openai-ultrafast-policylisted

# OpenAI Ultrafast Policy Use the deterministic `ultrafast-yolo` command. Treat Ultrafast as an access-controlled API preview, not a model name or a synonym for Fast/Priority. ## Workflow 1. Run `ultrafast-yolo doctor --json`. 2. If `ready` is false, keep the task on its existing subscription or local route. Do not ask for or expose a key; the command checks environment, the Hermes env file, and Keychain without printing values. 3. Run `ultrafast-yolo route --json --prompt "<task>"`. Only `latency-critical-high-value` work is eligible. Sensitive work stays local. 4. Run `ultrafast-yolo probe --json` only when the doctor shows an API key, contracted input/output rates, and available shared budget. 5. Run `ultrafast-yolo run --json --prompt "<task>" --max-output-tokens <n>` only after the same gates pass. 6. Claim Ultrafast execution only when the receipt has `exactServerTierConfirmed: true` and `serviceTier: "ultrafast"` from the server response. ## Hard rules - Never use ChatGPT or Codex OAuth tokens as `OPENAI_API_KEY`. - Never invent Ultrafast pricing. The preview requires `OPENAI_ULTRAFAST_INPUT_USD_PER_M` and `OPENAI_ULTRAFAST_OUTPUT_USD_PER_M` from the account's contract. - Never set Ultrafast globally for ordinary traffic. Restrict it to urgent, high-value interactive work. - Never bypass a failed reservation. The command aggregates tracked OpenRouter, GLM, and Ultrafast spend under one $10 monthly ceiling and fails closed on lock contention. - Never store prompt t