The Agent Skills Directory

[SAFE]: The skill uses official and trusted tools such as git, gh (GitHub CLI), and standard language toolchains (e.g., pnpm, cargo, ruff) for code analysis and repository interaction. All network activity is restricted to GitHub's official APIs. There are no external downloads, base64-obfuscated commands, or hardcoded credentials detected within the skill files or scripts.
[SAFE]: The skill processes untrusted pull request data (titles, bodies, and code diffs), which is a common surface for indirect prompt injection. However, this is inherent to its primary purpose as a code review tool, and the skill does not possess capabilities that would allow for significant privilege escalation or unauthorized system changes beyond standard repository interactions.
Ingestion points: Pull request content is ingested through scripts/pr-context.sh which calls the GitHub API.
Boundary markers: The instructions rely on standard LLM safety training when processing diffs and do not provide explicit text delimiters for the untrusted data.
Capability inventory: The skill can perform git repository operations, execute project-specific linting/testing commands, and post comments to GitHub pull requests.
Sanitization: Content is processed as structured JSON from the API but does not undergo specialized instruction-filtering before being presented to the agent.

review