Hermit/tests/test_plugins.c

/* Plugins: plugin.json exec tools (stdin JSON, env, timeout, perm), skills
 * and MCP entries from plugins, and workspace trust. */
#define _POSIX_C_SOURCE 200809L

#include "cst_agent.h"
#include "cst_ext.h"
#include "cst_mcp.h"
#include "cst_plugins.h"
#include "cst_skills.h"
#include "cst_tools.h"

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/stat.h>
#include <unistd.h>
#include "cst_test_util.h"

static int nfail;

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

static char g_root[256];

static void put(const char *rel, const char *text) {
  char p[512];
  snprintf(p, sizeof(p), "%s/%s", g_root, rel);
  for (char *q = p + strlen(g_root) + 1; *q; q++) {
    if (*q == '/') {
      *q = 0;
      test_mkdir(p, 0755);
      *q = '/';
    }
  }
  FILE *fp = fopen(p, "wb");
  if (fp) {
    fputs(text, fp);
    fclose(fp);
  }
}

static void run(const char *ws, const char *tool, const char *k1,
                const char *v1, const char *k2, const char *v2,
                cst_tool_result_t *r) {
  ctl_tool_call_t c;
  memset(&c, 0, sizeof(c));
  snprintf(c.name, sizeof(c.name), "%s", tool);
  if (k1) {
    snprintf(c.args[c.nargs].key, sizeof(c.args[0].key), "%s", k1);
    c.args[c.nargs].value = (char *)v1;
    c.args[c.nargs].value_len = strlen(v1);
    c.nargs++;
  }
  if (k2) {
    snprintf(c.args[c.nargs].key, sizeof(c.args[0].key), "%s", k2);
    c.args[c.nargs].value = (char *)v2;
    c.args[c.nargs].value_len = strlen(v2);
    c.nargs++;
  }
  cst_tool_ctx_t ctx;
  memset(&ctx, 0, sizeof(ctx));
  ctx.workspace_root = ws;
  memset(r, 0, sizeof(*r));
  cst_tool_run(&ctx, &c, r);
}

int main(void) {
  if (test_tmpdir(g_root, sizeof(g_root), "cst-plugins") != 0) {
    perror("tmpdir");
    return 1;
  }
  char ws[300], cfgdir[300];
  snprintf(ws, sizeof(ws), "%s/ws", g_root);
  snprintf(cfgdir, sizeof(cfgdir), "%s/cfg", g_root);
  setenv("XDG_CONFIG_HOME", cfgdir, 1);

  put("cfg/cst-agent/plugins/echo/plugin.json",
      "{\"name\": \"echo\", \"description\": \"Test plugin.\",\n"
      " \"tools\": [\n"
      "  {\"name\": \"echo_args\", \"description\": \"Echo stdin.\",\n"
      "   \"command\": [\"sh\", \"run.sh\", \"--flag\"], \"perm\": \"read\",\n"
      "   \"parameters\": {\"type\": \"object\", \"properties\": {\n"
      "     \"name\": {\"type\": \"string\"}, \"count\": {\"type\": "
      "\"integer\"},\n"
      "     \"tags\": {\"type\": \"array\"}}, \"required\": [\"name\"]}},\n"
      "  {\"name\": \"slow\", \"command\": \"sleep 5\", \"timeout\": 1},\n"
      "  {\"name\": \"missing\", \"command\": [\"cst-no-such-cmd-xyz\"], "
      "\"perm\": \"write\"},\n"
      "  {\"name\": \"piped\", \"command\": \"tr a-z A-Z | tr -d '{}'\", "
      "\"perm\": \"net\"},\n"
      "  {\"name\": \"bad\"}\n"
      " ],\n"
      " \"mcpServers\": {\"plug-srv\": {\"command\": \"true\"}}}\n");
  put("cfg/cst-agent/plugins/echo/run.sh",
      "cat\necho\necho \"arg=$1 dir=$(basename \"$CST_PLUGIN_DIR\") "
      "tool=$CST_TOOL ws=$(basename \"$CST_WORKSPACE\")\"\n");
  put("cfg/cst-agent/plugins/echo/skills/plug-skill/SKILL.md",
      "---\nname: plug-skill\ndescription: Shipped by a plugin.\n---\nP\n");
  put("cfg/cst-agent/plugins/notaplugin/readme.txt", "no plugin.json\n");
  put("ws/.agent/plugins/local/plugin.json",
      "{\"tools\": [{\"name\": \"local_tool\", \"command\": \"echo local\"}]}");
  put("ws/keep.txt", "x\n");

  int untrusted = -1;
  int n = cst_plugins_load(ws, cst_settings_trusted(ws), &untrusted);
  CHECK(n == 1, "one user plugin (workspace not trusted)");
  CHECK(untrusted == 1, "workspace plugin counted as untrusted");
  CHECK(!cst_tool_find("local_tool"), "untrusted workspace plugin not loaded");

  const cst_tool_def_t *d = cst_tool_find("echo_args");
  CHECK(d && d->perm == CST_TP_READ, "perm read");
  d = cst_tool_find("slow");
  CHECK(d && d->perm == CST_TP_SHELL, "perm defaults to shell");
  d = cst_tool_find("missing");
  CHECK(d && d->perm == CST_TP_WRITE, "perm write");
  d = cst_tool_find("piped");
  CHECK(d && d->perm == CST_TP_NET, "perm net");
  CHECK(!cst_tool_find("bad"), "tool without command skipped");

  cst_plugin_info_t info[CST_PLUGINS_MAX];
  int ni = cst_plugins_info(info, CST_PLUGINS_MAX);
  CHECK(ni == 1 && strcmp(info[0].name, "echo") == 0 && info[0].ntools == 4 &&
            info[0].nmcp == 1 && info[0].nskills_root == 1 &&
            strstr(info[0].error, "bad"),
        "plugin info");

  cst_tool_result_t r;
  run(ws, "echo_args", "count", "3", "name", "x y", &r);
  CHECK(r.exit_code == 0 && r.out && strstr(r.out, "\"count\":3") &&
            strstr(r.out, "\"name\":\"x y\""),
        "typed JSON on stdin");
  CHECK(r.out && strstr(r.out, "arg=--flag dir=echo tool=echo_args ws=ws"),
        "argv and CST_* environment");
  cst_tool_result_free(&r);

  run(ws, "echo_args", "count", "3", NULL, NULL, &r);
  CHECK(r.exit_code != 0 && r.out && strstr(r.out, "name"),
        "schema required enforced");
  cst_tool_result_free(&r);

  run(ws, "piped", "name", "abc", NULL, NULL, &r);
  CHECK(r.exit_code == 0 && r.out && strstr(r.out, "\"NAME\":\"ABC\""),
        "string command: whole pipeline reads stdin");
  cst_tool_result_free(&r);

  run(ws, "slow", NULL, NULL, NULL, NULL, &r);
  CHECK(r.exit_code == 124 && r.out && strstr(r.out, "timed out after 1 s"),
        "per-tool timeout");
  cst_tool_result_free(&r);

  run(ws, "missing", NULL, NULL, NULL, NULL, &r);
  CHECK(r.exit_code == 127 && r.out && strstr(r.out, "command not found"),
        "missing command hint");
  cst_tool_result_free(&r);

  cst_skill_t sv[CST_SKILLS_MAX];
  int ns = cst_skills_list(ws, sv, CST_SKILLS_MAX);
  CHECK(ns == 1 && strcmp(sv[0].name, "plug-skill") == 0 &&
            strcmp(sv[0].source, "plugin:echo") == 0,
        "plugin skills root");

  cst_mcp_info_t mi[CST_MCP_MAX];
  int nm = cst_mcp_info(mi, CST_MCP_MAX);
  CHECK(nm == 1 && strcmp(mi[0].name, "plug-srv") == 0 &&
            strcmp(mi[0].source, "plugin:echo") == 0,
        "plugin MCP server config");
  cst_mcp_stop_all();

  cst_plugins_unload();
  CHECK(!cst_tool_find("echo_args"), "unload drops tools");
  CHECK(cst_skills_list(ws, sv, CST_SKILLS_MAX) == 0, "unload drops skills");

  /* trust the workspace -> its plugin loads too */
  CHECK(cst_settings_trust(ws, 1) == 0 && cst_settings_trusted(ws),
        "trust saved");
  n = cst_plugins_load(ws, cst_settings_trusted(ws), &untrusted);
  CHECK(n == 2 && untrusted == 0 && cst_tool_find("local_tool"),
        "trusted workspace plugin loads");
  run(ws, "local_tool", NULL, NULL, NULL, NULL, &r);
  CHECK(r.exit_code == 0 && r.out && strstr(r.out, "local"), "local tool");
  cst_tool_result_free(&r);
  cst_plugins_unload();
  cst_mcp_stop_all();
  CHECK(cst_settings_trust(ws, 0) == 0 && !cst_settings_trusted(ws),
        "untrust saved");

  /* extensions: the minimal profile loads nothing */
  cst_agent_config_t cfg;
  memset(&cfg, 0, sizeof(cfg));
  snprintf(cfg.workspace_root, sizeof(cfg.workspace_root), "%s", ws);
  cfg.tool_profile = CST_TOOLS_MINIMAL;
  char rep[256];
  cst_extensions_load(&cfg, rep, sizeof(rep));
  CHECK(!rep[0] && !cst_tool_find("echo_args"), "minimal: no extensions");
  cfg.tool_profile = CST_TOOLS_DEFAULT;
  cst_extensions_load(&cfg, rep, sizeof(rep));
  CHECK(strstr(rep, "1 skill") && strstr(rep, "1 plugin (4 tools)") &&
            strstr(rep, "untrusted"),
        "extensions report");
  CHECK(cst_tool_find("skill") && cst_tool_find("echo_args"),
        "extensions register tools");
  cst_extensions_unload();
  CHECK(!cst_tool_find("skill") && !cst_tool_find("echo_args"),
        "extensions unload");

  if (test_rmrf(g_root) != 0)
    fprintf(stderr, "note: cleanup failed\n");
  if (nfail) {
    fprintf(stderr, "test_plugins: %d failure(s)\n", nfail);
    return 1;
  }
  printf("test_plugins: OK\n");
  return 0;
}