LetsPing OpenClaw Adapter

OpenClaw integration for LetsPing Agent Credentials & Behavioral Firewall.

This package provides a letsping_ask(tool_name, args_json, risk_reason) skill for OpenClaw agents. It routes high-risk actions through LetsPing’s firewall and (optionally) human-in-the-loop console using the same agent identity and guardrail model as other frameworks (LangGraph, CrewAI, Vercel AI SDK, custom Python). OpenClaw is just one adapter on top of LetsPing’s agent-first architecture.

Payloads are encrypted client-side using AES-GCM with the pairing secret. The relay server and database store only ciphertext.

Installation

Clone into your OpenClaw workspace:

git clone https://github.com/CordiaLabs/openclaw-skill ~/.openclaw/workspace/skills/letsping

Install dependencies:

cd ~/.openclaw/workspace/skills/letsping
npm install

Restart the OpenClaw gateway.

Configuration

Add to ~/.openclaw/openclaw.json:

{
  "skills": {
    "entries": {
      "letsping": {
        "env": {
          "LETS_PING_SECRET": "lp_live_..."
        }
      }
    }
  }
}

Note: The Supabase URL and Key are pre-configured for LetsPing. You can override them with LETSPING_SUPABASE_URL and LETSPING_SUPABASE_ANON_KEY for self-hosting.

Obtain LETS_PING_SECRET by pairing at https://letsping.co/openclaw/pair, or by using LetsPing’s Agent Credentials API to provision a dedicated agent key for your OpenClaw gateway.

Usage

Add the following to your agent's system prompt or AGENTS.md:

You have full authority for safe actions: reading files, logs, web searches, data analysis without side effects.

You MUST call letsping_ask BEFORE any high-risk action. NEVER execute high-risk actions directly.

High-risk includes:
- Financial: Spending money, bookings, transactions > $10
- Destructive: Delete/overwrite files, DB rows, configs
- Social: Posting, sending DMs/emails to new/public contacts
- Infrastructure: Modifying DNS, env vars, deployments, infra APIs

Provide:
- tool_name: exact tool name
- args_json: stringified JSON of arguments
- risk_reason: clear justification

After call:
- APPROVED: Use ONLY authorized_payload (may be patched)
- REJECTED or timeout: Abort, retry safely, or ask for guidance

Example:
letsping_ask(tool_name: "vercel_deploy", args_json: "{\"project\":\"my-app\",\"env\":\"production\",\"force\":true}", risk_reason: "Production deployment with force flag")

Security Model

Payloads are encrypted on the agent using AES-GCM derived from LETS_PING_SECRET. Only ciphertext is sent to the relay and stored in Supabase. Decryption occurs solely on paired devices using the same secret from local storage.

The relay cannot read payloads.

How It Works

1. Agent calls letsping_ask for a high risk action.
2. Skill encrypts payload locally and sends ciphertext to relay.
3. Relay stores request and notifies paired device.
4. Human reviews (decrypts), optionally edits, then approves/rejects.
5. On approval, relay returns patched ciphertext (or original).
6. Agent decrypts and resumes with authorized payload.

Default timeout: 10 minutes.

Troubleshooting

• Skill not loading: Check gateway logs (openclaw gateway --log-level debug). Ensure npm install succeeded.
• Agent skips call: Use strong models (Claude Opus, GPT-4o). Add more prompt examples.
• No notifications: Verify pairing and browser permissions.
• Timeout errors: Agent should handle gracefully.

Issues/PRs: https://github.com/CordiaLabs/openclaw-skill

https://letsping.co