Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
Add code-level enforcement hook (authensor-gate.sh) — bypass-proof PreToolUse hook with deterministic classification and redaction. Two-tier enforcement model.
What This Skill Does
Authensor Gateway is a policy gate that intercepts every OpenClaw tool call before it executes. It classifies each action as safe, high-risk, or dangerous, then automatically allows, requires your approval, or blocks it based on your policy. Only action metadata is sent to the control plane, never file contents, API keys, or conversation data.
Unlike blanket sandbox modes, it lets you auto-allow safe reads, require explicit approval for writes and shell commands, and hard-block dangerous operations, all without locking down the agent entirely.
When to use it
- Approving or blocking file writes from a third-party marketplace skill
- Preventing untrusted skills from reading SSH keys or .env files
- Getting an approval prompt before a skill runs npm install or curl commands
- Generating a timestamped audit trail of all agent tool calls for compliance
- Blocking destructive shell commands like rm -rf in automated workflows
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks a marketplace skill to write updated config to /src/config.js
- 1Classify the Write call as filesystem.write with resource /src/config.js
- 2POST to the control plane with the action type, resource, and tool name
- 3Control plane returns require_approval with receipt ID rec_abc123
- 4Pause execution and tell the user to run: openclaw approvals approve rec_abc123
- 5Poll GET /receipts/rec_abc123 every 30 seconds until the status changes to approved or rejected
File write executes only after the user approves; the receipt is stored with a timestamp for audit
Requirements
Accounts, API keys, or tools you or your AI assistant may need to set up while using this skill.