Trust Score
87
+6 this week updated 2m ago
Risk: flagged
23
-12 risk spike updated 1m ago
Score: stable
72
+2 trend updated 4m ago
Preflight Decision
adjusted 46
Review
$2,500 + new payee policy run now
NEW Agent Trust Infrastructure

Don't get pinched.
Know who to trust.

Trust score shows safety risk. Behavior score shows reliability over time. Check both before payments, key sharing, or delegated execution.

✓ Decayed Scoring ✓ Audit Export ✓ Trusted Ingest + Provider Mapping
Free tier · 20 agents · 1,000 events · No credit card

Where your API key goes (copy this)

Add your key to the machine where your agent runs. Do not put it in public frontend code.

CLAWTRUST_API_KEY=claw_your_api_key_here

Then download the OpenClaw bundle and reference this same env var in your skill config.

Download Skill Bundle (.zip)
Optional: run a quick API test

Already have a key? Run your first check now

Tip: save your key as CLAWTRUST_API_KEY in your backend/bot server.
Put your key in backend env var CLAWTRUST_API_KEY. Common issues: 401 = wrong/missing key · 402 = plan limit (use Upgrade This Key) · 429 = rate limit.
Current and future integrations
Live now
REST APIPayment PreflightCoinbase Wallet Connector Usage EndpointAudit Export (JSON/CSV)API Key Self-Serve
Coming soon
OpenClaw Skill WrapperMCP Server PackageTypeScript SDK Python SDKZapier / Make ConnectorIntegration Request Queue
Scoring Model

What Trust and Behavior actually mean

Two separate 0-100 scores so teams can distinguish safety risk from delivery quality.

Trust Score

Safety and abuse risk

Trust drops when high-risk incidents happen. It recovers over time with clean activity.

  • Lowered by fraud signals, abuse reports, security flags, and severe failures.
  • Recent incidents count more than old ones (decay model).
  • Used as a hard safety floor in policy decisions.
Behavior Score

Execution reliability

Behavior tracks whether the agent follows through on tasks and expected outcomes.

  • Raised by on-time task completion and successful executions.
  • Lowered by timeouts, cancellations, abandonments, and repeated misses.
  • Lets teams spot unreliable agents even without explicit abuse.
Decision Layer

How decisions are made

Policy uses Trust first, then applies Behavior and transaction risk context.

  • Output is allow, review, or block with a plain-language explanation.
  • Risk context can include amount, new payee status, and first-time counterparties.
  • Every decision is logged and exportable for audit/compliance.
Integrations

Auto-ingest trusted events in minutes

Use signed connectors so your trust network updates automatically from real activity.

Verified Connector Quickstart

Turn on signed ingest, map provider events, and start feeding trust+behavior without manual event posting. Includes Coinbase-compatible wallet event mapping via source=wallet.

  • 1. Check readiness: GET /v1/integrations/readiness?source=wallet
  • 2. Create secret: POST /v1/integrations/ingest/secret
  • 3. Validate mapping: POST /v1/integrations/map-event
  • 4. Send signed events: POST /v1/integrations/ingest/events
Open API Docs View Connector Quickstart

OpenClaw 10-line install

Install the official wrapper and gate risky actions with one policy preflight call.

npx clawhub@latest install claw-trust-scores

# .env
CLAWTRUST_API_KEY=claw_xxx
CLAWTRUST_BASE_URL=https://clawtrustscores.com

# tools
get_score(agentId)
preflight_payment(agentId, amountUsd, newPayee)
log_event(agentId, kind, eventType)
Download Skill Bundle (.zip) View on GitHub
Agent Finance Mode

Gate money movement before it happens

Use high-assurance presets, required attestations, and signed connector events for wallet-native agents.

Preset

Wallet Guarded

Strong default for agents that move funds from wallets or exchanges.

  • Requires verified sensitive events.
  • Requires connector.wallet.verified and operator.kya.completed on risk.
  • Attestation failure decision defaults to block.
POST /v1/policy/presets/wallet_guarded
Preset

Money Movement Strict

Maximum safety mode for payment rails and autonomous treasury operations.

  • High confidence and signal-quality thresholds.
  • Requires connector.payment.verified + KYA.
  • Blocks high-risk actions when credentials are missing.
POST /v1/policy/presets/money_movement_strict
Connectors

Wallet + Prediction + Runtime

Map live events without manual posting, then feed preflight decisions with verified context.

  • source=wallet for Coinbase-compatible transaction and key-risk events.
  • source=prediction_market for settlement and abuse flags.
  • source=runtime for execution and sandbox safety.
GET /v1/integrations/readiness?source=wallet
Features

Everything you need to
gate risky agent actions

One decision call. Dual scoring (trust + behavior). Full audit trail.

Preflight Gating

Block, review, or allow agent actions based on real-time trust + risk context before damage happens.

POST /v1/.../preflight
decision "review"
adjustedScore 48
riskPenalty -27

Dual Scores + Decay

Trust score (safety risk) and behavior score (execution quality), both 0–100 with time decay so recent activity matters most.

score 74 · medium-high

Decision Audit Export

Every block, review, and allow decision is logged with timestamp, agent ID, score, and reason. Export as JSON or CSV for compliance.

100%
Decisions logged
JSON + CSV
Export formats
GET /v1/audit/decisions?format=csv
timestamp2026-02-18T14:22:01Z
actionpayment_preflight
outcomeblock
score18
reason"Below hard min 35"
How It Works

One API call before risky actions

Integrate trust scoring into any agent workflow in minutes.

Log Events

POST positive or negative signals as agents interact. Each event is timestamped and weighted.

Check Scores

GET trust + behavior scores for any agent, plus explanation, signal quality, and event history.

Gate Actions

Use preflight to get allow/review/block decisions with risk-adjusted scoring before execution.

Webhook Alerts

Get notified when scores drop below thresholds. Automatic suppression prevents alert fatigue.

Audit Export

Export every decision as JSON or CSV. Compliance-ready. Up to 5,000 entries per API key.

Integrations

First-class payment preflight policy plus wallet/runtime connector templates. Framework adapters for TypeScript and Python coming soon.

Pricing

Simple, transparent pricing

Free for testing. Paid tiers when you go to production.

Free
$0 /month
  • ✓ 20 tracked agents
  • ✓ 1,000 events / month
  • ✓ 3,000 score checks / month
  • ✓ Basic audit log
  • ✓ Community support
Get Started
Starter
$19 /month
  • ✓ 100 tracked agents
  • ✓ 5,000 events / month
  • ✓ 10,000 score checks / month
  • ✓ Webhook alerts
  • ✓ CSV + JSON export
Get Started
Pro
$79 /month
  • ✓ 1,000 tracked agents
  • ✓ 100,000 events / month
  • ✓ 200,000 score checks / month
  • ✓ Priority support
  • ✓ Full audit + export path
Get Started
FAQ

Frequently Asked Questions

How does the trust score decay work? +
Scores use a 30-day half-life decay. Recent events count most. A negative event from 60 days ago has roughly 25% of the weight of one from today. This means good actors can recover, and old incidents don't permanently penalize agents.
What's the difference between trust score, behavior score, and preflight? +
A score check (GET /v1/score) returns both Trust and Behavior scores. Trust tracks safety/compliance risk. Behavior tracks delivery reliability. A preflight call combines both scores with risk context like payment amount and new payees to return allow/review/block.
Can I export audit logs for compliance? +
Yes. GET /v1/audit/decisions supports both JSON and CSV export. Every score check and preflight decision is automatically logged with timestamp, agent ID, outcome, score, and reason. Up to 5,000 entries per API key.
Is there a free tier? +
Yes. The free tier includes 20 tracked agents, 1,000 events, and 3,000 score checks per month. No credit card required. You can upgrade instantly when you need production volume.
What integrations are supported? +
Payment preflight policy is built-in, with connector templates for wallet, auth, marketplace, prediction-market, and runtime events. Framework adapters for TypeScript and Python are on the roadmap. The REST API works with any language or platform that can make HTTP requests.

Ready to score your first agent?

Start free. No credit card required. Upgrade when you're ready.