The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is configured to execute repository-defined shell commands such as build, lint, test, and migration scripts. These operations are intended for verification of the code change and are restricted to non-mutating actions, with explicit instructions forbidding edits to tracked files or deployment operations.- [EXTERNAL_DOWNLOADS]: The instructions authorize the use of web research tools to consult official primary sources, including vendor documentation, specifications, and security advisories, to verify the correctness of external contracts and standards.- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface (Category 8) because it processes untrusted data from repository code and git diffs.
- Ingestion points: Untrusted data enters the context via git diff, git status, and focused file reads of the implementation under review.
- Boundary markers: The skill defines a 'Prompt Contract' for subagents that uses a 'frozen base packet' to isolate context, providing a structural boundary, although it does not mandate specific character delimiters for code blocks.
- Capability inventory: The agent has the ability to execute shell commands (build/test tools), perform web research, and spawn autonomous subagents.
- Sanitization: There is no explicit logic provided for sanitizing or escaping the content of the code files being reviewed before they are processed by the agent.

ln-12-delivery-reviewer