Revolution Auto-Evolution

Multi-agent auto-evolution system — orchestrate review-execute-audit loops with 4 roles (Coordinator, Reviewer, Executor, Auditor). A single coordinator agen...

This skill appears to do what it claims: orchestrate multi-agent review→execute→audit loops using JSON task files in a workspace. However, it will read and write files under the resolved workspace (default: ~/.openclaw/agents/main/workspace), remove lock files, reset stuck tasks, and archive & delete completed task files. Before installing or enabling autonomous runs: (1) point OPENCLAW_WORKSPACE / EVOLUTION_TASKS_DIR to a dedicated sandbox directory; (2) back up any existing workspace/skills directories you care about; (3) review the scripts (heartbeat, monitor, pack-skill) to verify behavior you’re comfortable with—especially pack-skill which moves and deletes task files; (4) ensure sub-agents and models producing review/exec/audit outputs are trusted, since malformed or adversarial JSON in their outputs can influence state transitions; (5) start with manual/isolated runs (invoke scripts manually) before adding to cron/heartbeat for autonomous operation.

Type: OpenClaw Skill Name: ssa-revolution Version: 1.0.0 The 'ssa-revolution' bundle is a multi-agent orchestration framework designed to automate task execution through a review-execute-audit loop. The included Node.js scripts (heartbeat-coordinator.js, monitor.js, and pack-skill.js) manage task states, handle lock files, and archive results using standard local file system operations within the OpenClaw workspace. The logic is transparent, lacks any indicators of data exfiltration or unauthorized remote execution, and the instructions in SKILL.md are strictly aligned with the stated purpose of coordinating sub-agents for autonomous development.