The Agent Skills Directory

[SAFE]: The skill follows a 'Manual Invocation Only' principle, ensuring that no actions, especially deletions, are performed without explicit human request and confirmation.
[SAFE]: Implements comprehensive branch protection that aggregates data from multiple sources (Forge APIs, local worktree, fallbacks, and user config) to ensure critical branches are never targeted.
[SAFE]: Mitigates Indirect Prompt Injection (Category 8) by providing clear instructions to the agent to treat third-party text, such as PR titles or code comments, strictly as data and to disregard any instructions or commands embedded within them.
[SAFE]: The skill enforces a 'Plans First' transparency model, presenting the user with clear evidence (SHAs, PR numbers, committer dates) for each branch before asking for approval.
[SAFE]: Uses a least-privilege approach by restricting allowed tools to a specific set of read-only and local git commands, while explicitly requiring user intervention for remote write operations like 'git push'.
[SAFE]: Includes robust recovery instructions in the documentation, such as recording tip SHAs to allow users to restore deleted branches if necessary.

prune-branches