The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various shell commands using the Bash tool, including git (for diffing and logging), gh (to fetch PR metadata), and codex (to perform the external AI review). These are standard operations for a developer-oriented code review tool.
[DATA_EXFILTRATION]: The skill transmits source code diffs to an external service via the codex CLI. This is the primary intended function of the skill (providing an external peer review) and utilizes a well-known developer tool.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data in the form of code diffs and repository history. It provides structured instructions to the external reviewer and includes a manual triage step ('Iron Rule: Do not blindly trust Codex findings') to mitigate risks from inaccurate or malicious model outputs.

peer-review