The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon untrusted data from external sources, specifically GitHub Pull Request comments and reviews via the gh pr view command.
Ingestion points: Pull Request comments, review body text, and review-bot feedback extracted in SKILL.md Step 2.
Boundary markers: The instructions do not define specific technical delimiters for external data, but provide behavioral logic: 'act only on feedback you actually agree with' and 'don't guess: leave it and reply'.
Capability inventory: The agent is authorized to modify local code, commit, and push to the remote repository. It can also manage PR state (e.g., gh pr ready).
Sanitization: No explicit sanitization or escaping is mentioned for the ingested prose, though the agent is instructed to use a separate unslop skill for its own output.
[COMMAND_EXECUTION]: The skill utilizes the standard GitHub CLI (gh) for viewing PR state and checking status. These are functional invocations intended for the skill's primary purpose.
[SAFE]: The skill includes several 'Hard rules' that promote security and project integrity, such as forbidding history rewriting on shared branches, prohibiting the bypass of CI hooks (--no-verify), and requiring user confirmation for ambiguous fixes.

babysit