VeriClaw 爪印: Apple-native AI agent correction and OpenClaw companion workflow for LLM supervision, hallucination remediation, and role-drift diagnosis.

The control plane for AI desktop agents: watch, gate, and audit any agent before it touches a real machine. Sub-ms policy engine, independent AI verifier, human-in-the-loop gate, tamper-evident audit log, and a live Control Hub. Local-first and model-agnostic.

Active oversight layer (cockpit) for Claude Code agents — NOT an agent itself. Forces plan-first, surfaces decisions, makes thinking visible. Middle ground between yolo and baby-every-command.

Human-in-the-loop console for autonomous pentest/red-team agents — review live agent activity and approve or deny high-risk actions before they run.