The Agent Skills Directory

[COMMAND_EXECUTION]: The skill employs standard Git and GitHub CLI (gh) commands to facilitate repository operations, including git diff, git log, gh pr create, and gh pr merge. These commands are used to analyze changes and automate PR delivery workflows. All destructive operations (e.g., force-pushes, branch deletions) are explicitly gated behind user consent or mandatory backup protocols.
[INDIRECT_PROMPT_INJECTION]: As the skill ingests and analyzes untrusted data such as commit messages and pull request diffs (ingestion points in SKILL.md and pr-ship-flow.md), it identifies as having an indirect injection surface. However, it mitigates these risks using clear boundary markers like ## Review focus blocks (reference/pr-workflow-patterns.md) and a capability inventory limited to repository management tools. Furthermore, it implements sanitization via Hard Gates that require intent alignment verification and human-in-the-loop checkpoints for risk-heavy changes.
[SAFE]: The skill demonstrates a high level of security awareness, including specific logic for detecting credentials and sensitive file patterns (e.g., .env, .key) to trigger security escalation to a specialized agent (Sentinel). It enforces strict adherence to branch protection rules and prohibits the bypass of repository-defined safety gates unless explicitly authorized by the user.

guardian