The Agent Skills Directory

[SAFE]: The skill implements strong input validation for the Bug-ID using the regex ^[A-Za-z0-9_-]{1,64}$, which effectively mitigates path traversal (../), command injection, and branch naming attacks.
[SAFE]: It enforces strict environment isolation by requiring all sub-agents with write access to operate within independent git worktrees. This prevents cross-task interference and protects the main workspace.
[SAFE]: The skill uses structured communication delimiters (===== AGENT_COMPLETION_BEACON ===== and ===== END_BEACON =====) and instructs the master agent to ignore any content following the end marker, providing a defense against indirect prompt injection from sub-agent outputs.
[SAFE]: Local security is prioritized by creating worktrees in $TMPDIR (which is user-private on macOS) and applying umask 077 to ensure that temporary source code and build artifacts are not readable by other users on the system.
[SAFE]: Data exfiltration is mitigated by disabling git push capabilities (remote set-url --push origin PUSH_DISABLED) within the temporary worktrees and mandating the use of mock data for testing instead of real credentials.
[SAFE]: The workflow incorporates an independent review role that is isolated from the fixer role, ensuring that all code changes are verified by a separate agent before being merged.

bug-fix-expert