The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various git and gh (GitHub CLI) commands to interact with repositories, manage branches, and handle PR metadata. It also runs local bash scripts bundled with the skill (scripts/list-actionable-prs.sh, scripts/fetch-pr-data.sh) to query the GitHub GraphQL API and filter PRs. These operations are consistent with the skill's stated purpose of PR feedback management.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, specifically PR review comments and code suggestions (```suggestion blocks) fetched via the GitHub API. It addresses this surface in Step 3 by explicitly instructing the agent to independently verify claims against the source code or documentation before acting, particularly for suggestions from automated reviewers. This provides a robust verification and sanitization logic to prevent the agent from blindly following malicious or incorrect instructions embedded in review comments.
[DYNAMIC_CONTEXT_INJECTION]: The skill uses the !command`` syntax in SKILL.md to inject the output of benign environment discovery commands (git remote -v, git branch --show-current, and git status) into the prompt at load time. These commands are safe and used solely to provide the agent with necessary project context.
[REMOTE_CODE_EXECUTION]: The skill acts as an orchestrator in its multi-PR mode (--all), using the Task tool to dispatch subagents. These subagents are delegated tasks within isolated git worktrees and are restricted by a prompt template in REFERENCE.md that prevents them from performing GitHub-side mutations like pushing code or resolving threads, which the main orchestrator handles after validation.

git-pr-feedback