The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, untrusted data including code files, roadmaps, specifications, and pull request patches.
Ingestion points: Target codebases, individual files, git diffs, and project documentation are ingested as sources of intent and implementation as described in SKILL.md and lenses/implementation-review.md.
Capability inventory: The skill has permissions to write to the local filesystem to apply "Bucket I
Safe Fixes" and execute validation commands defined in the environment.
Boundary markers: The skill relies on the underlying agent's context window and instruction-following capabilities to separate review data from instructions; no specific boundary markers are mandated.
Sanitization: The instructions do not specify sanitization for ingested code or data.
[COMMAND_EXECUTION]: The skill uses local shell commands to facilitate the review workflow.
Evidence: The workflow involves running git status --short to determine the scope of changes and executing "relevant validation commands" to verify the correctness of applied fixes, as outlined in lenses/implementation-review.md. These are standard developer utility operations.
[EXTERNAL_DOWNLOADS]: The skill references public documentation from a well-known developer repository.
Evidence: The architecture and design guidance in lenses/design.md identifies Matt Pocock's GitHub repository (github.com/mattpocock/skills) as the source for its methodology. This is a static reference for documentation and attribution.

code-review