The Agent Skills Directory

[COMMAND_EXECUTION]: Runs local shell commands for builds, tests, and linting (bun test, npm run build, bun run lint, bunx tsc). These are standard developer operations performed on the local filesystem.
[COMMAND_EXECUTION]: Uses git rev-parse and git log to programmatically determine commit ranges for subagent-driven review tasks.
[PROMPT_INJECTION]: Establishes strict behavioral instructions, such as the "Forbidden Responses" and "Iron Law" of verification, to override default AI conversational patterns and ensure technical rigor.
[PROMPT_INJECTION]: Manages the ingestion of untrusted data from external code reviewers as an indirect prompt injection surface. The skill mitigates this through a mandatory verification protocol.
Ingestion points: External reviewer feedback and comments (SKILL.md, references/code-review-reception.md).
Boundary markers: Conceptual workflow isolation (READ -> UNDERSTAND -> VERIFY) without technical delimiters.
Capability inventory: Local shell execution (npm, bun, git) and internal task delegation (Task tool).
Sanitization: Requires evidence from fresh verification command outputs before accepting or claiming completion of external suggestions.

code-review