msd-core/get-shit-done/workflows/create-roadmap.md

Define the phases of implementation. Each phase is a coherent chunk of work that delivers value. The roadmap provides structure, not detailed tasks.

<required_reading> Read these files NOW:

1. ~/.claude/get-shit-done/templates/roadmap.md
2. ~/.claude/get-shit-done/templates/state.md
3. Read .planning/PROJECT.md if it exists </required_reading>

```bash cat .planning/PROJECT.md 2>/dev/null || echo "No brief found" ```

If no brief exists: Ask: "No brief found. Want to create one first, or proceed with roadmap?"

If proceeding without brief, gather quick context:

• What are we building?
• What's the rough scope?

Scan for available domain expertise:

ls ~/.claude/skills/expertise/ 2>/dev/null

Inference: Based on the brief/user request, infer applicable domains:

Keywords	Domain
"macOS", "Mac app", "menu bar", "AppKit"	expertise/macos-apps
"iPhone", "iOS", "iPad", "mobile app"	expertise/iphone-apps
"Unity", "game", "C#", "3D game"	expertise/unity-games
"MIDI", "sequencer", "music app"	expertise/midi
"ISF", "shader", "GLSL", "visual effect"	expertise/isf-shaders
"UI", "design", "frontend", "Tailwind"	expertise/ui-design
"Agent SDK", "Claude SDK", "agentic"	expertise/with-agent-sdk

If domain inferred:

Detected: [domain] project → expertise/[name]
Include this domain expertise? (Y / see options / none)

If multiple domains apply (e.g., ISF shaders for a macOS app):

Detected multiple domains:
- expertise/isf-shaders (shader development)
- expertise/macos-apps (native app)

Include both? (Y / select one / none)

If no domain obvious:

Available domain expertise:
1. macos-apps
2. iphone-apps
[... others found ...]

N. None - proceed without domain expertise

Select (comma-separate for multiple):

Store selected paths for inclusion in ROADMAP.md.

Derive phases from the actual work needed.

Check depth setting:

cat .planning/config.json 2>/dev/null | grep depth

<depth_guidance> Depth controls compression tolerance, not artificial inflation.

Depth	Typical Phases	Typical Plans/Phase	Tasks/Plan
Quick	3-5	1-3	2-3
Standard	5-8	3-5	2-3
Comprehensive	8-12	5-10	2-3

Key principle: Derive phases from actual work. Depth determines how aggressively you combine things, not a target to hit.

• Comprehensive auth system = 8 phases (because auth genuinely has 8 concerns)
• Comprehensive "add favicon" = 1 phase (because that's all it is)

For comprehensive depth:

• Don't compress multiple features into single phases
• Each major capability gets its own phase
• Let small things stay small—don't pad to hit a number
• If you're tempted to combine two things, make them separate phases instead

For quick depth:

• Combine related work aggressively
• Focus on critical path only
• Defer nice-to-haves to future milestones </depth_guidance>

Phase Numbering System:

Use integer phases (1, 2, 3) for planned milestone work.

Use decimal phases (2.1, 2.2) for urgent insertions:

• Decimal phases inserted between integers (2.1 between 2 and 3)
• Mark with "(INSERTED)" in phase title
• Created when urgent work discovered after planning
• Examples: bugfixes, hotfixes, critical patches

When to use decimals:

• Urgent work that can't wait for next milestone
• Critical bugs blocking progress
• Security patches needing immediate attention
• NOT for scope creep or "nice to haves" (capture with /gsd:add-todo instead)

Phase execution order: Numeric sort: 1 → 1.1 → 1.2 → 2 → 2.1 → 3

Deriving phases:

1. List all distinct systems/features/capabilities required
2. Group related work into coherent deliverables
3. Each phase should deliver ONE complete, verifiable thing
4. If a phase delivers multiple unrelated capabilities: split it
5. If a phase can't stand alone as a complete deliverable: merge it
6. Order by dependencies

Good phases are:

• Coherent: Each delivers one complete, verifiable capability
• Sequential: Later phases build on earlier
• Independent: Can be verified and committed on its own

Common phase patterns:

• Foundation → Core Feature → Enhancement → Polish
• Setup → MVP → Iteration → Launch
• Infrastructure → Backend → Frontend → Integration

**For each phase, determine if research is likely needed.**

Scan the brief and phase descriptions for research triggers:

<research_triggers> Likely (flag the phase):

Trigger Pattern	Why Research Needed
"integrate [service]", "connect to [API]"	External API - need current docs
"authentication", "auth", "login", "JWT"	Architectural decision + library choice
"payment", "billing", "Stripe", "subscription"	External API + compliance patterns
"email", "SMS", "notifications", "SendGrid", "Twilio"	External service integration
"database", "Postgres", "MongoDB", "Supabase"	If new to project - setup patterns
"real-time", "websocket", "sync", "live updates"	Architectural decision
"deploy", "Vercel", "Railway", "hosting"	If first deployment - config patterns
"choose between", "select", "evaluate", "which"	Explicit decision needed
"AI", "OpenAI", "Claude", "LLM", "embeddings"	Fast-moving APIs - need current docs
Any technology not already in codebase	New integration
Explicit questions in brief	Unknowns flagged by user

Unlikely (no flag needed):

Pattern	Why No Research
"add button", "create form", "update UI"	Internal patterns
"CRUD operations", "list/detail views"	Standard patterns
"refactor", "reorganize", "clean up"	Internal work
"following existing patterns"	Conventions established
Technology already in package.json/codebase	Patterns exist

</research_triggers>

For each phase, assign:

• Research: Likely ([reason]) + Research topics: [what to investigate]
• Research: Unlikely ([reason])

Important: These are hints, not mandates. The mandatory_discovery step during phase planning will validate.

Present research assessment:

Research needs detected:

Phase 1: Foundation
  Research: Unlikely (project setup, established patterns)

Phase 2: Authentication
  Research: Likely (new system, technology choice)
  Topics: JWT library for [stack], session strategy, auth provider options

Phase 3: Stripe Integration
  Research: Likely (external API)
  Topics: Current Stripe API, webhook patterns, checkout flow

Phase 4: Dashboard
  Research: Unlikely (internal UI using patterns from earlier phases)

Does this look right? (yes / adjust)

```bash cat .planning/config.json 2>/dev/null ``` Note: Config may not exist yet (project initialization). If missing, default to interactive mode. ``` ⚡ Auto-approved: Phase breakdown ([N] phases)

1. [Phase name] - [goal]
2. [Phase name] - [goal]
3. [Phase name] - [goal]

Proceeding to research detection...

Proceed directly to detect_research_needs step.
</if>

<if mode="interactive" OR="missing OR custom with gates.confirm_phases true">
Present the phase breakdown inline:

"Here's how I'd break this down:

1. [Phase name] - [goal]
2. [Phase name] - [goal]
3. [Phase name] - [goal]
   ...

Does this feel right? (yes / adjust)"

If "adjust": Ask what to change, revise, present again.
</step>

<step name="decision_gate">
<if mode="yolo">

⚡ Auto-approved: Create roadmap with [N] phases

Proceeding to create .planning/ROADMAP.md...

Proceed directly to create_structure step.
</if>

<if mode="interactive" OR="missing OR custom with gates.confirm_roadmap true">
Use AskUserQuestion:

- header: "Ready"
- question: "Ready to create the roadmap, or would you like me to ask more questions?"
- options:
  - "Create roadmap" - I have enough context
  - "Ask more questions" - There are details to clarify
  - "Let me add context" - I want to provide more information

Loop until "Create roadmap" selected.
</step>

<step name="create_structure">
```bash
mkdir -p .planning/phases

Use template from `~/.claude/get-shit-done/templates/roadmap.md`.

Initial roadmaps use integer phases (1, 2, 3...). Decimal phases added later via /gsd:insert-phase command (if it exists).

Write to .planning/ROADMAP.md with:

• Domain Expertise section (paths from detect_domain step, or "None" if skipped)
• Phase list with names and one-line descriptions
• Dependencies (what must complete before what)
• Research flags (from detect_research_needs step):
  • Research: Likely ([reason]) with Research topics: for flagged phases
  • Research: Unlikely ([reason]) for unflagged phases
• Status tracking (all start as "not started")

Create phase directories:

mkdir -p .planning/phases/01-{phase-name}
mkdir -p .planning/phases/02-{phase-name}
# etc.

Create STATE.md — the project's living memory.

Use template from ~/.claude/get-shit-done/templates/state.md.

Write to .planning/STATE.md:

# Project State

## Project Reference

See: .planning/PROJECT.md (updated [today's date])

**Core value:** [Copy Core Value from PROJECT.md]
**Current focus:** Phase 1 — [First phase name]

## Current Position

Phase: 1 of [N] ([First phase name])
Plan: Not started
Status: Ready to plan
Last activity: [today's date] — Project initialized

Progress: ░░░░░░░░░░ 0%

## Performance Metrics

**Velocity:**
- Total plans completed: 0
- Average duration: —
- Total execution time: 0 hours

**By Phase:**

| Phase | Plans | Total | Avg/Plan |
|-------|-------|-------|----------|
| — | — | — | — |

**Recent Trend:**
- Last 5 plans: —
- Trend: —

## Accumulated Context

### Decisions

Decisions are logged in PROJECT.md Key Decisions table.
Recent decisions affecting current work:

(None yet)

### Pending Todos

None yet.

### Blockers/Concerns

None yet.

## Session Continuity

Last session: [today's date and time]
Stopped at: Project initialization complete
Resume file: None

Key points:

• Project Reference points to PROJECT.md for full context
• Claude reads PROJECT.md directly for requirements, constraints, decisions
• This file will be read first in every future operation
• This file will be updated after every execution

Commit project initialization (brief + roadmap + state together):

git add .planning/PROJECT.md .planning/ROADMAP.md .planning/STATE.md
git add .planning/phases/
# config.json if exists
git add .planning/config.json 2>/dev/null
git commit -m "$(cat <<'EOF'
docs: initialize [project-name] ([N] phases)

[One-liner from PROJECT.md]

Phases:
1. [phase-name]: [goal]
2. [phase-name]: [goal]
3. [phase-name]: [goal]
EOF
)"

Confirm: "Committed: docs: initialize [project] ([N] phases)"

``` Project initialized: - Brief: .planning/PROJECT.md - Roadmap: .planning/ROADMAP.md - State: .planning/STATE.md - Committed as: docs: initialize [project] ([N] phases)

---

▶ Next Up

Phase 1: [Name] — [Goal from ROADMAP.md]

/gsd:plan-phase 1

<sub>/clear first → fresh context window</sub>

---

Also available:

• /gsd:discuss-phase 1 — gather context first
• /gsd:research-phase 1 — investigate unknowns
• Review roadmap

---

</step>

</process>

<phase_naming>
Use `XX-kebab-case-name` format:
- `01-foundation`
- `02-authentication`
- `03-core-features`
- `04-polish`

Numbers ensure ordering. Names describe content.
</phase_naming>

<anti_patterns>
- Don't add time estimates
- Don't create Gantt charts
- Don't add resource allocation
- Don't include risk matrices
- Don't impose arbitrary phase counts (let the work determine the count)

Phases are buckets of work, not project management artifacts.
</anti_patterns>

<success_criteria>
Roadmap is complete when:
- [ ] `.planning/ROADMAP.md` exists
- [ ] `.planning/STATE.md` exists (project memory initialized)
- [ ] Phases defined with clear names (count derived from work, not imposed)
- [ ] **Research flags assigned** (Likely/Unlikely for each phase)
- [ ] **Research topics listed** for Likely phases
- [ ] Phase directories created
- [ ] Dependencies noted if any
- [ ] Status tracking in place
</success_criteria>