The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple local shell commands including codex, git, jq, pnpm, make, awk, and sed to manage the lifecycle of the self-review loop.- [REMOTE_CODE_EXECUTION]: The skill utilizes the OpenAI Codex service to process repository contents and generate code fixes. Codebase diffs and context are sent to the external API for analysis.- [DYNAMIC_EXECUTION]: The skill automatically modifies the local codebase based on output from an external model. A 'HARD-GATE' policy explicitly restricts the agent from reasoning about these fixes, requiring them to be applied 'literally' and 'mechanically', which bypasses safety filters and could result in the application of malicious code.- [INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests untrusted data from the repository's source files and git diff output. 1. Ingestion points: Reads git diff, git log, and repository source files. 2. Boundary markers: Uses  sentinels for output parsing. 3. Capability inventory: Includes file-writing (apply_minimal_fix_from_mitigation), shell execution (pnpm test), and version control operations. 4. Sanitization: None; the skill explicitly forbids the agent from questioning the generated mitigations, potentially allowing malicious code to be committed if the codebase content influences Codex output.- [EXTERNAL_DOWNLOADS]: The skill instructs the user to install the @openai/codex package from the npm registry, which is a well-known service.

self-review