The Agent Skills Directory

[SAFE]: The skill implements a robust verification framework for external input, instructing the agent to treat reviewer feedback as suggestions to be technically validated rather than orders to be executed blindly. This reduces the risk of implementing malicious or incorrect code.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to handle untrusted data from external PR comments.
Ingestion points: Code review feedback and PR comments processed during the response pattern (SKILL.md).
Boundary markers: Explicit instructions to 'Verify before implementing' and 'Check against codebase reality' act as logic boundaries.
Capability inventory: File searching via grep and PR interaction via gh api (SKILL.md).
Sanitization: The skill mandates a verification process including checks for technical correctness, compatibility, and YAGNI (You Aren't Gonna Need It) principles before any action is taken.
[COMMAND_EXECUTION]: Uses standard development tools including grep and the GitHub CLI (gh api) for intended functionality. The commands use template placeholders for PR metadata, which is consistent with standard workflow automation.

receiving-code-review