draft-spec

# Draft Spec

Guide a collaborative discussion to explore a project idea, then synthesize the conversation into a comprehensive specification at `.turbo/specs/<slug>.md`.

## Task Tracking

At the start, use `TaskCreate` to create a task for each step:

1. Capture the vision and pick a slug
2. Consult task-specific skills and docs
3. Deep-dive discussion
4. Draft the spec
5. Resolve open questions
6. Present and finalize

## Step 1: Capture the Vision and Pick a Slug

Absorb whatever the user has provided — a sentence, a paragraph, a brain dump. Do not interrupt or ask questions yet. Restate the vision back in two or three sentences to confirm understanding.

Pick a slug for the spec file derived from the project or feature name:

- Lowercase
- Replace non-alphanumeric characters with hyphens
- Collapse consecutive hyphens
- Trim leading and trailing hyphens
- Truncate to 40 characters at a word boundary

Example: "Photo Sorter v2" → `photo-sorter-v2`. The user may pass an explicit slug; if so, honor it.

If `.turbo/specs/<slug>.md` already exists, use `AskUserQuestion` to ask whether to overwrite, append a numeric suffix (`-2`, `-3`, ...), or pick a different slug.

State the chosen slug and the resulting spec path before continuing.

Then use `AskUserQuestion` to ask 1-4 focused opening questions targeting the biggest unknowns. Skip anything the user already answered. Prioritize from:

- What problem does this solve, and for whom?
- Is this greenfield or does existing code/infrastructure exist?
- Are there strong technology preferences or constraints?
- What does the MVP look like versus the full vision?
- Are there hard deadlines, budget limits, or team size constraints?

## Step 2: Consult Task-Specific Skills and Docs

Ground architecture and tech-stack choices in current reality before the deep-dive discussion.

1. **Scan for matching skills.** Compare the vision and opening-question answers against available skill trigger descriptions. For each unambiguous match, run the skill via the Skill tool. This loads decision-level guidance (idiomatic patterns, known pitfalls, version constraints) before architectural choices are made. If unsure, do not load.
2. **Look up library or framework docs.** For any library, framework, or platform the user mentioned or the project clearly needs, query documentation MCP tools (or WebSearch as a fallback) when the decision hinges on current capabilities, supported versions, or known constraints.

Keep findings at the decision level: what tools can do, which approaches are idiomatic, which versions to target. Do not embed specific API signatures or code snippets into the spec. Those belong in implementation-time skill loads.

## Step 3: Deep-Dive Discussion

Interview the user relentlessly about every aspect of the project until you reach shared understanding. Gather behavioral requirements (the "what") before architectural design (the "how"), so design decisions land against a concrete set of requirements instead of being taken in the abstract. Track coverage internally but do not present the list as a rigid checklist. When the user jumps to architecture early, engage briefly then circle back to confirm the behavioral picture is complete.

### Requirements (gather first)

| Category | What to explore |
|---|---|
| **Users and personas** | Who uses this? Goals, pain points, technical sophistication |
| **Core behaviors** | Primary capabilities and user-facing workflows — the behaviors the system must exhibit |
| **Non-functional requirements** | Performance, security, accessibility, i18n, compliance |

### Design (gather after requirements are clear)

| Category | What to explore |
|---|---|
| **Architecture** | Client/server split, monolith vs services, real-time needs, offline support |
| **Tech stack** | Languages, frameworks, databases, hosting — preferences and constraints |
| **Data model** | Key entities, relationships, storage strategy |
| **Integrations** | Third-party APIs, auth providers, external data sources |

### Cross-cutting

| Category | What to explore |
|---|---|
| **MVP scope** | What ships first? What is explicitly deferred? |
| **Open questions** | Unknowns needing research, prototyping, or external input |

### Discussion Guidelines

- If a question can be answered by exploring the codebase, explore the codebase instead
- Use `AskUserQuestion` to ask one question at a time. Use options with descriptions to frame trade-offs and offer concrete suggestions. Use `multiSelect` when choices are not mutually exclusive.
- When the user gives a short answer, probe deeper before moving on
- Offer concrete suggestions and trade-off analysis — be a collaborator, not an interviewer
- For each question, recommend an answer with reasoning
- Walk down each branch of the design tree, resolving dependencies between decisions one-by-one
- When all categories have sufficient depth or the user signals readiness, confirm before moving to drafting

## Step 4: Draft the Spec

Synthesize the consulted skill and doc context plus the entire discussion into `.turbo/specs/<slug>.md` using the slug picked in Step 1. Use the fixed skeleton below.

````markdown
# <Project or Feature Name>

## Overview

<One or two paragraphs stating the problem being solved and the vision for the solution.>

## Users

<Personas and their goals. Omit this section if the project has no meaningful user role distinction (e.g., a single-integrator internal library).>

## Requirements

Enumerated behavioral requirements with stable IDs. Number requirements `R1`, `R2`, `R3`, ... IDs must stay stable once drafted so downstream artifacts can reference them reliably.

Pick either format per requirement; both can appear in the same spec.

**EARS format** — for unambiguous, testable behaviors:

- **R1.** When <trigger or condition>, the system shall <expected behavior>.

Adapt the slot to the EARS pattern that fits: `When` (event-driven), `While` (state-driven), `Where` (optional featur