The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with the local environment using standard git and gh (GitHub CLI) tools to manage repository state and close issues.
[SAFE]: Implements a robust 'dry-run' architecture via scripts/plan.sh. This script is read-only and only identifies candidates for deletion, ensuring no changes are made without oversight.
[SAFE]: Includes explicit protection logic to safeguard the default branch and 'staging' branch, preventing them from being deleted even if associated with a merged PR.
[SAFE]: Employs a mandatory human-in-the-loop confirmation step. The agent is instructed to show the full plan to the user and only proceed with deletions upon receiving clear, explicit consent.
[SAFE]: Uses the disable-model-invocation: true configuration, which limits the agent's ability to invoke other tools while this skill is active, reducing the risk of unauthorized tool usage.
[SAFE]: Indirect Prompt Injection Surface: The skill processes data from branch names and PR titles which could contain untrusted content.
Ingestion points: scripts/plan.sh ingests data from gh pr list.
Boundary markers: The skill does not use specific delimiters when presenting the plan to the user, but the output is intended for human verification.
Capability inventory: git push --delete and gh issue close are the primary actions.
Sanitization: Uses @tsv formatting in gh output to safely handle potential special characters in PR metadata during shell processing.

cleanup-merged-branches