Live on Sui Testnet

ClawAll

Constraint, Governance, and Audit Layer for OpenClaw on Sui

Autonomous agents should not hold unrestricted execution power.
ClawAll enforces policy checks, operator governance, and on-chain safety constraints.

View on GitHub Watch Demo $ npm run setup
Demo Shell current runtime flow
$ npm run gateway
[clawall-gateway] listening on http://127.0.0.1:3011
[clawall-gateway] transfer gate: enforced=true key_configured=true
$ npm run setup
Bootstrap and runtime refresh complete
$ openclaw clawall status
kill_switch=frozen:false policy_gate=enabled source=plugin
$ openclaw clawall transfer --amount 10000000 --recipient 0x...
risk=MEDIUM approval=required
$ openclaw clawall tx --limit 3
digest=D6gKSC...jfAS status=EXECUTED layer=CHAIN
digest=-- status=BLOCKED layer=RISK_ENGINE
Local dashboard exists for operators only. Use docs section "Local Dashboard" for setup.

The Problem

AI agents with raw
execution power are dangerous

Without a constraint layer, autonomous agents can drain wallets, execute destructive commands, and escalate privileges — silently.

01
Wallet Drain
Rapid micro-transactions bypass manual review, draining funds before anyone notices.
02
Destructive OS Commands
Unconstrained agents can execute rm -rf and other irreversible system commands.
03
Privilege Escalation
Cross-domain actions let agents bootstrap elevated access through chained operations.
04
No Audit Trail
Without enforcement events, there's no record of what an agent did or why.
05
Bypassed Human Intent
Agents optimizing for a goal can take paths that violate user intent — legally and technically.
06
No Override Mechanism
Once an agent is running, there's no reliable way to stop a single bad action mid-execution.

How it works

Seven-layer enforcement
pipeline

Every intent passes through all layers in sequence. Any layer can halt execution — permanently.

AI
AI Agent
Submits intent to the pipeline
01
Intent Firewall
Blocks unsafe commands instantly
02
Risk Engine
Multi-factor behavioral scoring
03
Policy Engine
Allow / Alert / Block / Require Approval
04
Telegram Governance
Human-in-the-loop approval
05
Move Enforcement
On-chain constraint execution
06
Walrus Audit
Immutable event trail
8
Risk Factors Scored
3
Policy Levels
∞
Kill-switch survives restarts
100%
On-chain audit coverage

Security Layers

Every layer is a hard stop

Layers are independent — a failure in one doesn't bypass the others.

Layer 01
Intent Firewall
Inspects raw intent before any scoring. Blocks destructive OS commands, unauthorized domain escalation, and known attack patterns outright.
rm -rf domain escalation pattern match
Layer 02
Risk Engine
Computes a 0–100 risk score from eight independent signals including transfer size, recipient novelty, velocity, and behavioral anomalies.
score 0–100 8 signals compounding
Layer 03
Policy Engine
Maps risk level to action. LOW auto-approves. MEDIUM allows with Telegram alert. HIGH requires explicit human approval before execution proceeds.
LOW → allow MEDIUM → alert HIGH → approval
Layer 04
Telegram Governance
Real-time human oversight via Telegram bot. Inline approve/reject buttons, live logs, freeze/resume commands, and on-chain transaction lookup.
/freeze /resume /logs /status
Layer 05
Kill Switch
Persistent file-based kill switch survives process restarts. Triggered by critical OS violations or manual /freeze. Cross-domain: OS violation freezes blockchain actions.
persistent cross-domain kill-switch.json
Layer 06
Move Enforcement
Transfer constraints are minted on-chain with max_amount, allowed_recipient, and expiry_ms. Execution is enforced by the Move runtime — not just application logic.
max_amount expiry TransferExecuted
Risk Level Score Range Action Telegram
● LOW 0 – 39 Auto-approved Silent
● MEDIUM 40 – 69 Allowed with alert Alert sent
● HIGH 70 – 100 Blocked pending approval Approve / Reject

Demo

Watch it in action

Live runtime simulation: risk scoring, governance approval, freeze controls, and on-chain execution records.

Runtime Demo

Risk scoring, governance approval, and execution flow in a single run.

clawall-demo.mp4

ClawAll Setup Walkthrough

Step-by-step bootstrap flow for local setup with subtitle support.

clawall-setup.mp4

Tech Stack

Built on real infrastructure

No mocks. Real on-chain transactions, real Telegram bot, real Walrus blobs.

01 Sui Move Smart contract enforcement layer — on-chain constraint minting & execution
02 Sui JSON-RPC Transaction signing, execution, and event querying
03 Node.js ESM Agent runtime — modular, async, fully typed pipeline
04 Telegram Bot API Human governance interface — approvals, freeze, audit commands
05 Walrus Decentralized audit blob storage — every execution logged immutably
06 Custom Risk Engine Multi-factor behavioral scoring — 8 signals, compounding logic

Get Started

Run it locally

Clone
$ git clone https://github.com/tejas0111/clawall
Install
$ npm install
Configure .env
PRIVATE_KEY=<your_private_key>
GUARD_CAP_ID=<your_guard_cap_id>
PACKAGE_ID=<deployed_package_id>
TG_BOT_TOKEN=<telegram_bot_token>
TG_CHAT_ID=<telegram_chat_id>
CLAWALL_PLUGIN_KEY=<plugin_key>
Run Services
$ npm run gateway
Gateway started on 127.0.0.1:3011
$ openclaw clawall status
Runtime status available

============================================================================
===     ===  ===========  =====  ====  ====  =====  =====  ========  =======
==  ===  ==  ==========    ====  ====  ====  ====    ====  ========  =======
=  ========  =========  ==  ===  ====  ====  ===  ==  ===  ========  =======
=  ========  ========  ====  ==  ====  ====  ==  ====  ==  ========  =======
=  ========  ========  ====  ==   ==    ==  ===  ====  ==  ========  =======
=  ========  ========        ===  ==    ==  ===        ==  ========  =======
=  ========  ========  ====  ===  ==    ==  ===  ====  ==  ========  =======
==  ===  ==  ========  ====  ====    ==    ====  ====  ==  ========  =======
===     ===        ==  ====  =====  ====  =====  ====  ==        ==        =
============================================================================
ClawAll Demo Shell
Real OpenClaw path with security and tamper controls.

1 → Normal Transaction
2 → Medium Risk Transaction
3 → High Risk Transaction (Approval Required)
4 → Simulate OS Attack
5 → Show Frozen Status
6 → Reset In-Memory State
0 → Exit

clawall>

Roadmap

What's next

01 Multi-sig governance with operator quorum approvals
02 Expanded tx-history and chainlogs analytics in local dashboard
03 Policy bundle lifecycle tooling (sign, anchor, verify)
04 Stronger automated red-team and forensics pipelines
05 Production-safe proxy pattern for hosted operations UI
06 Attested agent identity and verifiable governance actions