The Agent Skills Directory

[SAFE]: The skill implements a structured multi-agent pipeline using the Agent tool. Reviewers are dispatched with subagent_type: Explore, ensuring they operate in a read-only capacity during the analysis phase.
[SAFE]: Includes a dedicated hygiene lane that specifically scans for hardcoded secrets, credentials, and private keys in the code diff, treating them as critical findings to prevent accidental exposure.
[SAFE]: Uses standard Git commands via Bash to resolve review scopes and manage background tasks using worktree isolation, following best practices for local development tooling.
[SAFE]: Orchestration logic is transparent and focuses on producing a REVIEW.md report. It does not perform unexpected network operations, exfiltrate data, or attempt to modify system configurations.

code-review