The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses various command-line tools to interact with the local repository and fetch PR data.
Evidence: Use of git diff, git show, git status, git log, gh pr diff, and glab mr diff across SKILL.md and references/diff-checks.md.
Context: These tools are necessary for the skill's primary function of reviewing code changes.
[EXTERNAL_DOWNLOADS]: The skill fetches diff data from remote repositories if local CLI tools are unavailable.
Evidence: Uses curl -L to fetch .diff files from github.com and gitlab.com URLs in SKILL.md (Step 1).
Context: Fetches data from well-known version control services as a fallback mechanism for PR review.
[INDIRECT_PROMPT_INJECTION]: The skill has an inherent vulnerability surface because it processes untrusted data (git diffs and PR descriptions) that could contain malicious instructions designed to influence the agent's review.
Ingestion points: The skill ingests PR URLs, commit hashes, and raw diff text in SKILL.md (Step 1).
Boundary markers: No explicit boundary markers or "ignore embedded instructions" warnings are used when interpolating diff content into the analysis context.
Capability inventory: The skill has the capability to run shell commands (git, grep), make network requests (curl), and output text to the chat. It can also save files on explicit user request.
Sanitization: No explicit sanitization of the diff content or PR descriptions is mentioned.
Mitigation: The skill follows a "Human-in-the-loop" workflow (Step 10) and makes all output ephemeral by default, requiring user confirmation for critical actions like saving files.

pr-antipattern-review