The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local system commands including git diff to determine the review scope and uses codex exec or spawn-codex-worker to launch analysis sub-agents. These actions are fundamental to the tool's stated purpose of reviewing code.
[DATA_EXFILTRATION]: The skill accesses repository history and file contents to provide context to the reviewers. This data is used within the agent's workflow to generate the review verdict.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests untrusted code from the repository and includes it in prompts sent to sub-agent reviewers.
Ingestion points: SKILL.md (Step 1) reads repository diffs and file contents.
Boundary markers: references/reviewer-prompt.md uses markdown headers to delineate the code under review but lacks explicit instructions to the sub-agent to ignore any instructions embedded within that code.
Capability inventory: The skill can execute CLI tools and present synthesized results to the user.
Sanitization: No specific sanitization or escaping of the ingested code is performed before it is placed in the reviewer templates.

interrogate