# Authoring a service

*Describe the outcome you want in plain English, and Tredy assembles a governed, runnable service from a best-practice playbook.*

You don't build a Tredy service by wiring stages by hand. You **describe the outcome**,
and Tredy's Service Designer assembles a governed, runnable definition from a proven
**playbook** — then you refine it in an editor.

## How it works

1. **Describe it in plain English.** Write a one-paragraph description of the outcome you want.

2. **Tredy picks a playbook.** Your description is matched (lexically, then semantically) to a best-practice playbook — competitor intelligence, coverage / policy / security-control / regulatory / contract / vendor gap analysis, document analysis or comparison, regulatory monitoring, denials prevention. The playbook brings a proven tool chain, quality gates, and a deliverable format. Cross-cutting best-practice **skills** are always attached.

3. **The model drafts, code assembles.** An LLM proposes the inputs, agents, and stages; Tredy deterministically expands that into a strict service definition and overlays the playbook's governance. Only **read-only** capabilities are freely selectable — anything that exports or publishes is reachable **only through an approval-gated stage**, never freehand.

4. **Auto-lint, grade, and accept.** The draft is linted, graded by a reviewer model, and put through a deterministic **acceptance gate** (required capabilities, quality gates, review gates, deliverable sections). Tredy self-corrects up to twice, keeping a revision only if it genuinely improves.

5. **Refine in the Service Creator.** The result opens in the editor for you to adjust — nothing is persisted until you save.

## Lifecycle & governance

Once you save, a service moves through a **maker–checker** lifecycle, fully audited at
every step:

```
draft → validated → submitted → approved → published → installed
```

Publishing hot-registers the service with zero runtime changes, and every definition
change is recorded with a hash for provenance.

> **Tip:** Because side-effecting steps are always forced behind approval gates, an authored service is safe to run before you fully trust it — it can't publish or export without a human approving.