The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local tools including bwwl, git, and the GitHub CLI (gh). It explicitly prohibits the agent from attempting to install bwwl if it is not already present on the system, which prevents unauthorized software installation.
[INDIRECT_PROMPT_INJECTION]: The skill processes contents of .github/workflows/ and output from the bwwl linter, which represent ingestion points for untrusted data. 1. Ingestion points: Workflow files in .github/workflows/ and linter results from bwwl. 2. Boundary markers: The skill requires showing a git diff and obtaining explicit user confirmation before finalizing changes. 3. Capability inventory: File editing (restricted to .github/), branch creation, and draft PR creation via gh pr create. 4. Sanitization: Relies on mandatory human-in-the-loop review of code changes and manual execution of sensitive git commands (add, commit, push) to prevent unauthorized actions.

workflow-fix