The Agent Skills Directory

[COMMAND_EXECUTION]: The skill extensively uses command-line utilities including gh, git, find, grep, jq, yq, sed, awk, and python3. These tools are used for legitimate repository mapping, diff analysis, and interacting with the GitHub API. The implementation includes specific mitigations (such as URL encoding and shell-safe substitution) to prevent command injection from untrusted filenames found in the repository.
[EXTERNAL_DOWNLOADS]: The skill provides a hint to the user to download a configuration template from the author's public GitHub repository (github.com/bhagyamudgal/skills) for CodeRabbit integration. This is a documented, vendor-owned resource used for setup guidance.
[DATA_EXFILTRATION]: The skill reads pull request diffs, linked issue descriptions, and repository source code. This data is used to generate a review summary and per-finding comments, which are then transmitted to GitHub via authenticated API calls. All network traffic is directed to official GitHub domains for the intended purpose of pull request review.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection because it ingests untrusted data from PR content and issue bodies.
Ingestion points: Reads external content via gh pr view, gh pr diff, and gh issue view.
Boundary markers: Uses a 'Grounding Pass' to mechanicalize changes before review, but does not explicitly use delimiters for the diff content itself.
Capability inventory: Possesses write capabilities to the PR (via gh api), shell execution permissions, and local file system write access for state management.
Sanitization: Employs robust sanitization using jq for JSON construction and custom Python encoding for shell-bound file paths to prevent secondary injections.

review-pr