Overview
Browser Agent Trust Hub is the pre-flight safety layer every OpenClaw operator needs before putting a browser or computer-use agent into production. Define your governance policy once and audit any planned workflow in seconds. The trust scoring engine evaluates each proposed action across six finding categories and produces a three-verdict result (ALLOW / REVIEW / BLOCK) with full evidence and remediation steps.
Key Features
- Policy-based trust scoring 0–100 with configurable minimum threshold
- Three-verdict engine: ALLOW, REVIEW, BLOCK with automatic BLOCK on CRITICAL findings
- Six detection categories: sensitive data, tool/domain violations, audit gaps, approval gates, evidence gaps
- Approval gate enforcement for irreversible actions (submit, purchase, delete, pay)
- Path-sandboxed file I/O — absolute paths and traversal rejected
- Zero external dependencies: pure Python stdlib, no network calls
Use Cases
- Pre-flight governance before any browser automation goes live
- Auditing planned workflows for sensitive data exposure
- Enforcing approval gates on irreversible browser actions
- Pairing with browser-automation and agentic-browser-workflow-templates
Requirements
- OpenClaw v2026.3.23 or later
- Python 3.8+ (stdlib only — no pip installs required)