The Agent Skills Directory

[SAFE]: The skill enforces isolation by performing work in temporary git worktrees or dedicated branches, protecting the operator's primary environment from unintended side effects.
[SAFE]: Permitted shell operations are confined to a whitelist of standard development and version control utilities as specified in the allowed-tools configuration.
[SAFE]: The provided scripts for analyzing symbol sets and reference counts perform local operations only and do not exhibit any network exfiltration behavior.
[SAFE]: Extensive documentation on LLM failure modes and verification steps ensures that the refactoring process is robust and behavior-preserving.

refactor-verify