The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of GitHub PR comments, review bodies, and bot summaries.
Ingestion points: The scripts/collect_pr_feedback.py script and gh pr view commands ingest data from GitHub PRs into the agent's context.
Boundary markers: The skill does not use specific structural delimiters for external content, but the instructions include a requirement to 'Verify proposals against current code and PR intent' and to identify 'incorrect, or harmful suggestions'.
Capability inventory: The skill can check out branches, modify code, push commits, and post comments/replies via the GitHub API (gh api).
Sanitization: The instructions provide a logical gate ('Active-review gate') and require a 'local implementation review' of the diff, which serves as a verification step.
[COMMAND_EXECUTION]: The skill relies on executing shell commands via the GitHub CLI (gh) and a local Python helper script to interact with repository metadata.
Evidence: SKILL.md contains multiple examples of gh commands and python3 invocations for the collector script. scripts/collect_pr_feedback.py uses subprocess.run to call gh with various API and view arguments.

address-pr-feedback