cc-codex-run 0.1.1 → 0.1.2

package/CHANGELOG.md

@@ -1,5 +1,9 @@
 # Changelog
 
+## 0.1.2 — 2026-11-04
+
+- The `codex` skill calls its script as `node <path>`, so it works when an installer drops the executable bit (cc-switch installs from GitHub did, giving "Permission denied").
+
 ## 0.1.1 — 2026-11-04
 
 - Published to npm as `cc-codex-run`. No code changes.

package/README.md

@@ -34,6 +34,14 @@ npm install -g cc-codex-run
 codex-run --install-skills          # add --force to move existing codex/codex-delegate skills aside
 ```
 
+With the [skills CLI](https://github.com/vercel-labs/skills), into the current project (`.claude/skills/`), or user-wide with `-g`:
+
+```bash
+npx skills@latest add kcng0/cc-codex-run -a claude-code -s '*'
+```
+
+The plugin namespaces its skills (`codex-run:codex`). The other two methods install plain `codex` and `codex-delegate` skills. If you already have a personal skill with one of those names in `~/.claude/skills`, it takes precedence over a project-level install, and a user-level install may replace it.
+
 ## Skills
 
 - **`codex`** teaches Claude how to dispatch, answer, resume, attach to, stop and review with `codex-run`.

package/README.zh-CN.md

@@ -34,6 +34,14 @@ npm install -g cc-codex-run
 codex-run --install-skills          # 已有同名 codex / codex-delegate skill 时，加 --force 会先把它们改名备份
 ```
 
+也可以用 [skills CLI](https://github.com/vercel-labs/skills) 安装到当前项目（`.claude/skills/`），加 `-g` 则装到用户目录：
+
+```bash
+npx skills@latest add kcng0/cc-codex-run -a claude-code -s '*'
+```
+
+插件安装的 skill 带命名空间（`codex-run:codex`）。另外两种方式装的是不带命名空间的 `codex` 和 `codex-delegate`：如果 `~/.claude/skills` 里已有同名的个人 skill，它会优先于项目级安装的 skill 生效；用户级安装则可能把它覆盖掉。
+
 ## Skills
 
 - **`codex`**：教 Claude 用 `codex-run` 派发任务、答复提问、续接、重新跟随、停止和 review。

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "cc-codex-run",
-  "version": "0.1.1",
+  "version": "0.1.2",
   "description": "Claude Code skills and CLI for delegating work to OpenAI Codex through the local Codex app-server daemon",
   "type": "module",
   "bin": {

package/plugin/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "codex-run",
-  "version": "0.1.1",
+  "version": "0.1.2",
   "description": "Delegate tasks, questions and reviews to OpenAI Codex through the local Codex app-server daemon",
   "author": {
     "name": "kcng0"

package/plugin/skills/codex/SKILL.md

@@ -1,12 +1,12 @@
 ---
 name: codex
 description: Run tasks on OpenAI Codex through the local Codex app-server daemon — pick a role or model, run several in the background, answer Codex's questions, resume a thread by id, or run Codex's built-in review. Use whenever you need to call Codex.
-allowed-tools: Bash(${CLAUDE_SKILL_DIR}/scripts/codex-run.mjs *)
+allowed-tools: Bash(node ${CLAUDE_SKILL_DIR}/scripts/codex-run.mjs *)
 ---
 
 # Calling Codex
 
-Script: `${CLAUDE_SKILL_DIR}/scripts/codex-run.mjs` (always call it by this full path; `-h` lists options). Tasks run on the local Codex app-server daemon (started automatically). Thread names start with `[codex-run·<role>]`, and the user can watch or take over them in the Codex app.
+Script: `node ${CLAUDE_SKILL_DIR}/scripts/codex-run.mjs` (always call it with `node` and this full path, since some installers drop the executable bit; `-h` lists options). Tasks run on the local Codex app-server daemon (started automatically). Thread names start with `[codex-run·<role>]`, and the user can watch or take over them in the Codex app.
 
 ## Dispatching
 
@@ -14,7 +14,7 @@ Script: `${CLAUDE_SKILL_DIR}/scripts/codex-run.mjs` (always call it by this full
 2. Run it with Bash `run_in_background: true`:
 
    ```bash
-   ${CLAUDE_SKILL_DIR}/scripts/codex-run.mjs --role <role> -C <workdir> <task file>
+   node ${CLAUDE_SKILL_DIR}/scripts/codex-run.mjs --role <role> -C <workdir> <task file>
    ```
 
    - **Roles** set the responsibilities and the default model, effort and sandbox. `--roles` lists them; roles in `$CODEX_HOME/agents/` shadow the bundled examples. If the task is clearly lighter or harder than the role's default, override with `-m`/`-e` (pick from `--models`). If results are weak, first tighten the task (definition of done, verification), then start a new thread with higher effort.
@@ -65,7 +65,7 @@ Start a new thread, and hand over the needed conclusions in the task file, when:
 When unsure, start a new thread.
 
 ```bash
-${CLAUDE_SKILL_DIR}/scripts/codex-run.mjs --resume <thread> --role <same role> -C <same dir> <follow-up file>
+node ${CLAUDE_SKILL_DIR}/scripts/codex-run.mjs --resume <thread> --role <same role> -C <same dir> <follow-up file>
 ```
 
 Resume by thread id (parallel runs make "last thread" ambiguous). The follow-up file holds only the new instructions.
@@ -75,7 +75,7 @@ A resumed thread keeps its model and effort and does not accept `-m`/`-e`. `--ro
 ## Review
 
 ```bash
-${CLAUDE_SKILL_DIR}/scripts/codex-run.mjs --review [--base <branch> | --commit <sha>] -C <repo> [focus file]
+node ${CLAUDE_SKILL_DIR}/scripts/codex-run.mjs --review [--base <branch> | --commit <sha>] -C <repo> [focus file]
 ```
 
 Runs Codex's built-in read-only review of git changes; uncommitted changes (including untracked files) by default. The optional focus file names the risks to look at. `--review` takes no `--role`, `--resume`, `-s` or `--schema`.

package/plugin/skills/codex/scripts/codex-run.mjs

@@ -12290,7 +12290,7 @@ async function run(a, { role, schema, validate, version }) {
 }
 
 // src/cli.mjs
-var VERSION = true ? "0.1.1" : "dev";
+var VERSION = true ? "0.1.2" : "dev";
 var HELP = `usage: codex-run [--role ROLE] [-m MODEL] [-e EFFORT] [-s SANDBOX] [-t SECONDS] [--allow-subagents]
                  -C DIR [--schema FILE] [--json] TASK_FILE
        codex-run --resume THREAD -C DIR [--role ROLE] [-s SANDBOX] [--schema FILE] FOLLOW_UP_FILE