The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a Node.js script to execute local git and gh (GitHub CLI) commands. This is used to inspect branch status, check PR states, and perform branch deletions. The script employs execFileSync with argument arrays, which is a secure practice that prevents shell injection attacks.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources (Git metadata and GitHub PR states) which could theoretically contain malicious instructions.
Ingestion points: Local branch names retrieved via git for-each-ref and PR information fetched via gh pr list in scripts/classify-branches.mjs.
Boundary markers: The script formats results into a structured text table for the agent to present to the user.
Capability inventory: The script is capable of writing to a local recovery log file (fs.appendFileSync) and executing local branch deletion commands (git branch -d).
Sanitization: The skill avoids shell interpolation by passing arguments as arrays to subprocesses, and it provides a dry-run report for user review before execution.
[SAFE]: The skill follows security best practices by implementing a 'report-first' approach, ensuring all deletions are logged for recovery, and strictly limiting operations to the local repository without impacting remote branches.

safe-cleanup