The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and acts upon external data provided via CodeRabbit review comments, which is a potential vector for malicious instructions.
Ingestion points: gh api graphql fetches review thread data in SKILL.md (Step 3) and github.md (Section 2).
Boundary markers: Explicit instructions in SKILL.md (Step 4) and github.md warn the agent that external prompts are untrusted evidence and should not be executed as commands.
Capability inventory: The skill uses git commit, git push, and various gh commands to modify the repository and interact with PRs.
Sanitization: The agent is directed to validate fixes against a "restored task contract" and perform multi-step verification before committing changes.
[COMMAND_EXECUTION]: The skill involves the execution of local CLI tools to manage repository state and GitHub PRs.
Evidence: The workflow utilizes git and gh (GitHub CLI) for operations such as git status, git push, and gh api calls.

autofix