The Agent Skills Directory

[SAFE]: The skill implements a comprehensive multi-stage pipeline (SELECT, CHECK YOURSELF, PILOT, FAN-OUT, REPORT) that ensures human oversight at critical junctures. The mandatory pilot stage prevents non-deterministic changes from being applied to the fleet without line-by-line diff review.
[SAFE]: Security scanning is integrated into the workflow, with instructions to scan staged diffs for secrets using gitleaks, detect-secrets, or specific regex patterns before every commit. This prevents the accidental mass-distribution of credentials.
[SAFE]: The skill provides strong protections against indirect prompt injection (CWE-1427) by explicitly instructing sub-agents to treat repository content like CLAUDE.md and PR templates as untrusted data. It uses structural delimiting and per-target isolation to prevent cross-contamination or hijacking from poisoned targets.
[SAFE]: Destructive operations are protected by a required reference-check pre-step, which verifies dependencies before allowing deletions or rewrites, ensuring project stability across bulk updates.

force-multiplier