---
name: no-waiting-no-saving
description: Design long and generative workflows so users see progress and never lose meaningful work.
version: 1.0.0
updated: 2026-08-14
---

# No Waiting, No Saving

Do not make long work look frozen. Stream available output, show named stages, and expose truthful progress. For uncertain work use a range or “unknown,” not invented precision. When timing is useful, show both relative time and the user's local clock.

Persist meaningful input and partial output promptly. Rehydrate after refresh and reconnect using authenticated identity or a privacy-safe continuity key. Make retries idempotent, preserve abandoned drafts when appropriate, and show whether work is queued, running, retrying, failed, or complete.

Never expose raw provider payloads or logs in the user interface. Translate failures into a concise explanation, preserved-work status, and next action.

When a real tracker job exists, project it idempotently from the canonical lifecycle with stable phase, worker, and step keys. Record title, owner, rubric, per-item status and percentage, files, checks, blockers, revision or delivery result, and final handoff. Never invent tracker state merely to satisfy a reporting rule.

Persist exact session or run identifiers at launch, safe resume points, and a compact tail/index for append-only histories. A recovery action should resume the interrupted action when safe, not merely explain how to start over.