The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses git and grep to audit the repository state and identify markers such as TODOs or FIXMEs within changed files.
[COMMAND_EXECUTION]: In Step 6.5, the skill orchestrates cross-model reviews by invoking codex or claude CLI tools. This execution is designed with security in mind, utilizing read-only sandboxes, ephemeral sessions, and piping untrusted data via standard input to prevent command injection.
[PROMPT_INJECTION]: The skill processes untrusted input in the form of git diff outputs. It mitigates indirect prompt injection risks by using isolation: "worktree" for cold reviews and explicitly instructing sub-agents to treat diff content as untrusted data rather than commands.

review