# private-intent-broker-ops

Use this skill when operating, verifying, extending, or judging Private Intent Broker from a deployed shell.

## Start here

1. Read `/submission-readiness.json`.
2. Read `/agent-manifest.json`.
3. If a backend is available, switch to `/api/agent/manifest` and follow its workflow.

## Product truth

- Do not claim Venice is sponsor-grade unless a live artifact shows a Venice private model response.
- Do not claim Uniswap is sponsor-grade unless there is a real transaction hash or order proof.
- Do not claim open-track publish readiness while the demo video is missing.

## Static shell workflow

1. Use the page as a sponsor-facing overview.
2. Read `/llms.txt` for concise discovery.
3. Use `/submission-readiness.json` to identify the proof gaps.
4. Treat the missing proof list as the source of truth.

## API-backed workflow

1. `GET /api/scenarios`
2. `POST /api/intent/analyze`
3. Inspect `trustSummary`, `riskMatrix`, `policyChecklist`, `executionPlan`, `evidencePlan`, and `judgePreview`
4. `GET /api/workbench/overview`
5. `POST /api/intent/execute`
6. `POST /api/judges/run`

## Judge posture

- Reward explicit blocker labeling.
- Penalize fake integrations, placeholder proof, or vague sponsor claims.
- Prefer the machine-readable artifacts over scraping prose from the UI.