The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from GitHub Pull Request comments and reviews.\n
Ingestion points: The scripts/pr-watch.py utility fetches PR comments and review bodies from the GitHub API.\n
Boundary markers: The skill employs [!NOTE] blocks for human notes and mandatory AI disclosure footers to distinguish agent activity.\n
Capability inventory: The agent has extensive capabilities to modify the repository, including git commit, git push, git rebase, and gh pr edit or gh pr merge.\n
Sanitization: Instructions in references/review-responses.md direct the agent to "assess each [comment] on its own before acting" and to "judge on merit," though no programmatic sanitization of comment bodies is implemented before processing.\n- [COMMAND_EXECUTION]: The skill relies on a local Python script to execute GitHub CLI commands.\n
Evidence: scripts/pr-watch.py uses subprocess.run([\"gh\", *args]) to fetch PR data and state.\n
Evidence: references/review-responses.md includes a GraphQL mutation command for resolving threads using gh api graphql.\n- [DYNAMIC_EXECUTION]: The skill instructs the agent to run background tasks and sub-agents to monitor activity.\n
Evidence: references/watch.md describes launching scripts/pr-watch.py as a background bash task or spawning a secondary sub-agent model to relay event information.

git