The Agent Skills Directory

[COMMAND_EXECUTION]: The skill coordinates a multi-round debate by executing shell commands, including running the Codex CLI via node and managing discussion transcripts using a local Python script (scripts/discuss.py). It also uses standard utilities like git, mkdir, and cp for repository analysis and file management.
[EXTERNAL_DOWNLOADS]: The skill documentation lists @openai/codex as a requirement, which is a well-known package from the npm registry. Installation instructions also refer to the skills CLI and the author's public GitHub repository.
[COMMAND_EXECUTION]: The helper script discuss.py reads the user's ~/.claude/plugins/installed_plugins.json file. This is performed to programmatically locate the installation path of the Codex plugin required for the skill's functionality.
[PROMPT_INJECTION]: The skill includes an inherent indirect prompt injection surface as it ingests untrusted repository data (such as PR diffs or design specs) for analysis. However, it implements a 'conclusion protocol' requiring evidence-based claims and manual user sign-off to mitigate risks of model misbehavior.

cc-codex-discussion