
  cst_session_lock(s);
  int n = cst_session_count(s), idx = -1;
  for (int i = n - 1; i > 0; i--) {
    const cst_message_t *m = cst_session_get(s, i);
    if (m->role == CST_ROLE_USER && !m->tool_call_id[0] && m->content &&
        !msg_is_note(m) &&
        strncmp(m->content, "[session compacted", 18) != 0) {
      idx = i;
      break;
    }
  }
  char *prompt = idx > 0 ? strdup(cst_session_get(s, idx)->content) : NULL;
  int nimg = 0;
  cst_attachment_t *imgs = NULL;
  if (prompt) {
    const cst_message_t *m = cst_session_get(s, idx);
    imgs = cst_attachments_dup(m->attach, m->nattach);
    nimg = imgs ? m->nattach : 0;
  }
  cst_session_unlock(s);
  if (!prompt)
    return NULL;
  if (cst_session_rewrite(s, cst_session_get(s, 0), idx) != 0) {
    free(prompt);
    cst_attachments_free(imgs, nimg);
    return NULL;
  }
  /* the prompt's images go back to the queue with its text */
  for (int i = 0; i < nimg && a->npending < CST_ATTACH_MAX_COUNT; i++)
    cst_attachments_push(&a->pending, &a->npending, &a->cappending, &imgs[i]);
  cst_attachments_free(imgs, nimg);
  a->write_epoch++;
  return prompt;
}

int cst_agent_undo_available(const cst_agent_t *a) {
  if (!a)
    return 0;
  char sid[256];
  cst_agent_session_id(a, sid, sizeof(sid));
  return cst_undo_count(a->cfg.workspace_root, sid);
}

/* --- frozen system prompt ------------------------------------------------
 * Every request starts with the session's system message. It is written
 * once (Build mode, the state at that moment) and changes only at a rebase;
 * later changes reach the model as persisted notes appended to the history,
 * which keeps the cached prefix intact. */

#define PROMPT_VERSION 3

typedef struct {
  int plan;
  char approval[96];
  char mode[16];
  char branch[128];
  char day[16];
  char notes[24]; /* hash of AGENTS.md, "" = none */
} pstate_t;

static void state_now(const cst_agent_t *a, pstate_t *s) {
  memset(s, 0, sizeof(*s));
  s->plan = a->cfg.plan_mode;
  snprintf(s->approval, sizeof(s->approval), "%s", approval_desc(&a->cfg));
  snprintf(s->mode, sizeof(s->mode), "%s", cst_mode_name(a->cfg.mode));
  git_branch(a->cfg.workspace_root, s->branch, sizeof(s->branch));
  today(s->day, sizeof(s->day));
  if (a->notes && a->notes[0])
    snprintf(s->notes, sizeof(s->notes), "%016llx",
             (unsigned long long)cst_fnv1a64(CST_FNV_INIT, a->notes,
                                             strlen(a->notes)));
}

static cJSON *state_json(const pstate_t *s) {
  cJSON *o = cJSON_CreateObject();
  cJSON_AddNumberToObject(o, "plan", s->plan);
  cJSON_AddStringToObject(o, "approval", s->approval);
  cJSON_AddStringToObject(o, "mode", s->mode);
  cJSON_AddStringToObject(o, "branch", s->branch);
  cJSON_AddStringToObject(o, "day", s->day);
  cJSON_AddStringToObject(o, "notes", s->notes);
  return o;
}

static void state_parse(const cJSON *o, pstate_t *s) {
  memset(s, 0, sizeof(*s));
  const cJSON *v = cJSON_GetObjectItemCaseSensitive(o, "plan");
  s->plan = cJSON_IsNumber(v) ? v->valueint : 0;
#define STATE_STR(key, field)                                                  \
  v = cJSON_GetObjectItemCaseSensitive(o, key);                                \
  if (cJSON_IsString(v))                                                       \
    snprintf(s->field, sizeof(s->field), "%s", v->valuestring);
  STATE_STR("approval", approval)
  STATE_STR("mode", mode)
  STATE_STR("branch", branch)
  STATE_STR("day", day)
  STATE_STR("notes", notes)
#undef STATE_STR
}

static char *system_meta(const cst_agent_t *a, const pstate_t *s) {
  cJSON *o = cJSON_CreateObject();
  cJSON_AddStringToObject(o, "protocol", cst_protocol_name(a->cfg.protocol));
  cJSON_AddNumberToObject(o, "pv", PROMPT_VERSION);
  cJSON_AddNumberToObject(o, "tools", a->cfg.tool_profile);
  cJSON_AddStringToObject(o, "mode", cst_mode_name(a->cfg.mode));
  cJSON_AddItemToObject(o, "state", state_json(s));
  char *out = cJSON_PrintUnformatted(o);
  cJSON_Delete(o);
  return out;
}

static char *frozen_prompt(const cst_agent_t *a, pstate_t *st) {
  cst_agent_config_t c = a->cfg;
  c.plan_mode = 0;
  state_now(a, st);
  st->plan = 0;
  return cst_agent_build_prompt(&c, c.protocol, a->notes);
}

static int system_current(const cst_message_t *m, const cst_agent_config_t *c) {
  cJSON *meta = m->meta_json ? cJSON_Parse(m->meta_json) : NULL;
  const cJSON *pv = cJSON_GetObjectItemCaseSensitive(meta, "pv");
  const cJSON *pr = cJSON_GetObjectItemCaseSensitive(meta, "protocol");
  const cJSON *tp = cJSON_GetObjectItemCaseSensitive(meta, "tools");
  const cJSON *md = cJSON_GetObjectItemCaseSensitive(meta, "mode");
  int ok = cJSON_IsNumber(pv) && pv->valueint == PROMPT_VERSION &&
           cJSON_IsString(pr) &&
           strcmp(pr->valuestring, cst_protocol_name(c->protocol)) == 0 &&
           cJSON_IsNumber(tp) && tp->valueint == c->tool_profile;
  ok = ok && cJSON_IsString(md) &&
       strcmp(md->valuestring, cst_mode_name(c->mode)) == 0;
  cJSON_Delete(meta);
  return ok;
}

/* Sessions from older versions, or after a protocol / tool profile change,
 * get a new system prompt (a one-time cache miss). */
static void ensure_system(cst_agent_t *a) {
  cst_session_t *s = a->session;
  int n = cst_session_count(s);
  for (int i = 0; i < n; i++) {
    const cst_message_t *m = cst_session_get(s, i);
    if (m->role != CST_ROLE_SYSTEM)
      continue;
    if (system_current(m, &a->cfg))
      return;
    break;
  }
  pstate_t st;
  char *sp = frozen_prompt(a, &st);
  char *meta = system_meta(a, &st);
  if (n == 0) {
    append_msg(a, CST_ROLE_SYSTEM, sp, NULL, NULL, NULL, meta);
    free(sp);
    return;
  }
  cst_message_t *v = calloc((size_t)n + 1, sizeof(*v));
  if (v) {
    v[0].role = CST_ROLE_SYSTEM;
    v[0].content = sp;
    v[0].meta_json = meta;
    int k = 1;
    for (int i = 0; i < n; i++) {
      const cst_message_t *m = cst_session_get(s, i);
      if (m->role != CST_ROLE_SYSTEM)
        v[k++] = *m;
    }
    cst_session_rewrite(s, v, k);
    free(v);
  }
  free(sp);
  free(meta);
  reset_prefix_state(a);
}

/* State as of the newest note, or the system prompt. */
static void last_state(cst_agent_t *a, pstate_t *st) {
  memset(st, 0, sizeof(*st));
  for (int i = cst_session_count(a->session) - 1; i >= 0; i--) {
    const cst_message_t *m = cst_session_get(a->session, i);
    if (!m->meta_json || !strstr(m->meta_json, "\"state\""))
      continue;
    cJSON *meta = cJSON_Parse(m->meta_json);
    state_parse(cJSON_GetObjectItemCaseSensitive(meta, "state"), st);
    cJSON_Delete(meta);
    return;
  }
}

/* One persisted note for whatever changed since the last one. */
static void sync_notes(cst_agent_t *a) {
  pstate_t was, now;
  last_state(a, &was);
  state_now(a, &now);
  cst_buf_t b = {0};
  if (now.plan != was.plan)
    cst_buf_puts(&b,
                 now.plan
                     ? "[cst] Plan mode is now ON: you may only read and "
                       "research; tools that change files or run shell "
                       "commands are refused. Investigate, then answer with a "
                       "concrete step-by-step plan: files to touch, what to "
                       "change, how to verify. The user switches to Build "
                       "mode to carry it out.\n"
                     : "[cst] Plan mode is now OFF (Build mode): every tool "
                       "is available again.\n");
  if (strcmp(now.approval, was.approval) != 0)
    cst_buf_printf(&b, "[cst] Approval mode is now %s.\n", now.approval);
  if (strcmp(now.mode, was.mode) != 0)
    cst_buf_printf(&b,
                   "[cst] Working mode is now %s%s. Earlier instructions about "
                   "how to frame the work no longer apply.\n",
                   now.mode,
                   strcmp(now.mode, "general") == 0
                       ? " (general: documents, data, research)"
                       : " (coding: repo, build, tests)");
  if (strcmp(now.branch, was.branch) != 0)
    cst_buf_printf(&b, "[cst] The git branch is now %s.\n",
                   now.branch[0] ? now.branch : "(none)");
  if (strcmp(now.day, was.day) != 0)
    cst_buf_printf(&b, "[cst] Today's date is now %s.\n", now.day);
  if (strcmp(now.notes, was.notes) != 0) {
    if (now.notes[0]) {
      size_t len = strlen(a->notes);
      if (len > PROJECT_NOTES_MAX)
        len = cst_utf8_floor(a->notes, PROJECT_NOTES_MAX);
      cst_buf_puts(&b, "[cst] The workspace's AGENTS.md changed; follow "
                       "these project instructions from now on:\n");
      cst_buf_append(&b, a->notes, len);
      cst_buf_puts(&b, "\n");
    } else {
      cst_buf_puts(&b, "[cst] The workspace's AGENTS.md was removed; its "
                       "earlier instructions no longer apply.\n");
    }
  }
  if (!b.len)
    return;
  while (b.len && b.buf[b.len - 1] == '\n')
    b.buf[--b.len] = 0;
  cJSON *o = cJSON_CreateObject();
  cJSON_AddStringToObject(o, "kind", "note");
  cJSON_AddItemToObject(o, "state", state_json(&now));
  char *meta = cJSON_PrintUnformatted(o);
  cJSON_Delete(o);
  append_msg(a, CST_ROLE_USER, b.buf, NULL, NULL, NULL, meta);
  cst_buf_free(&b);
}

cst_agent_t *cst_agent_create(const cst_agent_config_t *cfg) {
  if (!cfg)
    return NULL;
  cst_agent_t *a = calloc(1, sizeof(*a));
  if (!a)
    return NULL;
  a->cfg = *cfg;
  if (a->cfg.max_steps <= 0)
    a->cfg.max_steps = CST_DEFAULT_MAX_STEPS;
  if (a->cfg.tool_out_cap == 0)
    a->cfg.tool_out_cap = CST_DEFAULT_TOOL_OUT_CAP;
  /* The generation parameters the user saved for this provider (Settings ->
   * Advanced in the app, "params" in settings.json). They are read here so
   * that the CLI and the TUI send what the GUI was set to; a caller that
   * named one explicitly wins. */
  {
    cst_settings_t st;
    if (cst_settings_load(&st) == 0) {
      const cst_gen_params_t *g = cst_settings_params(&st, a->cfg.provider);
      if (g) {
        if (a->cfg.temperature <= 0)
          a->cfg.temperature = g->temperature;
        if (a->cfg.top_p <= 0)
          a->cfg.top_p = g->top_p;
        if (a->cfg.max_tokens <= 0)
          a->cfg.max_tokens = g->max_tokens;
      }
    }
  }
  a->session = cst_session_open(cfg->session_path[0] ? cfg->session_path
                                                     : "session.jsonl");
  if (!a->session) {
    free(a);
    return NULL;
  }
  a->metrics = cst_metrics_open(a->cfg.workspace_root,
                                cst_session_path(a->session));
  snprintf(a->cfg.session_path, sizeof(a->cfg.session_path), "%s",
           cst_session_path(a->session));
  a->perms = cst_perms_open(a->cfg.workspace_root);
  cst_agent_reload_notes(a);
  cst_model_caps(a->cfg.provider, a->cfg.api_base, a->cfg.model, &a->caps);
  ensure_system(a);
  restore_todo(a);
  estimate_tokens(a);
  return a;
}

void cst_agent_destroy(cst_agent_t *a) {
  if (!a)
    return;
  os_set_cancel_flag(NULL);
  cst_metrics_close(a->metrics);
  cst_session_close(a->session);
  free(a->notes);
  free(a->todo);
  free(a->tools_json);
  free(a->prev_hash);
  cst_agent_pending_clear(a);
  cst_perms_close(a->perms);
  free(a);
}

cst_session_t *cst_agent_session(cst_agent_t *a) { return a ? a->session : NULL; }

const cst_agent_config_t *cst_agent_config(const cst_agent_t *a) {
  return a ? &a->cfg : NULL;
}

cst_agent_config_t *cst_agent_config_mut(cst_agent_t *a) {
  return a ? &a->cfg : NULL;
}

void cst_agent_request_cancel(cst_agent_t *a) {
  if (a)
    a->cancel = 1;
}

int cst_agent_run_turn(cst_agent_t *a, const char *user_text, char *err,
                       size_t err_sz) {
  return cst_agent_run_turn_ex(a, user_text, NULL, err, err_sz);
}

static void emit_step_metrics(turn_t *t, const cst_llm_stats_t *st, int lrc,
                              double cost) {
  cJSON *o = cJSON_CreateObject();
  cJSON_AddStringToObject(o, "type", "step");
  cJSON_AddNumberToObject(o, "turn", t->a->turn_no);
  cJSON_AddNumberToObject(o, "step", t->step);
  cJSON_AddStringToObject(o, "protocol", cst_protocol_name(t->proto));
  cJSON_AddStringToObject(o, "model", t->a->cfg.model);
  cJSON_AddBoolToObject(o, "final_pass", t->final_pass);
  cJSON_AddBoolToObject(o, "ok", lrc == 0);
  cJSON_AddNumberToObject(o, "http", (double)st->http_status);
  cJSON_AddStringToObject(o, "finish_reason", st->finish_reason);
  cJSON_AddNumberToObject(o, "t_first_byte", ms2(st->t_first_byte_ms));
  cJSON_AddNumberToObject(o, "t_stream_end", ms2(st->t_end_ms));
  cJSON_AddNumberToObject(o, "prompt_tokens", (double)st->prompt_tokens);
  cJSON_AddNumberToObject(o, "completion_tokens",
                          (double)st->completion_tokens);
  cJSON_AddNumberToObject(o, "cached_tokens", (double)st->cached_tokens);
  cJSON_AddNumberToObject(o, "cache_miss_tokens",
                          (double)st->cache_miss_tokens);
  if (st->cache_write_tokens)
    cJSON_AddNumberToObject(o, "cache_write_tokens",
                            (double)st->cache_write_tokens);
  if (cost >= 0)
    cJSON_AddNumberToObject(o, "cost", cost);
  cJSON_AddNumberToObject(o, "prefix_break", t->prefix_break);
  if (t->prefix_reason)
    cJSON_AddStringToObject(o, "prefix_reason", t->prefix_reason);
  if (t->cont)
    cJSON_AddNumberToObject(o, "continuation", t->cont);
  cJSON_AddNumberToObject(o, "retries", st->retries);
  cJSON_AddBoolToObject(o, "client_stopped", st->client_stopped);
  cJSON_AddNumberToObject(o, "calls", t->ncalls);
  cJSON_AddNumberToObject(o, "text_bytes", (double)t->text.len);
  if (t->ctl) {
    ctl_parser_stats_t cs;
    ctl_parser_stats(t->ctl, &cs);
    cJSON *c = cJSON_AddObjectToObject(o, "ctl");
    cJSON_AddNumberToObject(c, "calls_open", (double)cs.calls_open);
    cJSON_AddNumberToObject(c, "calls_ready", (double)cs.calls_ready);
    cJSON_AddNumberToObject(c, "args_nbytes", (double)cs.args_nbytes);
    cJSON_AddNumberToObject(c, "args_inline", (double)cs.args_inline);
    cJSON_AddNumberToObject(c, "args_heredoc", (double)cs.args_heredoc);
    cJSON_AddNumberToObject(c, "pending_peak", (double)cs.pending_peak);
    cJSON_AddStringToObject(c, "error_kind",
                            ctl_error_kind_name(cs.error_kind));
  }
  if (t->parse_error)
    cJSON_AddStringToObject(o, "parse_error", t->perr);
  if (lrc != 0 && st->error[0])
    cJSON_AddStringToObject(o, "error", st->error);
  cst_metrics_emit(t->a->metrics, o);
}

static void emit_turn_metrics(cst_agent_t *a, int steps, int rc,
                              double total_ms, const char *err) {
  cJSON *o = cJSON_CreateObject();
  cJSON_AddStringToObject(o, "type", "turn");
  cJSON_AddNumberToObject(o, "turn", a->turn_no);
  cJSON_AddStringToObject(o, "protocol", cst_protocol_name(a->cfg.protocol));
  cJSON_AddStringToObject(o, "model", a->cfg.model);
  cJSON_AddNumberToObject(o, "steps", steps);
  cJSON_AddNumberToObject(o, "rc", rc);
  cJSON_AddNumberToObject(o, "t_total", ms2(total_ms));
  if (err && err[0])
    cJSON_AddStringToObject(o, "error", err);
  cst_metrics_emit(a->metrics, o);
}

/* DeepSeek can resume a reply that was cut off (max tokens, dropped stream)
 * from its own text via prefix completion, instead of a fresh step. */
static int can_continue(const cst_agent_t *a, const turn_t *t,
                        const cst_llm_stats_t *st, int lrc,
                        cst_protocol_t proto) {
  if (!cst_protocol_is_native(proto) || !a->caps.prefix_completion ||
      a->prefix_off || a->cancel || t->stop || t->cont >= 2 || t->ncalls ||
      t->native_opened || t->in_think || !t->raw.len)
    return 0;
  if (lrc == 0)
    return strcmp(st->finish_reason, "length") == 0;
  return st->http_status > 0 && st->http_status < 400 && !st->client_stopped;
}

/* 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)
    return atol(e);
  long win = cst_model_context_tokens(a->cfg.model);
  if (win <= 0)
    return 0;
  long b = win / 4 * 3;
  return b > 262144 ? 262144 : b;
}

/* Compact before a turn when the last request crossed the budget. */
static void maybe_auto_compact(cst_agent_t *a, const cst_agent_hooks_t *hooks) {
  const char *e = getenv("CST_AUTO_COMPACT");
  if (e && e[0] == '0')
    return;
  long budget = ctx_budget(a);
  if (a->last_prompt_tokens <= 0 || budget <= 0 ||
      a->last_prompt_tokens < budget)
    return;
  if (hooks && hooks->on_status) {
    char msg[128];
    snprintf(msg, sizeof(msg),
             "prompt %ldK tokens (budget %ldK) — compacting...",
             a->last_prompt_tokens / 1000, budget / 1000);
    hooks->on_status(msg, hooks->userdata);
  }
  char err[256];
  if (cst_agent_compact(a, err, sizeof(err)) == 0) {
    a->last_prompt_tokens = 0;
    if (hooks && hooks->on_status)
      hooks->on_status("context compacted", hooks->userdata);
  }
}

int cst_agent_run_turn_ex(cst_agent_t *a, const char *user_text,
                          const cst_agent_hooks_t *hooks, char *err,
                          size_t err_sz) {
  char dummy[8];
  if (!err || !err_sz) {
    err = dummy;
    err_sz = sizeof(dummy);
  }
  err[0] = 0;
  if (!a || !user_text)
    return -1;

  a->cancel = 0;
  maybe_auto_compact(a, hooks);
  a->turn_no++;
  a->nreads = 0;
  a->nfails = 0;
  a->nwrites = 0;
  {
    time_t now = time(NULL);
    struct tm tmv;
    os_localtime(&now, &tmv);
    char ts[32];
    strftime(ts, sizeof(ts), "%Y%m%d-%H%M%S", &tmv);
    snprintf(a->undo_tag, sizeof(a->undo_tag), "%s-%d-%03d", ts, os_getpid(),
             a->turn_no % 1000);
    if (a->cfg.undo_tag[0])
      snprintf(a->undo_tag, sizeof(a->undo_tag), "%s", a->cfg.undo_tag);
  }
  os_set_cancel_flag(&a->cancel);
  const cst_protocol_t proto = a->cfg.protocol;
  const double turn_t0 = os_now_ms();
  memset(&a->turn_usage, 0, sizeof(a->turn_usage));

  ensure_system(a);
  sync_notes(a);
  append_msg_full(a, CST_ROLE_USER, user_text, NULL, NULL, NULL,
                  meta_json(NULL, proto), a->pending, a->npending, NULL);
  cst_agent_pending_clear(a); /* the images now live in the session */

  int empty_retries = 0, ctl_retries = 0, tools_ran = 0;
  const char *nudge = NULL;
  int rc = -1;
  int step = 0;

  for (step = 0; step < a->cfg.max_steps; step++) {
    turn_t t;
    memset(&t, 0, sizeof(t));
    t.a = a;
    t.hooks = hooks;
    t.proto = proto;
    t.step = step;
    t.final_pass = tools_ran && step == a->cfg.max_steps - 1;
    if (t.final_pass)
      nudge = NUDGE_FINAL;
    if (cst_protocol_is_ctl(proto))
      t.ctl = ctl_parser_create(on_ctl, &t);
    else
      t.mm = mm_scan_create(on_mm, &t);
    if (!t.ctl && !t.mm) {
      snprintf(err, err_sz, "parser create failed");
      break;
    }

    if (step > 0)
      sync_notes(a);
    req_msgs_t rm;
    if (build_request_msgs(a, proto, nudge, &rm) != 0) {
      turn_free(&t);
      snprintf(err, err_sz, "oom");
      break;
    }
    nudge = NULL;
    {
      char s[64];
      snprintf(s, sizeof(s), "calling model (step %d)...", step + 1);
      hook_status(&t, t.final_pass ? "final answer pass..." : s);
    }

    /* The final pass keeps tools and tool_choice as they were: changing
     * either would miss the cache for the whole prompt. Calls it still
     * makes are dropped below. */
    cst_llm_stats_t st;
    cst_llm_request_t req;
    fill_request(a, &req, &rm, proto);
    req.on_delta = on_delta;
    req.userdata = &t;
    req.stop_flag = &t.stop;
    req.stats = &st;
    if (cst_protocol_is_native(proto)) {
      req.on_native_tool = on_native_tool;
      req.on_native_open = on_native_open;
    }
    check_prefix(a, &t, &rm, req.tools_json);

    t.t0 = os_now_ms();
    char lerr[512];
    int lrc = cst_llm_chat_stream(&req, lerr, sizeof(lerr));
    if (lrc != 0 && st.http_status == 400 &&
        req.cache_style != CST_CACHE_AUTO && !t.raw.len && !t.ncalls &&
        !a->cancel) {
      a->hints_off = 1; /* the gateway rejects cache hints */
      req.cache_style = CST_CACHE_AUTO;
      req.cache_key = NULL;
      req.cache_ttl = NULL;
      lrc = cst_llm_chat_stream(&req, lerr, sizeof(lerr));
    }
    if (lrc != 0 && st.http_status == 400 && !req.images_off &&
        msgs_have_images(&rm) && !t.raw.len && !t.ncalls && !a->cancel) {
      a->images_off = 1; /* unknown model turned out not to take images */
      snprintf(a->images_off_model, sizeof(a->images_off_model), "%s",
               a->cfg.model);
      req.images_off = 1;
      if (req.tools_json)
        req.tools_json = agent_tools_json(a);
      hook_status(&t, "the model refused images — sending notes instead...");
      lrc = cst_llm_chat_stream(&req, lerr, sizeof(lerr));
    }
    const int lrc0 = lrc;
    while (can_continue(a, &t, &st, lrc, proto)) {
      emit_step_metrics(&t, &st, lrc, account(a, &st));
      if (st.prompt_tokens > 0)
        a->last_prompt_tokens = st.prompt_tokens;
      t.cont++;
      t.prefix_break = -1;
      t.prefix_reason = NULL;
      cst_buf_t pre = {0};
      cst_buf_append(&pre, t.raw.buf ? t.raw.buf : "", t.raw.len);
      cst_buf_append(&pre, t.tagbuf, t.taglen);
      push_prefix(&rm, pre.buf);
      cst_buf_free(&pre);
      if (a->caps.reasoning_passback && t.think.len)
        rm.msgs[rm.n - 1].reasoning = t.think.buf;
      char beta[1024];
      beta_base(a->cfg.api_base, beta, sizeof(beta));
      req.api_base = beta;
      req.msgs = rm.msgs;
      req.nmsgs = rm.n;
      hook_status(&t, "output cut off — continuing where it stopped...");
      lrc = cst_llm_chat_stream(&req, lerr, sizeof(lerr));
      rm.n--;
      rm.ntail--;
      req.api_base = a->cfg.api_base;
      if (lrc != 0 && st.http_status >= 400) {
        if (st.http_status == 400 || st.http_status == 404)
          a->prefix_off = 1; /* no prefix completion behind this base */
        lrc = lrc0;
        break;
      }
    }
    req_msgs_free(&rm);

    /* flush a dangling partial tag */
    if (t.taglen && !t.in_think)
      feed_visible(&t, t.tagbuf, t.taglen);
    t.taglen = 0;
    if (t.mm)
      mm_scan_finish(t.mm);
    if (t.ctl && !t.stop)
      ctl_parser_finish(t.ctl); /* EOF may confirm a final call */
    if (proto == CST_PROTO_XML && t.ncalls == 0 && !t.parse_error &&
        t.raw.buf &&
        (strstr(t.raw.buf, "<minimax:tool_call") ||
         strstr(t.raw.buf, "<invoke"))) {
      t.parse_error = 1;
      snprintf(t.perr, sizeof(t.perr),
               "malformed tool block, nothing could be parsed");
    }

    emit_step_metrics(&t, &st, lrc, account(a, &st));
    if (st.prompt_tokens > 0)
      a->last_prompt_tokens = st.prompt_tokens;

    if (lrc != 0) {
      /* side effects of eagerly executed calls must still be recorded */
      int kept = 0;
      for (int i = 0; i < t.ncalls; i++) {
        if (t.calls[i].done) {
          t.calls[kept++] = t.calls[i];
        } else {
          cst_tool_call_clear(&t.calls[i].call);
          free(t.calls[i].raw_args);
          free(t.calls[i].output);
          cst_attachments_free(t.calls[i].attach, t.calls[i].nattach);
        }
      }
      t.ncalls = kept;
      if (t.ncalls > 0)
        persist_calls(&t);
      turn_free(&t);
      snprintf(err, err_sz, "llm: %s", lerr);
      break;
    }

    exec_pending(&t); /* eagerly executed calls are already done */
    if (t.ncalls > 0 && a->cfg.tool_profile != CST_TOOLS_MINIMAL &&
        a->cfg.max_steps - (step + 1) == 3)
      append_output(&t.calls[t.ncalls - 1],
                    "\n[cst: 3 model calls left in this turn. Finish the "
                    "essential part now, verify it, and then summarize what "
                    "is done and what remains.]");
    if (a->cancel) {
      if (t.ncalls > 0)
        persist_calls(&t);
      else if (t.text.len)
        append_msg(a, CST_ROLE_ASSISTANT, trim_left(t.text.buf), NULL, NULL,
                   NULL, meta_json("cancelled", proto));
      turn_free(&t);
      snprintf(err, err_sz, "cancelled");
      break;
    }

    const char *body = t.text.buf ? trim_left(t.text.buf) : "";
    int has_text = text_is_meaningful(body, strlen(body));

    if (t.ncalls > 0) {
      persist_calls(&t);
      tools_ran = 1;
      empty_retries = 0;
      if (t.parse_error && cst_protocol_is_text(proto)) {
        char msg[768];
        snprintf(msg, sizeof(msg),
                 "[%s] contract error after your last confirmed call: %s. "
                 "Frames after it were ignored; re-send them with exact "
                 "syntax.",
                 proto == CST_PROTO_XML ? "xml" : "ctl", t.perr);
        append_msg(a, CST_ROLE_USER, msg, NULL, NULL, NULL,
                   meta_json("ctl_error", proto));
      }
      turn_free(&t);
      hook_status(&t, "tool round done...");
      continue;
    }

    if (t.parse_error && cst_protocol_is_text(proto)) {
      if (t.raw.len)
        append_msg(a, CST_ROLE_ASSISTANT, t.raw.buf, NULL, NULL, NULL,
                   meta_json(NULL, proto));
      char msg[768];
      if (proto == CST_PROTO_XML)
        snprintf(msg, sizeof(msg),
                 "[xml] contract error: %s. Nothing was executed. Re-send the "
                 "tool block with exact syntax:\n<minimax:tool_call>\n"
                 "<invoke name=\"TOOL\">\n<parameter name=\"KEY\">VALUE"
                 "</parameter>\n</invoke>\n</minimax:tool_call>",
                 t.perr);
      else
        snprintf(msg, sizeof(msg),
                 "[ctl] contract error: %s. Nothing was executed. Re-send the "
                 "call with exact CTL syntax (byte counts must be exact; use "
                 "inline values for short single-line arguments).",
                 t.perr);
      append_msg(a, CST_ROLE_USER, msg, NULL, NULL, NULL,
                 meta_json("ctl_error", proto));
      turn_free(&t);
      if (++ctl_retries > MAX_CTL_RETRIES) {
        snprintf(err, err_sz, "%s: repeated contract errors",
                 cst_protocol_name(proto));
        break;
      }
      hook_status(&t, "contract error — asking model to re-send...");
      continue;
    }

    if (has_text) {
      append_reply(&t, body, NULL, meta_json(NULL, proto));
      turn_free(&t);
      hook_status(&t, "done");
      rc = 0;
      break;
    }
    turn_free(&t);

    if (empty_retries < 2 && step + 1 < a->cfg.max_steps) {
      empty_retries++;
      nudge = NUDGE_EMPTY;
      hook_status(&t, "empty response — retrying...");
      continue;
    }
    /* Last resort: surface the latest tool output so the turn isn't blank. */
    if (tools_ran) {
      int n = cst_session_count(a->session);
      for (int i = n - 1; i >= 0; i--) {
        const cst_message_t *m = cst_session_get(a->session, i);
        if (!m->tool_call_id[0] || !m->content)
          continue;
        cst_buf_t b = {0};
        cst_buf_puts(&b, "（模型未返回正文，以下为最近工具输出）\n");
        cst_buf_append(&b, m->content, cst_utf8_floor(m->content,
                                                     strlen(m->content) > 2000
                                                         ? 2000
                                                         : strlen(m->content)));
        append_msg(a, CST_ROLE_ASSISTANT, b.buf, NULL, NULL, NULL,
                   meta_json("fallback", proto));
        cst_buf_free(&b);
        rc = 0;
        break;
      }
      if (rc == 0)
        break;
    }
    snprintf(err, err_sz, "empty model response");
    break;
  }
  if (rc != 0 && !err[0])
    snprintf(err, err_sz, "step limit (%d) reached", a->cfg.max_steps);
  emit_turn_metrics(a, step + (step < a->cfg.max_steps ? 1 : 0), rc,
                    os_now_ms() - turn_t0, err);
  return rc;
}

static int agent_reopen_session(cst_agent_t *a, const char *path) {
  cst_session_t *ns = cst_session_open(path);
  if (!ns)
    return -1;
  cst_session_close(a->session);
  a->session = ns;
  cst_agent_pending_clear(a); /* images belong to the session that made them */
  snprintf(a->cfg.session_path, sizeof(a->cfg.session_path), "%s", path);
  cst_metrics_close(a->metrics);
  a->metrics = cst_metrics_open(a->cfg.workspace_root, path);
  reset_prefix_state(a);
  a->nprev = 0;
  a->prefix_off = a->hints_off = a->images_off = 0;
  memset(&a->usage, 0, sizeof(a->usage));
  memset(&a->turn_usage, 0, sizeof(a->turn_usage));
  ensure_system(a);
  restore_todo(a);
  estimate_tokens(a);
  return 0;
}

int cst_agent_new_session(cst_agent_t *a, char *id_out, size_t id_sz) {
  if (!a)
    return -1;
  if (cst_agent_store_ensure(a->cfg.workspace_root) != 0)
    return -1;
  char id[128], path[4096];
  /* ids are per second: never reopen (and append to) one that exists */
  for (int tries = 0;; tries++) {
    if (tries == 16 || cst_agent_store_new_id(id, sizeof(id)) != 0 ||
        cst_agent_store_path_for_id(a->cfg.workspace_root, id, path,
                                    sizeof(path)) != 0)
      return -1;
    FILE *fp = fopen(path, "rb");
    if (!fp)
      break;
    fclose(fp);
  }
  if (agent_reopen_session(a, path) != 0)
    return -1;
  cst_agent_store_set_current(a->cfg.workspace_root, id);
  if (id_out && id_sz)
    snprintf(id_out, id_sz, "%s", id);
  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;
  char path[4096];
  if (cst_agent_store_path_for_id(a->cfg.workspace_root, id, path,
                                  sizeof(path)) != 0)
    return -1;
  if (agent_reopen_session(a, path) != 0)
    return -1;
  char bare[128];
  snprintf(bare, sizeof(bare), "%s", id);
  size_t n = strlen(bare);
  if (n > 6 && strcmp(bare + n - 6, ".jsonl") == 0)
    bare[n - 6] = 0;
  cst_agent_store_set_current(a->cfg.workspace_root, bare);
  return 0;
}

static void acc_feed(const char *data, size_t len, void *ud) {
  cst_buf_append((cst_buf_t *)ud, data, len);
}

#define COMPACT_ASK                                                            \
  "Summarize the coding session so work can continue from the summary "        \
  "alone. Keep: the user's goals, files and paths involved, decisions made, "  \
  "current state (what works, what fails), and open items. Write it in the "   \
  "language the user used. Do not call tools; output only the summary."
#define COMPACT_HEAD "[session compacted — summary of earlier conversation]\n"

/* The summary request reuses the session's own prefix (system prompt, tools,
 * history) and only appends the ask, so it reads almost entirely from the
 * cache. DeepSeek also gets the summary heading as an assistant prefix. */
static int compact_same_prefix(cst_agent_t *a, cst_buf_t *acc, char *lerr,
                               size_t lerr_sz) {
  const cst_protocol_t proto = a->cfg.protocol;
  req_msgs_t rm;
  if (build_request_msgs(a, proto, "[cst] " COMPACT_ASK, &rm) != 0)
    return -1;
  cst_llm_stats_t st;
  cst_llm_request_t req;
  fill_request(a, &req, &rm, proto);
  req.on_delta = acc_feed;
  req.userdata = acc;
  req.stats = &st;
  char beta[1024];
  int prefix = cst_protocol_is_native(proto) && a->caps.prefix_completion &&
               !a->prefix_off;
  if (prefix) {
    push_prefix(&rm, COMPACT_HEAD);
    beta_base(a->cfg.api_base, beta, sizeof(beta));
    req.api_base = beta;
    req.nmsgs = rm.n;
  }
  int rc = cst_llm_chat_stream(&req, lerr, lerr_sz);
  if (rc != 0 && prefix && !a->cancel &&
      (st.http_status == 400 || st.http_status == 404)) {
    a->prefix_off = 1;
    req.nmsgs = --rm.n;
    rm.ntail--;
    req.api_base = a->cfg.api_base;
    acc->len = 0;
    if (acc->buf)
      acc->buf[0] = 0;
    rc = cst_llm_chat_stream(&req, lerr, lerr_sz);
  }
  account(a, &st);
  req_msgs_free(&rm);
  return rc;
}

/* Fallback: a standalone request carrying a condensed transcript. */
static int compact_blob(cst_agent_t *a, cst_buf_t *acc, char *lerr,
                        size_t lerr_sz) {
  int n = cst_session_count(a->session);
  cst_buf_t blob = {0};
  cst_buf_puts(&blob, COMPACT_ASK "\n\n---\n");
  for (int i = 0; i < n && blob.len < 120000; i++) {
    const cst_message_t *m = cst_session_get(a->session, i);
    if (!m || m->role == CST_ROLE_SYSTEM)
      continue;
    const char *role = m->role == CST_ROLE_USER
                           ? (m->tool_call_id[0] ? "tool" : "user")
                       : m->role == CST_ROLE_ASSISTANT ? "assistant"
                                                       : "tool";
    const char *c = m->content ? m->content : "";
    size_t cl = strlen(c);
    if (cl > 2000)
      cl = cst_utf8_floor(c, 2000);
    cst_buf_printf(&blob, "\n[%s]\n", role);
    cst_buf_append(&blob, c, cl);
  }

  cst_message_t reqs[2];
  memset(reqs, 0, sizeof(reqs));
  reqs[0].role = CST_ROLE_SYSTEM;
  reqs[0].content = (char *)"You summarize coding-agent sessions. No tools.";
  reqs[1].role = CST_ROLE_USER;
  reqs[1].content = blob.buf;

  cst_llm_stats_t st;
  cst_llm_request_t req;
  memset(&req, 0, sizeof(req));
  req.api_base = a->cfg.api_base;
  req.api_key = a->cfg.api_key;
  req.model = a->cfg.model;
  req.reasoning_effort =
      cst_effort_wire(a->cfg.provider, a->cfg.model, a->cfg.effort);
  req.msgs = reqs;
  req.nmsgs = 2;
  req.on_delta = acc_feed;
  req.userdata = acc;
  req.cancel_flag = &a->cancel;
  req.stats = &st;
  int rc = cst_llm_chat_stream(&req, lerr, lerr_sz);
  cst_buf_free(&blob);
  account(a, &st);
  return rc;
}

/* Strip <think>...</think> blocks and surrounding blanks; returns the
 * length left. */
static size_t clean_summary(cst_buf_t *acc) {
  if (!acc->buf)
    return 0;
  char *src = acc->buf, *dst = acc->buf;
  while (*src) {
    if (strncmp(src, "<think>", 7) == 0) {
      char *end = strstr(src, "</think>");
      if (end) {
        src = end + 8;
        continue;
      }
    }
    *dst++ = *src++;
  }
  *dst = 0;
  acc->len = (size_t)(dst - acc->buf);
  while (acc->len > 0 &&
         (acc->buf[acc->len - 1] == '\n' || acc->buf[acc->len - 1] == ' '))
    acc->buf[--acc->len] = 0;
  return acc->len;
}

int cst_agent_compact(cst_agent_t *a, char *err, size_t err_sz) {
  char dummy[8];
  if (!err || !err_sz) {
    err = dummy;
    err_sz = sizeof(dummy);
  }
  err[0] = 0;
  if (!a || !a->session)
    return -1;

  int n = cst_session_count(a->session);
  if (n <= 1) {
    snprintf(err, err_sz, "nothing to compact");
    return -1;
  }

  cst_buf_t acc = {0};
  char lerr[512];
  lerr[0] = 0;
  int rc = compact_same_prefix(a, &acc, lerr, sizeof(lerr));
  if ((rc != 0 || clean_summary(&acc) < 8) && !a->cancel) {
    cst_buf_free(&acc);
    lerr[0] = 0;
    rc = compact_blob(a, &acc, lerr, sizeof(lerr));
    if (rc == 0)
      clean_summary(&acc);
  }
  if (rc != 0 || acc.len == 0) {
    cst_buf_free(&acc);
    snprintf(err, err_sz, "compact llm: %s", lerr[0] ? lerr : "empty");
    return -1;
  }
  if (acc.len < 8) {
    cst_buf_free(&acc);
    snprintf(err, err_sz, "compact produced empty summary");
    return -1;
  }

  const char *body = trim_left(acc.buf);
  if (strncmp(body, COMPACT_HEAD, strlen(COMPACT_HEAD) - 1) == 0)
    body += strlen(COMPACT_HEAD) - 1;
  cst_buf_t summary = {0};
  cst_buf_printf(&summary, COMPACT_HEAD "%s", trim_left(body));
  cst_buf_free(&acc);

  pstate_t st;
  cst_message_t keep[2];
  memset(keep, 0, sizeof(keep));
  keep[0].role = CST_ROLE_SYSTEM;
  keep[0].content = frozen_prompt(a, &st);
  keep[0].meta_json = system_meta(a, &st);
  keep[1].role = CST_ROLE_USER;
  keep[1].content = summary.buf;
  rc = cst_session_rewrite(a->session, keep, 2);
  free(keep[0].content);
  free(keep[0].meta_json);
  cst_buf_free(&summary);
  if (rc != 0) {
    snprintf(err, err_sz, "rewrite session failed");
    return -1;
  }
  reset_prefix_state(a);
  return 0;
}