The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various shell commands to manage the git lifecycle and verify project health. These include standard git operations (git status, switch, pull, branch deletion), GitHub CLI commands (gh pr view, gh run list), and deployment-specific CLIs (vercel, wrangler, gcloud). It also runs project-defined scripts found in package.json, Makefiles, or justfiles for local verification.
[EXTERNAL_DOWNLOADS]: The skill performs a fast-forward pull from the remote repository (git pull --ff-only origin <primary>) to sync local state with the remote branch. It also accesses deployment status and smoke-tests production/live URLs as part of the verification workflow.
[INDIRECT_PROMPT_INJECTION]: The skill incorporates an attack surface for indirect prompt injection by reading untrusted data from the repository, including documentation files (AGENTS.md, CLAUDE.md, README.md) and PR metadata via gh pr view. This content is used to inform the agent's verification and next-step recommendations. However, the skill maintains safe defaults by preferring non-destructive git operations and requiring explicit user approval for implementation of new features.
Ingestion points: Local repository documentation (README, AGENTS.md, CLAUDE.md) and GitHub PR metadata.
Boundary markers: Not explicitly defined in instructions.
Capability inventory: File system access via git, network access via CLI tools and smoke tests, and execution of local build/test scripts.
Sanitization: Relies on standard CLI tool behavior; no explicit sanitization of documentation content mentioned.

merged