Yanez Agent Authorization
View as Markdown Open in Claude Open in ChatGPT llms.txt

Agent Authorization

Let an AI agent request verifiable human approval for a sensitive action, and let the action’s executor (the relying party) verify the signed receipt before anything runs.

The HTTP/OpenAPI contract is the source of truth. Every SDK, the CLI, the MCP server, and the skill are adapters over that same contract — pick the highest layer your runtime supports.

The one rule that matters

A receipt authorizes nothing by itself. The action executor (the relying party, not the agent) must verify both signatures — Yanez's, and the approver's own — compare the signed terms against the proposed action, apply its own freshness and assurance policy, and consume single-use receipts.

How it works

sequenceDiagram
    participant A as Agent
    participant Y as Yanez Pulse
    participant U as User (YID app)
    participant R as Relying party
    A->>Y: 1. POST /api/agent/authorizations (yak_ key, terms)
    Y-->>U: 2. Push notification
    U->>Y: 3. Approve or reject, gated on biometrics
    A->>Y: 4. GET /api/agent/authorizations/{id}?wait=25
    Y-->>A: approved + signed receipt, or rejected / expired
    A->>R: Proposed action + receipt
    R->>Y: 5. GET /api/authz/public-keys, POST /api/authz/introspect
    R->>R: Verify, consume, then act
  1. The agent creates a request. It sends the exact terms to Yanez Pulse with its yak_ agent API key. The key can ask, not act.
  2. Yanez Pulse notifies the user. A push notification reaches the user’s device.
  3. The user decides. They approve or reject in the YID app, gated on a fresh biometric scan. Their own key signs the decision, and approval produces a receipt carrying both that signature and Yanez’s: user-signed approvals.
  4. The agent polls for the decision. It long-polls the request until it is approved (with the receipt), rejected, or expired.
  5. The relying party checks the receipt. The action executor verifies Yanez’s signature offline against Yanez’s public keys and the approver’s signature against the key inside the receipt, compares the signed terms with the proposed action, consumes the receipt when the action is single-use, and only then acts.

Start here

Two ways in. Pick one.

Let an AI agent integrate it
Paste one prompt into Claude Code, Cursor, Codex, or a chat assistant. It reads llms.txt and wires the integration in.
Integrate it yourself
Python, TypeScript, CLI + skill, MCP, or raw HTTP. Pick the highest layer your runtime supports, then follow its quickstart.

Understand the model

Terms
The object the human actually approves, field by field.
Receipts
What the signed artifact contains, how it is signed, and how its keys rotate.
User-signed approvals
The approver's own signature: what changed in the schema, and the steps to verify both signatures.
Action enforcement
The contract for the boundary where a receipt is turned into an action.