The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes git and the GitHub CLI (gh) to perform its primary functions. It executes commands to retrieve branch diffs, pull request metadata, and issue descriptions, and it can post automated comments to GitHub via the CLI.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from GitHub, including PR descriptions and code diffs. This data is interpolated into the agent's context for evaluation, which could allow a malicious PR author to attempt to influence the agent's behavior or override the review rubric.\n
Ingestion points: Untrusted data enters the skill via gh pr view, gh pr diff, and gh issue view calls in SKILL.md.\n
Boundary markers: The skill employs a logical 'Inquisitor Doctrine' in RUBRIC.md to filter findings, but it lacks strict data delimiters or instructions to ignore embedded commands within the ingested text.\n
Capability inventory: The skill has the ability to read and write files, interact with the GitHub API, and execute local development tools and test suites.\n
Sanitization: No explicit sanitization or escaping is performed on the ingested PR content before analysis.\n- [DYNAMIC_EXECUTION]: The skill provides an interactive option to 'Apply fixes locally', which involves running project-specific test suites (e.g., dart test, go test). Running tests on unvetted code from a pull request is a known vector for code execution attacks, although the skill mitigates this by requiring explicit user confirmation before proceeding.

review-pr