The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains an explicit instruction to override and violate platform-level guidelines stored in 'CLAUDE.md' (e.g., "explicit CLAUDE.md violation"). This is used to suppress specific polite conversational responses. Additionally, it defines a hidden signaling phrase ("Strange things are afoot at the Circle K") to be used when the agent is uncomfortable with a situation.
[COMMAND_EXECUTION]: The skill utilizes shell-based commands for codebase analysis and platform interaction:
Uses grep to check for code usage before implementing features.
Uses gh api to interact with GitHub pull request comments via the repos/{owner}/{repo}/pulls/{pr}/comments/{id}/replies endpoint.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from "External Reviewers" which could contain malicious instructions designed to exploit the agent's capabilities.
Ingestion points: External code review feedback (SKILL.md).
Boundary markers: None identified; feedback is processed as natural language instructions.
Capability inventory: File system modifications ("IMPLEMENT"), codebase search (grep), and GitHub API access (gh api).
Sanitization: The skill relies on manual "skepticism" and verification steps rather than automated sanitization or sandboxing of the input.

receiving-code-review