## openai/codex-plugin-cc — v1.0.3…v1.0.4

_6 commits._

### Fixes
- **Plugin version bumped to 1.0.4** — Updated the Codex plugin version from 1.0.3 to 1.0.4 across all configuration files, including package metadata and marketplace listings. (807e03a)
- **Fix rescue command routing recursion** — Fixed an infinite loop in the rescue command by changing how it routes requests to the Codex rescue subagent. Instead of using ambiguous prose that caused the system to call itself recursively, the command now explicitly uses the Agent tool with proper routing and runs inline rather than in a forked context, ensuring the Agent tool remains available. (bb38412)
- **Secure job ID argument handling** — Fixed a security vulnerability where unquoted job IDs in cancel, result, and status commands could allow shell metacharacters to be executed. Job ID arguments are now properly quoted to prevent command injection attacks. (6a5c2ba)
- **Fixed invalid reasoning effort setting** — Corrected an unsupported 'xhigh' value to 'high' in the README configuration example to prevent users from encountering errors when applying the documented settings. (8e9a38c)
- **Specify model for rescue agent** — The codex-rescue agent now explicitly declares which AI model it uses (Sonnet), ensuring predictable performance and cost. Previously it relied on a default assignment, which could vary. (b115623)

### Backend
- **Fix session runtime with --cwd flag** — The setup and status commands now correctly read session runtime information when using the --cwd flag to specify a different working directory, ensuring they look in the right location for shared session data regardless of where the command is invoked from. (c24afe8)

_Recap by [Repo Wrapped](https://repowrapped.com/gh/openai/codex-plugin-cc?utm_source=github-action)._