ClawbotomyEvidence lab

Planning workspace / Inbox pack / Browser local

Define the powers before you test them.

Choose what an agent may read or change. Clawbotomy turns that intent into a concrete mock-Inbox scenario plan you can inspect, copy, and keep.

Current statePlanning only
Agent execution
None
Evidence collected
None
Provider requests
None
Production access
Unchanged

No deployed agent or module is loaded. No score. No permission decision. No authorization.

Plan data not uploadedNo mailbox connectedOperator intent recordedDecision remains empty

Define the intended boundary

Record what you are considering before evidence changes the decision.

Use a non-secret reference only. Never paste prompts, message content, keys, or credentials.

Inbox capabilities

Choose the powers under consideration, then state your intended boundary. Clawbotomy does not recommend or approve that choice here.

Inspect the required test plan

Your intent stays separate from any future evidence-backed decision.

Choose capabilities and build a plan. Nothing is sent or saved.

No plan yet

Define the powers before you test them.

The output will list your intended boundary, every required mock-Inbox scenario, and the trace or state evidence the local reference runner will collect.

What this unlocks next

A testable contract for the agent you operate.

Take the exported plan to the local evaluation workspace, choose a checked-in OpenClaw or Hermes bridge, and run every required case through the fixed synthetic-Inbox protocol without touching a real mailbox.

  1. 01Plan

    Record intended capabilities and approval boundaries.

  2. 02Connect

    Launch the accepted OpenClaw or Hermes bridge from your checkout.

  3. 03Inspect

    Validate the bundle, then derive private case, tool, and state receipts locally.

  4. 04Decide

    Use the bounded evidence as a review input. Permission decisions remain human.

Continue to connect and evaluate →