The Agent Skills Directory

[SAFE]: The skill defines a coordination protocol for multi-agent workflows. It includes security-positive constraints such as a five-round limit on review loops to prevent infinite loops, disjoint write ownership to prevent race conditions or collisions, and a mandatory 'root verification gate' ensuring the primary agent validates all sub-agent output before acceptance. The instructions explicitly prohibit sub-agents from contacting each other or releasing work without approval. No code execution, network exfiltration, or obfuscation was found. Findings are documented neutrally and adhere to standard architectural patterns.

bounded-orchestration-contract