The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from GitHub, creating a surface for potential instruction injection.
Ingestion points: Untrusted data is ingested from PR bodies, code diffs, and Issue comments via gh pr view, gh pr diff, and gh issue view commands as described in SKILL.md.
Boundary markers: The instructions lack explicit delimiters or "ignore previous instructions" warnings when processing the PR and Issue content.
Capability inventory: The skill has powerful capabilities, including merging code (gh pr merge), force-pushing to branches (git push --force-with-lease), and modifying repository labels and issue states.
Sanitization: There is no evidence of sanitization or filtering applied to the external text before it is analyzed by the agent.
[COMMAND_EXECUTION]: The skill makes extensive use of system commands through the GitHub CLI (gh) and Git.
It performs repository management operations such as fetching PR lists, viewing diffs, commenting, and merging.
Safety features are incorporated, such as using --match-head-commit to prevent merging unreviewed commits and requiring user consent for all modifications and merges.

github-pr-merge