ĀML™ / ACTION LAB
PROPOSE · POLICY · APPROVAL · RECEIPT
What exactly did the agent ask to do?
Edit a proposed tool call. ĀML's reference contract hashes its exact meaning and asks a host-owned policy for an exact match. Try Replay one-use grant : approve once, rehearse dispatch twice, and watch the second attempt block.
Browser rehearsal only. No tool runs, no message is sent, and the button does not obtain a person's consent. The visible policy is a fixed local fixture.
01 · Send needs approval
02 · Different destination
03 · Allowed read
04 · Uncertain result
05 · Replay one-use grant
THE ACTION PATH Changes with the local scenario and your edits
01 / AGENT PROPOSAL Calculating 02 / HOST POLICY Calculating 03 / SIMULATED APPROVAL Calculating 04 / REHEARSAL OUTCOME Not attempted
This diagram shows a project-authored browser rehearsal. A green simulated path is not tool execution, delivery, or human consent.
Can an outside implementation make the same decision? Explore the 13-case challenge ↗
01 / Proposed action
These fields are the agent's declared intent. The host must independently map the actual tool and destination.
A small JSON object. This pilot's cross-runtime vectors cover a limited JSON subset.
Inspect the host policy fixture
02 / Inspect the boundary
Each edit recalculates the SHA-256 digest and the exact rule decision.
POLICY DECISION
Calculating
PROPOSED ACTION SHA-256
Calculating…
Simulate host approval Rehearse dispatch Copy local report
No simulated approval recorded.
SIMULATED OUTCOME
Not attempted
CANONICAL PROPOSAL AND REPORT
01 · Exact match An undeclared destination or effect is denied by the local fixture.
02 · Approval binding and replay Editing arguments invalidates the digest. In the replay scenario, a spent local grant blocks the second attempt.
03 · Honest outcome A simulated callback error is unknown. It cannot prove an external action happened or did not happen.