CST-agent/tests/test_embed.c

/* The Qt app embeds the engine in-process: it creates an agent from a
 * config, registers its own platform tools with owner "app:*" and dispatches
 * them exactly like the model would. No HTTP server, no port, no token and
 * no second process -- this is the embedding path app/ (Qt) relies on, kept
 * runnable without Qt so it guards the C side on every `make check`. */

#define _POSIX_C_SOURCE 200809L

#include "cJSON.h"
#include "cst_agent.h"
#include "cst_test_util.h"
#include "cst_tools.h"

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

static int nfail;

#define EXPECT(cond, msg)                                                      \
  do {                                                                         \
    if (!(cond)) {                                                             \
      fprintf(stderr, "FAIL %s:%d: %s\n", __FILE__, __LINE__, msg);            \
      nfail++;                                                                 \
    }                                                                          \
  } while (0)

/* What a platform tool implemented by the front end looks like on the C
 * side: a callback plus whatever the front end needs to keep. */
struct app_ctx {
  int calls;
  char last[128];
};

static int app_notify_run(const cst_tool_ctx_t *ctx, const ctl_tool_call_t *c,
                          cst_tool_result_t *r, void *ud) {
  (void)ctx;
  struct app_ctx *a = ud;
  const char *msg = NULL;
  for (int i = 0; i < c->nargs; i++)
    if (strcmp(c->args[i].key, "message") == 0)
      msg = c->args[i].value;
  a->calls++;
  snprintf(a->last, sizeof(a->last), "%s", msg ? msg : "");
  char buf[256];
  snprintf(buf, sizeof(buf), "notified: %s", msg ? msg : "");
  r->out = strdup(buf);
  r->out_len = r->out ? strlen(r->out) : 0;
  r->exit_code = 0;
  return 0;
}

static void dispatch_notify(const cst_tool_ctx_t *ctx,
                            const ctl_tool_call_t *call, cst_tool_result_t *r,
                            struct app_ctx *actx) {
  EXPECT(cst_tool_run(ctx, call, r) == 0, "the app tool runs");
  EXPECT(r->out && strstr(r->out, "notified: hi") != NULL,
         "its output is the tool result");
  EXPECT(actx->calls == 1 && strcmp(actx->last, "hi") == 0,
         "the C callback saw the argument");
  cst_tool_result_free(r);
}

int main(void) {
  char root[1024];
  if (test_tmpdir(root, sizeof(root), "cst-embed") != 0) {
    fprintf(stderr, "cannot create a scratch dir\n");
    return 1;
  }

  cst_agent_config_t cfg;
  memset(&cfg, 0, sizeof(cfg));
  snprintf(cfg.workspace_root, sizeof(cfg.workspace_root), "%s", root);
  snprintf(cfg.session_path, sizeof(cfg.session_path), "%s/s.jsonl", root);
  snprintf(cfg.provider, sizeof(cfg.provider), "deepseek");
  snprintf(cfg.model, sizeof(cfg.model), "deepseek-flash");
  snprintf(cfg.api_base, sizeof(cfg.api_base), "https://api.deepseek.com/v1");
  snprintf(cfg.api_key, sizeof(cfg.api_key), "test-key");
  snprintf(cfg.mode, sizeof(cfg.mode), "general");
  cfg.max_steps = 1;
  cfg.headless = 1;

  cst_agent_t *a = cst_agent_create(&cfg);
  EXPECT(a != NULL, "the app can create an engine in-process");
  if (!a) {
    test_rmrf(root);
    return 1;
  }
  EXPECT(cst_agent_session_path(a) != NULL, "the agent owns its session");

  /* the front end registers its own tool: android-only, policy "notify" */
  static const char *const ANDROID_ONLY[] = {"android", NULL};
  struct app_ctx actx;
  memset(&actx, 0, sizeof(actx));
  cst_tool_def_t def;
  memset(&def, 0, sizeof(def));
  def.name = "app_notify";
  def.desc = "Send a system notification (provided by the app).";
  def.perm = CST_TP_ASK;
  def.perm_name = "notify";
  def.platforms = ANDROID_ONLY;
  def.run_ex = app_notify_run;
  def.ud = &actx;
  def.owner = "app:test";
  EXPECT(cst_tool_register(&def) == 0, "the app registers a tool in-process");

  const cst_tool_def_t *t = cst_tool_find("app_notify");
  EXPECT(t != NULL, "the registry knows it");
  if (!t) {
    cst_agent_destroy(a);
    test_rmrf(root);
    return 1;
  }
  EXPECT(strcmp(cst_tool_perm_name(t), "notify") == 0,
         "its policy name survives registration");
  EXPECT(t->run_ex == app_notify_run && t->ud == &actx,
         "so does the callback and its userdata");

  /* on a desktop host it must not be offered (the probe is the only gate) */
  setenv("CST_PLATFORM", "linux", 1);
  cst_tool_meta_reload();
  char why[160];
  EXPECT(cst_tool_available(t, why, sizeof(why)) != 0,
         "an android-only tool is not available on this host");
  EXPECT(!cst_tool_active(cst_agent_config(a), t),
         "the agent's config does not offer it");

  /* ... and once the host is the phone it is, through the same config */
  setenv("CST_PLATFORM", "android", 1);
  cst_tool_meta_reload();
  EXPECT(cst_tool_available(t, why, sizeof(why)) == 0, "available on android");
  EXPECT(cst_tool_active(cst_agent_config(a), t), "and offered there");
  char *names = cst_tools_names(cst_agent_config(a));
  EXPECT(names && strstr(names, "app_notify") != NULL,
         "it is part of the active tool set");
  free(names);
  char *prompt = cst_tools_prompt(cst_agent_config(a));
  EXPECT(prompt && strstr(prompt, "app_notify") != NULL,
         "and the prompt describes it like any built-in");
  free(prompt);
  void *schema = cst_tools_schema_cfg(cst_agent_config(a));
  char *sj = schema ? cJSON_PrintUnformatted(schema) : NULL;
  EXPECT(sj && strstr(sj, "app_notify") != NULL, "tools[] carries it too");
  free(sj);
  if (schema)
    cJSON_Delete(schema);

  /* dispatch: the front end's callback is what runs */
  ctl_tool_call_t call;
  memset(&call, 0, sizeof(call));
  snprintf(call.name, sizeof(call.name), "app_notify");
  snprintf(call.args[0].key, sizeof(call.args[0].key), "message");
  call.args[0].value = strdup("hi");
  call.args[0].value_len = strlen("hi");
  call.nargs = 1;
  cst_tool_result_t r;
  cst_tool_ctx_t ctx;
  memset(&ctx, 0, sizeof(ctx));
  ctx.workspace_root = root;
  ctx.agent = a;
  dispatch_notify(&ctx, &call, &r, &actx);
  free(call.args[0].value);

  /* a name that is already taken by a tool that *can* run here is refused */
  cst_tool_def_t dup;
  memset(&dup, 0, sizeof(dup));
  dup.name = "read_file";
  dup.desc = "not allowed to shadow a working built-in";
  dup.perm = CST_TP_READ;
  dup.run_ex = app_notify_run;
  dup.owner = "app:test";
  EXPECT(cst_tool_find("read_file")->owner == NULL, "read_file is a built-in");
  EXPECT(cst_tool_register(&dup) != 0,
         "a working built-in cannot be shadowed");

  /* dropping the app's tools leaves the built-in registry untouched */
  EXPECT(cst_tool_unregister_owner("app:test") == 1, "the app unregisters");
  EXPECT(cst_tool_find("app_notify") == NULL, "its tool is gone");
  EXPECT(cst_tool_find("read_file") != NULL, "the built-ins are still there");

  cst_agent_destroy(a);
  test_rmrf(root);
  if (nfail) {
    fprintf(stderr, "test_embed: %d failed\n", nfail);
    return 1;
  }
  printf("test_embed: OK\n");
  return 0;
}