spec-brainstorm

Files (1)

• SKILL.md

SKILL.md (first lines)

# Spec Brainstorm

Conversational design workshop for substantial work that produces a focused, reviewed spec.

One question at a time. Multiple approaches explored. Design approved in
sections. Ruthless scope control. No implementation until design is approved.

Run this skill only after `atelier-orchestrator` selects a Spec-backed Plan or the human
explicitly requests a spec. Bounded work should go directly to `spec-plan` for an Inline
Plan. Do not reclassify the planning mode here.

## Artifact

```
docs/specs/YYYY-MM-DD-<feature>/
└── design.md  ← This skill's output
```

Requirements are inline — no separate requirements.json needed.

### Exclusive output contract

This skill may create or update only the `design.md` shown above. It may inspect the
repository and discuss drafts in conversation, but it must not modify any other file, create
`plan.json`, create tracker entries, invoke another workflow skill, or write implementation
code. This boundary still applies when the human asks to brainstorm, plan, and implement in
one request. Finish `design.md`, report it, and stop.

---

## Lessons

These principles apply to every spec, every time.

### Specs must earn their cost

Persisted specs are for work whose discovery, architecture, dependencies, or coordination
needs justify a durable artifact. Do not pull bounded work into this workflow merely because
it touches multiple files or takes time. That work belongs in an Inline Plan.

### Design for isolation and clarity

Break the system into units with one clear purpose each. Well-defined interfaces between
them. Each unit independently understandable and independently testable. If you can't
explain a unit's job in one sentence, it's doing too much. Apply **oracle-codebase-design**:
deep modules, seams placed where behaviour varies, and the interface as the test surface.

### Working in existing codebases

Explore the current structure first. Follow existing patterns. Targeted improvements only.
No unrelated refactoring. Understand why things are the way they are before proposing
changes. Treat loaded skills as relevant guidance, not as a requirement to apply every pattern
they contain.

### Decomposition

If the request describes multiple independent subsystems, flag it immediately. Decompose
into sub-projects before diving into details. Each substantial sub-project gets its own spec
and plan; bounded sub-projects can use Inline Plans. A spec that tries to cover three
subsystems helps no one.

Keep migrations separate from authorization, product behavior, infrastructure, schema, and
test-platform projects. Do not use a migration as permission to redesign adjacent systems.

### YAGNI ruthlessly

Remove unnecessary features from all designs. If a capability isn't needed for the first
user story, it doesn't go in the spec. Every feature is a cost — to build, to test,
to maintain, to understand later. Push back on scope creep during discovery.

Future consumers do not justify shared infrastructure. A "reusable foundation" may describe
an architectural quality, but it is not a user story or a current requirement.

---

## Step 1: Orient

Before diving in, understand where you are.

1. **Read project context** — AGENTS.md, README, existing architecture docs
2. **Check existing specs** — Scan `docs/specs/` for previous work. What domain model
   exists? What patterns are established? What has been built before?
3. **Read recent specs** — What was the last thing built? Is this feature building
   on existing work, extending it, or something greenfield?

This is silent — don't narrate it. Let the context inform where you focus.

---

## Step 2: Discovery

Ask questions to understand what to build. Skip this step if requirements are already
clear from context (existing specs, human provided details, etc.).

### Interview style

Ask **one question at a time**. Multiple choice preferred when possible — give 2-4
concrete options rather than open-ended prompts. Keep the conversation moving.

**Good:** "Should this be real-time or batch-processed? (a) Real-time via WebSocket,
(b) Periodic polling every 30s, (c) On-demand when user requests it."

**Bad:** "How should the data synchronization work?"

### When to skip discovery

- Human provides clear, detailed requirements
- Feature extends existing work with well-defined scope
- Human says "spec out X" and X is specific enough

### Decomposition check

Before asking any detail questions, assess scope. If the request describes multiple
independent subsystems (e.g., "build a notification system with email, SMS, push, and
an admin dashboard"):

1. **Flag it immediately:** "This looks like multiple independent projects. Let me
   propose a decomposition."
2. **Break it down:** Identify the subsystems and their dependencies.

From martinffx/atelier, MIT.

More from martinffx/atelier