The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local commands such as git, gh, and codex to perform diffing and analysis. It uses security flags like -a never and -s read-only to ensure the code review process cannot modify files and runs within a restricted environment.
[PROMPT_INJECTION]: The skill processes external data from git diffs and pull requests, which presents a surface for indirect prompt injection. This risk is effectively mitigated by specific instructions in SKILL.md requiring the agent to manually verify every finding against the source code before reporting it to the user.

codex-review