The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides scripts and workflow templates that execute shell commands using the gh (GitHub CLI) and jq utilities. These tools are used to query repository state and perform authenticated API calls to GitHub for managing branch settings and deleting git references.
[SAFE]: The skill implements a safety-first architecture, specifically including a mandatory guard that verifies no open Pull Requests target a branch as their base before allowing deletion. This prevents the unrecoverable closure of dependent PRs, which is a common failure mode in automated branch cleanup.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection by processing external repository data.
Ingestion points: The workflow ingests branch names and PR metadata from GitHub Actions event payloads and API responses (gh pr list, GraphQL queries).
Boundary markers: There are no explicit delimiters or instruction-isolation markers around the branch names used in script variables.
Capability inventory: The skill has permissions to delete git references (contents: write) and modify repository configuration via the GitHub API.
Sanitization: The logic includes specific URL-encoding for potentially problematic characters like # and % in branch names before they are used in API paths, although it largely depends on Git's own ref-format validation.

branch-cleanup