The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests untrusted data from GitHub PR comments and reviews via the gh pr-review-check CLI and the scripts/collect-review.sh script.
Ingestion points: Data enters the context from the output of GitHub CLI commands and REST/GraphQL API responses.
Boundary markers: The protocol enforces structural parsing using jq and maintains a clear separation between the classification (Phase 3) and execution (Phase 6) phases.
Capability inventory: The agent can perform git push, issue GraphQL mutations via gh api, and execute local test suites like npm test or make test.
Sanitization: The documentation recommends using structured file inputs for GraphQL mutation bodies to prevent command or string injection, and uses jq to extract only specific validated fields from API responses.
[COMMAND_EXECUTION]: The execution pipeline involves several shell commands including git (pull, push, branch), gh (api, view, extension), and project-specific testing tools. These operations are essential to the skill's primary function of managing software review cycles and are triggered in a structured manner within the defined execution phases.
[EXTERNAL_DOWNLOADS]: The documentation references the installation of the gh-pr-review-check extension from a public GitHub repository. This is an explicit, user-facing dependency required for the skill's orchestration logic and is not a silent or hidden download.

review-pr-check