Multi Agent Protocol

OpenClaw-native v2 protocol for spec-first multi-agent delivery. Use when you need 2+ agents with explicit phase control, dual review gates, bounded retry/ci...

This skill is a coherent protocol specification (no code or installs). Before installing or relying on it: (1) confirm you actually run the referenced components (OpenProse, Lobster, task-store plugin and SQLite, and ACP if needed); (2) verify the orchestrator identity and enforcement of role checks in your task-store implementation (task_transition must be limited to the orchestrator); (3) ensure workspace paths and artifact URIs recorded by task-store do not leak sensitive files or secrets and that the SQLite DB is stored with appropriate access controls; and (4) note the publisher/source is 'unknown' and there is no homepage—if you need accountability, ask the owner for provenance or prefer a vetted internal copy. Overall the package appears internally consistent and appropriate for its stated purpose.

Type: OpenClaw Skill Name: multi-agent-protocol Version: 2.0.0 The skill bundle defines a structured multi-agent orchestration protocol (v2) focusing on state management via a SQLite-backed 'task-store' plugin. It emphasizes security best practices such as least privilege, explicit phase transitions, and human-in-the-loop approvals for side effects via the 'Lobster' component. No evidence of malicious intent, data exfiltration, or unauthorized execution was found; the instructions in SKILL.md and the tool definitions in task-store-plugin.md are consistent with the stated purpose of coordinating complex agent workflows.