README.md

metadata

title: ClawCoat
emoji: 🦞
colorFrom: purple
colorTo: indigo
sdk: gradio
sdk_version: 5.20.0
app_file: app.py
pinned: true
license: apache-2.0
tags:
  - ai-governance
  - ai-agents
  - mcp-gateway
  - agent-security
  - trust-tiers
  - active-decision-making
  - human-in-the-loop
  - audit-trail
  - zero-trust
  - self-hosted
  - mcp
  - goose
  - kill-switch
  - agentic-ai
short_description: 'Every agent call: allow, gate, or block. Five tiers.'

ClawCoat - Active Decision Making

Every OpenClaw tool call passes through ClawCoat before execution. Allow, gate for human approval, or block. Every time.

This Space connects to a live ClawCoat deployment running the full governance pipeline. Select a demo agent, pick an action, and see an actual governance decision.

What You're Seeing

This is not a simulation. The decisions come from a live ClawCoat instance running the full stack - trust tier enforcement, Manners Engine scoring in real time, HITL approval gates, and a cryptographic audit chain recording every event.

• Active decision making - every call evaluated before execution. Allow, gate, or block.
• Five trust tiers - QUARANTINE → PROBATION → RESIDENT → CITIZEN → AGENT, earned by behavior
• Manners Engine - 8-factor behavioral score (0.0–1.0) updated in real time on every action
• Kill switch - suspend an agent instantly, every subsequent action rejected
• HITL gates - actions held for human approval regardless of trust tier
• Audit chain - every governance decision recorded, cryptographically chained

Trust Tier Flow

QUARANTINE → PROBATION → RESIDENT → CITIZEN → AGENT
 (all gated)  (safe tools)  (read/write)  (broad)   (apex)

No agent arrives at AGENT tier. Every agent starts at QUARANTINE and earns its way up through demonstrated behavioral compliance. The Manners Engine drives automatic demotion. Human operators drive promotion.

The Agent Autonomy SLA

ClawCoat is the working implementation of the Agent Autonomy SLA - a formal per-tier commitment framework defining what an OpenClaw agent may do, under what conditions, and with what audit trail. Jouneaux et al. identified this as an open challenge in November 2025 (arXiv:2511.02885). ClawCoat is the answer.

Full Repository

github.com/QuietFireAI/ClawCoat

Source code, documentation, proof sheets, deployment guide, and full test suite. 746 tests passing. 0 high-severity findings. Apache 2.0. Self-hosted.

---

ClawCoat v11.0.1 · Quietfire AI · Apache 2.0