The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides behavioral guidance on how to handle external instructions (review feedback). It explicitly directs the agent to treat such feedback as 'technical claims to evaluate' rather than 'orders to follow'. This posture increases the agent's resilience to potentially malicious or incorrect instructions provided during the review process.
[COMMAND_EXECUTION]: The skill references standard development activities such as opening functions, tracing callers, and using grep to verify claims against the actual codebase. It also mentions a required sub-skill tdd, implying the execution of test suites to validate changes.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from external reviewers or subagents.
Ingestion points: External review feedback and PR comments provided via user input or external tools.
Boundary markers: The skill lacks explicit syntactic boundary markers for input data but implements a logical boundary through its mandatory sequence (Understand, Verify, Evaluate).
Capability inventory: The agent has the capability to read, grep, and modify the local codebase and execute tests via the tdd sub-skill.
Sanitization: Technical verification against the source of truth (the codebase) serves as the primary sanitization mechanism before any changes are implemented.

receive-review