| name | devmd-guide |
| description | Interactive guide for writing DevMD files from scratch. Walks through files one by one in dependency order, asking the right questions to fill each template. |
| triggers | ["devmd guide","devmd init","devmd start","devmd 시작","devmd 가이드"] |
| user-invocable | true |
| allowed-tools | ["Read","Write","Edit","AskUserQuestion","Glob","Grep","Bash"] |
DevMD Guide
Interactive walkthrough for creating DevMD files from scratch for a new project.
Arguments
/devmd-guide # Start from scratch (auto-detect or ask tier)
/devmd-guide app # Start with App tier (16 files)
/devmd-guide SCHEMA.md # Write a single specific file
/devmd-guide --resume # Continue where you left off (check which files exist)
Step 0: Setup
0a. Detect existing state
for f in PRODUCT GLOSSARY BRAND ARCHITECTURE SCHEMA API ERRORS LOGGING \
DESIGN UI SCREENS FLOWS SEO TESTING SECURITY INFRA CONFIG DEVOPS \
OPERATIONS CHANGELOG CLAUDE AGENTS HARNESS RUNTIME LIFECYCLE; do
[ -f "${f}.md" ] && echo "EXISTS: ${f}.md"
done
If files already exist, show what's done and what's remaining. Offer to continue from next file.
0b. Ask tier (if not specified)
Which tier fits your project?
| Tier | Files | Best For |
|------|-------|----------|
| Starter (8) | PRODUCT, BRAND, CLAUDE, DESIGN, UI, SCREENS, SEO, INFRA | Landing pages, static sites, portfolios |
| App (16) | + GLOSSARY, ARCHITECTURE, SCHEMA, API, ERRORS, LOGGING, FLOWS, TESTING, SECURITY | SaaS, web apps, APIs |
| AI-Native (21) | + AGENTS, HARNESS, LIFECYCLE, RUNTIME, CONFIG | AI-powered products, agent systems |
| Enterprise (25) | + DEVOPS, OPERATIONS, CHANGELOG, remaining | Full teams at scale |
0c. File selection (filter)
Tier gives a starting list, but not every file is needed for every project. After tier selection, present the full file list with per-file relevance hints based on project characteristics, and let the user deselect files that don't apply.
Relevance hints (auto-detect from Step 0a/0b context + ask if unclear):
| Signal | Files to question |
|---|
| Internal tool / no public traffic | SEO — likely skip |
| No AI agent orchestration (just LLM API calls) | AGENTS, HARNESS, LIFECYCLE — likely skip |
| Solo dev / small team, no on-call | OPERATIONS — likely skip |
| No CI/CD pipeline yet / simple deploy | DEVOPS — likely skip |
| Greenfield project (no changelog yet) | CHANGELOG — likely skip |
| Background jobs described in FLOWS | RUNTIME — may be redundant |
Process:
- Show the tier's full file list as a table:
| File | Wave | Purpose (one line) | Recommended? |
- Mark each file as ✓ Recommended or ? Review based on the signals above.
- Ask via AskUserQuestion (multiSelect): "Which files should we skip? (deselect = skip)"
- Store the final selected list. Only these files are generated in subsequent waves.
- Skipped files are noted in the Post-Completion summary.
If the user specifies --all, skip this step and generate every file in the tier.
0d. Create output directory
Files are written to the current directory (project root). Read the corresponding template from the DevMD plugin's templates/ directory for each file.
Writing Order
Files are written in 7 waves. Each wave builds on the previous. Within a wave, order doesn't matter — but complete one wave before starting the next.
Wave 1: "What are we building?" (Every tier)
PRODUCT.md → BRAND.md → CLAUDE.md
These three answer: what is this, who is it for, and how do we talk about it.
Wave 2: "How does the data work?" (App+ tiers)
GLOSSARY.md → SCHEMA.md → ERRORS.md
Domain terms first (GLOSSARY defines the vocabulary), then data structure, then what can go wrong.
Wave 3: "How is it structured?" (App+ tiers)
ARCHITECTURE.md → API.md → LOGGING.md
System shape, interface contract, observability.
Wave 4: "How does it look and flow?" (Every tier)
DESIGN.md → UI.md → SCREENS.md → FLOWS.md → SEO.md
Visual tokens, component structure, screen catalog, user journeys, search strategy.
Wave 5: "How is it secured and tested?" (App+ tiers)
SECURITY.md → TESTING.md
Wave 6: "How does it run?" (Varies by tier)
INFRA.md → CONFIG.md → DEVOPS.md → OPERATIONS.md → CHANGELOG.md
Wave 7: "How does the AI work?" (AI-Native+ tiers)
AGENTS.md → HARNESS.md → RUNTIME.md → LIFECYCLE.md
Per-File Writing Process
For each file, follow this exact sequence:
1. Show context
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Wave {N}/{total} — {FILE_NAME}
{one-line purpose from spec}
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
This file answers: {core question}
It connects to:
← reads from: {previous files that inform this one}
→ feeds into: {future files that will use this one}
2. Ask key questions
Each file has 3–5 key questions. Ask via AskUserQuestion. Questions are designed to extract the REQUIRED frontmatter fields and body content.
Do NOT ask all 25 files' questions upfront. Ask one file at a time.
3. Generate the file
MUST read the spec BEFORE generating. Read spec/{FILE_NAME} from the DevMD plugin directory to understand:
- REQUIRED frontmatter fields (types, constraints)
- REQUIRED body sections (order, content rules)
- Cross-reference requirements (SHOULD/MUST)
- Validation rules (e.g., DESIGN.md colors must be #hex)
Then read the template from DevMD templates/{FILE_NAME}. Fill in the user's answers, ensuring all REQUIRED fields and sections from the spec are present. Write the file.
4. Show and confirm
Show a 5-line summary of what was written. Ask: "Good to move on, or want to edit?"
5. Move to next file
Key Questions Per File
PRODUCT.md
- What is this product in one sentence?
- Who is the primary user? What pain do they have?
- What exists today that they use instead? Why is that not good enough?
- How will you know this product is successful? (1–2 metrics)
BRAND.md
- What tone should the product use? (casual / professional / playful / technical)
- Give an example sentence that sounds like your product.
- Any words or phrases to always avoid?
CLAUDE.md
- What language/framework? (e.g., TypeScript + Next.js, Python + FastAPI)
- Any strict coding rules? (e.g., no
any, always use absolute imports)
- Key commands? (dev, build, test, deploy)
GLOSSARY.md
- What are the 5–10 most important domain terms? For each: what does it mean, and what do people mistakenly call it?
- Are there terms that MUST NOT be used? (e.g., "user" → always say "member")
SCHEMA.md
- What are the main entities/tables? (e.g., User, Organization, Project, Task)
- For each: what are the key fields (not every field — the important ones)?
- What are the main relationships? (User has many Projects, Project belongs to Organization)
- What database? (PostgreSQL, MySQL, SQLite, MongoDB)
ARCHITECTURE.md
- Monolith or microservices? Monorepo or polyrepo?
- What are the main layers? (e.g., API → Service → Repository → DB)
- Any firm dependency rules? (e.g., "domain layer must not import infrastructure")
API.md
- REST, GraphQL, tRPC, or mixed?
- What authentication method? (JWT, session, API key)
- What are the main resource groups? (e.g., /users, /projects, /billing)
ERRORS.md
- What error format? (Show a JSON example or describe)
- Any custom error codes? (e.g., AUTH_001, BILLING_002)
- Retry policy for failed operations?
LOGGING.md
- What log format? (JSON structured, plain text)
- What log levels do you use?
- Any tracing/observability tools? (OpenTelemetry, Datadog, Sentry)
DESIGN.md
- What UI framework? (Tailwind, CSS Modules, Styled Components, etc.)
- Primary colors? (or "I don't know yet" is fine)
- Dark mode needed?
- Any design system reference? (existing Figma, Storybook, or "from scratch")
UI.md
- What are the main pages/screens? (e.g., Dashboard, Settings, Profile, List view)
- What layout pattern? (sidebar + main, top nav + content, single column)
- Any reusable component patterns? (card grid, data table, form wizard)
SCREENS.md
- For each main page from UI.md: what does the user see in the default state?
- Any important empty states, loading states, or error states to define?
FLOWS.md
- What are the 3–5 most important user journeys? (e.g., signup → onboarding → first action)
- For each: what happens step by step (user action → system response)?
SEO.md
- Does this product need to be found via search? If so, what keywords matter?
- Any AI search (GEO) considerations? (e.g., should LLMs be able to cite your content?)
- SSR or CSR?
SECURITY.md
- What auth method? (email+password, OAuth, SSO, magic link)
- Any role-based access control? What roles?
- Any compliance requirements? (SOC2, GDPR, HIPAA)
TESTING.md
- What test frameworks? (Jest, Vitest, Playwright, pytest)
- What coverage target?
- Any E2E tests planned?
INFRA.md
- Where will this run? (Vercel, AWS, GCP, self-hosted, Cloudflare)
- What database hosting?
- Any specific infrastructure constraints? (budget, region, compliance)
CONFIG.md
- What environment variables are needed? (DB URL, API keys, feature flags)
- How many environments? (dev, staging, prod)
DEVOPS.md
- What CI/CD? (GitHub Actions, GitLab CI, none yet)
- Any deploy automation?
- Branch strategy? (trunk-based, gitflow, simple main+feature)
OPERATIONS.md
- Who gets paged when something breaks?
- Any SLO targets? (99.9% uptime, <200ms p95 latency)
- Incident response process?
CHANGELOG.md
- What versioning? (semver, calver, just dates)
- Any breaking change policy?
AGENTS.md
- What AI agents will this product have? Name and one-line purpose for each.
- What tools/skills can each agent use?
HARNESS.md
- What LLM provider? (OpenAI, Anthropic, local)
- Any RAG setup? What knowledge sources?
- Any guardrails or safety rules?
RUNTIME.md
- Any background jobs/workers? What do they do?
- Any cron/scheduled tasks?
- Any real-time features? (WebSocket, SSE)
LIFECYCLE.md
- What are the main states this system can be in? (e.g., setup → active → suspended → archived)
- What triggers transitions between states?
Resume Mode
When --resume is used:
- Run Step 0a to find existing files
- Determine tier from what's present (or ask)
- Find the first missing file in the wave order
- Start from that file
Single-File Mode
When invoked as /devmd-guide SCHEMA.md:
- Read any existing files that SCHEMA.md depends on (GLOSSARY.md, PRODUCT.md) for context
- Ask only SCHEMA.md questions
- Write only SCHEMA.md
Post-Wave Spec Compliance Check
After ALL waves are complete (or after each wave if --strict flag is set), run a spec compliance check on every generated file:
Process
- For each generated file, read the corresponding spec from the DevMD plugin's
spec/{FILE_NAME} directory.
- Validate:
a. REQUIRED frontmatter fields: every field marked REQUIRED in the spec exists and has a value of the correct type
b. REQUIRED body sections: every section marked REQUIRED in the spec exists and has content (not empty)
c. Cross-references: every
@FILE.md reference marked SHOULD or MUST in the spec is present
d. Validation rules: any spec-specific rules (e.g., DESIGN.md colors must be #hex, PRODUCT.md tagline max 120 chars)
- Report as a compliance table:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Spec Compliance Check — {N} files
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
| File | Frontmatter | Sections | Cross-refs | Status |
|------|-------------|----------|------------|--------|
| PRODUCT.md | 6/6 ✓ | 5/5 ✓ | 2/3 | PASS |
| DESIGN.md | 2/5 ✗ | 1/2 ✗ | 0/3 | FAIL |
...
FAIL: {N} files need fixes. Fix now? (Y/n)
- If any file FAILs:
- Re-read the spec
- Read the generated file
- Add missing REQUIRED fields/sections
- Do NOT remove existing content — only add what's missing
- Re-run the check until all files PASS.
Why this step exists
Agent-generated files frequently miss spec-mandated structure (frontmatter maps, required sections, cross-references) even when the spec is available. This check catches gaps before the developer starts building from these docs.
Post-Completion
After all files pass spec compliance:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
DevMD Complete — {tier} tier ({N} files)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Written: {list of files}
Spec compliance: {N}/{N} PASS
What's next:
/devmd-gap-analysis --consistency-only → check for conflicts between files
/pdca plan {feature} → start building a feature using these specs
/devmd-scan --overwrite → later, regenerate from code to keep specs in sync