| name | tutorial-authoring |
| description | Author hands-on tutorials that take readers from zero to a working result. Use when producing a new tutorial, getting-started guide, onboarding doc, or restructuring docs that drifted from tutorial into reference. Trigger terms: tutorial, getting started, onboarding, step-by-step guide, walkthrough, hands-on, learning guide. |
Tutorial Authoring
A tutorial teaches. It holds the reader's attention across many steps, surfaces only what's needed at each step, and leaves the reader with both a working result and a model of why it worked.
A tutorial is not a README (orients in seconds), not a runbook (executed under time pressure), not a reference (looked up). It is a guided learning experience.
Spine and Reference Split
Every tutorial has two surfaces:
- Spine modules — the linear path. One module per logical step, ordered by dependency. Numbered prefixes (
00-mental-model.md, 01-environment.md). Each ends with the reader having done something concrete.
- Reference pages — lookups needed along the way. Datasets, config fields, troubleshooting. Visited on demand, not read in order.
Test: Imagine the reader doing the tutorial a second time. The thing they walk through is the spine. The thing they look up is the reference.
Module Structure
Every spine module has the same frame:
# N. Module Title (the outcome, not the activity)
> [← index](../README.md) · [← prev](NN-prev.md) · [next →](NN-next.md)
**What You'll Learn**: One sentence — the concrete outcome.
## Section heading
Prose. State the goal, then show the action.
\```bash
the-command --flag value
\```
What the command did. What the reader sees now.
⚠️ **Pitfall**: Concrete trap with symptom and recovery.
---
> [← index](../README.md) · [← prev](NN-prev.md) · [next →](NN-next.md)
Required Patterns
Module sizing. 80-180 lines. Below 80 → collapse with neighbour. Above 180 → extract a reference page.
Outcome-named slugs. Name by what the reader HAS at the end: 02-positive-samples.md (you have samples), not 02-recording-audio.md (you did recording).
Lazy linking via <details>. Collapse additive context at point of mention. Reader mid-action skips it; reader who pauses opens it without losing their place.
<details>
<summary>Other paths the script writes to</summary>
See [config-fields.md](../reference/config-fields.md#paths) for the full list.
</details>
Pitfall dual-storage. Each ⚠️ Pitfall appears:
- Inline at the moment of risk in the spine module
- Mirrored in
reference/troubleshooting.md indexed by symptom
Two reader populations: first-time linear walker vs. broke-something-now-searching.
Every step independently verifiable. Show expected output after each command. The reader must be able to confirm they're on track without proceeding to the next step.
Prerequisites stated explicitly. Exact versions, tools, prior knowledge. Never assume without stating.
Banned Patterns
- Reference manual masquerading as tutorial — tables and flag lists in a spine module. Extract to reference.
<details> hiding required steps — collapsed blocks are for additive context only. Required steps stay inline.- Forward links from module M to module N — invites reader to leave the spine. Foreshadow with one sentence; don't link.
- Assuming prior knowledge without stating it — if the reader needs to know X, say so in prerequisites.
- Decorative diagrams — every diagram answers a question the reader has at that point.
Tutorial vs How-To (Diátaxis Distinction)
Self-Check
- Can the reader complete without opening a reference page?
- Does every step show expected output?
- Are prerequisites complete and version-specific?
- Does each module fit one screenful of attention?
