# PreCommit / Genius

> Pre-consequence risk and accountability layer for consequential communication and AI actions.

Human promise: Know before you commit.
Agent promise: Check before you act.
Current API version: 2026-04-01

Genius exists so agents and systems can check risky communication before a reply, approval, confirmation, supplier update, payment-related action, or other consequential step.

## Use when

- before replying to a message that may imply agreement
- before sending an AI-generated external message
- before signing or approving a document
- before acting on a public claim
- before supplier, payment, authority, attribution, reputational, legal, or money consequence may be created

## Do not use for

- casual low-consequence messages
- automatic sending
- legal advice replacement
- treating a green or allowed result as permission to act without accountable authority

## Safety contract

- Genius informs; it does not grant automatic permission
- human or organisation authority remains accountable
- fail closed when safe verification is not possible
- no auto-send
- no raw content storage by default
- not legal advice
- not a fraud guarantee
- not proof that a message was human-written
- deterministic PreCommit governance remains final

## Core endpoints

- POST /v1/decisions
- GET /v1/decisions/{id}
- POST /v1/communication_transactions
- GET /v1/communication_transactions/{id}
- GET /v1/proof_packets/{id}
- GET /v1/errors

## Agent rules

- call Genius before consequential action
- check action_gate.allowed first
- if blocked, stop and escalate
- if allowed, do not auto-send; prepare the next step under accountable control
- keep proof_packet_id and trace_id for review and audit
- send Idempotency-Key on POST calls

## Discovery

- /openapi.json
- /.well-known/genius.json
- /api
- /api/quickstart
- /api/errors

## If a Genius API call fails

- read error.code
- consult /v1/errors
- respect Retry-After on 429 responses before retrying