Arch

Global Rules: Adheres to rules/archflow-rules.md. Code Mapping # Rule: Always use max(existing #) + 1 for new rows. NEVER reuse deleted numbers. Document Version Control: After changes, commit recommended. Message: docs({serviceName}): arch - {summary}. If git unavailable, skip. Req ID Parsing: Extract from spec.md ## 0. Requirement Summary table. Each Req ID maps to Code Mapping Spec Ref (1:N).

Model: Opus required. Design quality determines implementation quality.

Arch Workflow

Two perspective sub-agents collaborate to derive optimal design.

Tool Fallback

Tool	Alternative
Read/Grep	Request file path, ask for copy-paste
AskQuestion	"Please select: 1) A 2) B 3) C" format
Task	Use Self-Debate Pattern (see below)

Self-Debate Pattern (when Task unavailable)

Play two roles sequentially:

**Role A: Domain Architect** - Full project context, design that fits this project
  Considerations: Existing patterns, technical constraints, team familiarity

**Role B: Best Practice Advisor** - Feature requirements only, ideal design
  Considerations: Industry standards, scalability, maintainability

Phase 1: Role A provides design proposal

Phase 2 (Cross-Review): Switch to Role B.
  Review the Role A design and provide:
  1. What you agree with
  2. Concerns or risks
  3. Alternative suggestions
  4. Can we reach consensus? [Yes / Need discussion]

Phase 3 (Counter-Review): Return to Role A.
  Review Role B's feedback and provide:
  1. Valid points you can accept
  2. Points you must reject (with project-specific reasons)
  3. Proposed compromises
  4. Final stance: [Agreement / Need user decision]

Phase 4 (Synthesis): Synthesize both perspectives into final design.
  Document in "Risks & Tradeoffs":
  - Chosen approach
  - Rejected alternatives and why
  - Assumptions made
  - Points requiring user decision (if any)

When to use: Only when Task tool is unavailable.

Document Structure

docs/{serviceName}/
  spec.md       <- input
  arch-be.md    <- output (Backend)
  arch-fe.md    <- output (Frontend)
  ui.md         <- input (FE only, from /ui skill)

serviceName inference: Extracted from input file path docs/{serviceName}/spec.md

Role Definitions

Agent	Role	Context
domain-architect	Project context-based design	Requirements + project structure
best-practice-advisor	Best practice suggestions	Feature request only (no context)

Phase -1: Service Discovery

-1.1. Scan Docs Directory

1. List docs/ subdirectories
2. If 1 found (e.g., docs/auth): auto-select, confirm with user If multiple: ask user to select. If none: manual input.
3. Auto-resolve paths:
   • spec.md = docs/{serviceName}/spec.md
   • arch-be.md = docs/{serviceName}/arch-be.md
   • arch-fe.md = docs/{serviceName}/arch-fe.md
   • ui.md = docs/{serviceName}/ui.md

Phase 0: Skill Entry

0-0. Model Guidance (Display at start)

WARNING: This skill strongly recommends Opus model. Design quality determines implementation quality, so maintain Opus even when cost savings are needed.

Input: docs/{serviceName}/spec.md | Output: docs/{serviceName}/arch-be.md or arch-fe.md

0-1. Collect Input

If serviceName resolved in Phase -1: verify spec.md exists, proceed to 0-1.5. If not exists: "Requirements not found at docs/{serviceName}/spec.md. Run /spec or /reverse first."

If NOT resolved:

{"title":"Start Design Debate","questions":[{"id":"has_requirements","prompt":"Do you have a requirements document? (docs/{serviceName}/spec.md)","options":[{"id":"yes","label":"Yes - I will provide via @filepath"},{"id":"no","label":"No - Need requirements refinement first"}]}]}

• yes -> Request file path -> 0-1.5
• no -> Guide to spec skill

0-1.5. Select Architecture Type (BE/FE)

{"title":"Architecture Type","questions":[{"id":"arch_type","prompt":"What type of architecture are you designing?","options":[{"id":"be","label":"Backend - API server, business logic, database"},{"id":"fe","label":"Frontend - Web app, SPA, components"}]}]}

• be -> Read templates/be.md, use as output template
• fe -> Check for ui.md first, then read templates/fe.md

WARNING: MUST read the template file before proceeding to Phase 1.

0-1.6. Frontend Prerequisites (FE only)

Verify docs/{serviceName}/ui.md exists:

{"title":"UI Specification Check","questions":[{"id":"has_ui_spec","prompt":"Do you have a UI specification? (docs/{serviceName}/ui.md)","options":[{"id":"yes","label":"Yes - I will provide via @filepath"},{"id":"no","label":"No - I need to run /ui first"}]}]}

• yes -> Request ui.md path -> Proceed to Phase 1
• no -> Show guidance:

  WARNING: UI specification required for Frontend architecture. Run /ui first to generate UI specification from:

  • docs/{serviceName}/spec.md
  • docs/{serviceName}/arch-be.md

  The ui.md defines screens, components, and interactions needed for arch-fe.

NOTE: Frontend arch input = spec.md + ui.md (NOT arch-be.md directly).

0-1.7. Project Type

{"title":"Project Type","questions":[{"id":"project_type","prompt":"Is this a new project or existing project?","options":[{"id":"new","label":"New project - Start from scratch"},{"id":"existing","label":"Existing project - Has code already"}]}]}

• new -> 0-1.8 (Setup Strategy)
• existing -> 0-1.7a (Path Detection)

0-1.7a. Existing Project - Path Detection

Auto-detect by scanning for dependency files:

Language	Files to Check	Detected Path
Python	pyproject.toml, requirements.txt	Parent directory of file
Node.js	package.json	Parent directory of file

{"title":"Detected Project Paths","questions":[{"id":"path_confirm","prompt":"I found:\n- BE: {detected_be_path}\n- FE: {detected_fe_path}\n\nCorrect?","options":[{"id":"accept","label":"Yes - Use these paths"},{"id":"modify","label":"No - I will specify manually"}]}]}

• accept -> 0-1.7b | modify -> manual input -> 0-1.7b

0-1.7b. Existing Project - Auto-detect Dependencies

Read dependency files from detected paths. Extract:

• Installed packages with versions
• Package manager (from lock files: package-lock.json, yarn.lock, pnpm-lock.yaml, poetry.lock, uv.lock)
• Run commands (from scripts in package.json or pyproject.toml)

-> Proceed to 0-1.9

0-1.8. Setup Strategy (New Projects)

{"title":"Setup Strategy","questions":[{"id":"setup_strategy","prompt":"How would you like to configure the project?","options":[{"id":"recommend","label":"LLM Recommendation - AI proposes optimal setup based on requirements"},{"id":"manual","label":"Manual Selection - I will choose each option"}]}]}

• recommend -> 0-1.8a | manual -> 0-1.8b

0-1.8a. LLM Full Recommendation

Analyze spec.md and propose complete setup:

1. Project Structure (Monorepo / Monolith / Custom)
2. BE/FE Paths
3. Language, Framework, Database, ORM
4. Package Manager, Essential Libraries
5. Run Commands

{"title":"Proposed Project Setup","questions":[{"id":"full_proposal","prompt":"Based on analysis:\n\n**Structure**: {type}\n- BE: {be_path} | FE: {fe_path}\n\n**Stack**: {lang} / {fw} / {db} / {orm}\n**PM**: {pm}\n**Run**: BE: {cmd_be} | FE: {cmd_fe}\n**Libraries**: {libs}\n\nAccept?","options":[{"id":"accept","label":"Accept - Proceed with this setup"},{"id":"modify","label":"Modify - Switch to manual selection"}]}]}

• accept -> Save all, proceed to 0-2 | modify -> 0-1.8b

0-1.8b. Manual Setup - Project Structure

{"title":"Project Structure","questions":[{"id":"project_structure","prompt":"What project structure do you want?","options":[{"id":"monorepo","label":"Monorepo - BE+FE in same repo (apps/{serviceName}-api|web)"},{"id":"monolith","label":"Monolith - Single service (backend/, frontend/)"},{"id":"custom","label":"Custom - I will specify paths"}]}]}

• monorepo -> apps/{serviceName}-api, apps/{serviceName}-web
• monolith -> backend/, frontend/
• custom -> manual path input

0-1.8c. Manual Setup - Language & Framework

{"title":"Language Selection","questions":[{"id":"language","prompt":"What programming language for backend?","options":[{"id":"python","label":"Python"},{"id":"typescript","label":"TypeScript (Node.js)"},{"id":"go","label":"Go"},{"id":"java","label":"Java"},{"id":"other","label":"Other (specify)"}]}]}

Then ask framework based on language:

Language	Framework Options
Python	FastAPI, Django, Flask
TypeScript	NestJS, Express, Fastify
Go	Gin, Echo, Fiber
Java	Spring Boot, Quarkus

0-1.8d. Manual Setup - Database & ORM

{"title":"Database Selection","questions":[{"id":"database","prompt":"What database will you use?","options":[{"id":"postgresql","label":"PostgreSQL"},{"id":"mysql","label":"MySQL / MariaDB"},{"id":"mongodb","label":"MongoDB"},{"id":"sqlite","label":"SQLite"},{"id":"none","label":"No database"},{"id":"other","label":"Other (specify)"}]}]}

ORM by language:

Language	ORM Options
Python	SQLAlchemy, Tortoise ORM, Raw SQL
TypeScript	Prisma, TypeORM, Drizzle, Raw SQL
Go	GORM, sqlx, Raw SQL
Java	JPA/Hibernate, MyBatis, Raw SQL

0-1.8e. Manual Setup - Package Manager

Python:

{"title":"Python Package Manager","questions":[{"id":"python_pkg_manager","prompt":"Which package manager?","options":[{"id":"uv","label":"uv - Fast, modern"},{"id":"pip","label":"pip + venv - Standard Python"},{"id":"poetry","label":"Poetry - Dep management + packaging"}]}]}

Node.js:

{"title":"Node Package Manager","questions":[{"id":"node_pkg_manager","prompt":"Which package manager?","options":[{"id":"npm","label":"npm - Default Node.js"},{"id":"yarn","label":"Yarn - Fast, reliable"},{"id":"pnpm","label":"pnpm - Efficient disk space"}]}]}

-> After all manual selections, proceed to 0-1.9

0-1.9. Library Review

Analyze spec.md requirements, recommend libraries per feature:

{"title":"Library Review: {feature_name}","questions":[{"id":"lib_{library_name}","prompt":"{library_name} ({version}) - {purpose}","options":[{"id":"use","label":"Use as recommended"},{"id":"skip","label":"Don't use"},{"id":"alt","label":"Use alternative (specify)"},{"id":"recommend","label":"Get LLM recommendation"}]}]}

For existing projects: show installed with markers, highlight missing. Record decisions in Dependencies section.

0-2. Infer serviceName

Extract from path: docs/alert/spec.md -> serviceName = "alert" Output: docs/alert/arch-be.md or docs/alert/arch-fe.md

Phase 1: Initial Input Collection

Confirm requirements doc path (e.g., @docs/alert/spec.md). If already provided -> Phase 1.5.

Phase 1.5: Parse Requirement Summary

Read spec.md and extract Requirement Summary grid:

1. Find ## 0. Requirement Summary section
2. Parse table: Req ID (FR-001, FR-002...), Category, Requirement, Priority
3. Store for Code Mapping: each Req ID -> Code Mapping Spec Ref (1:N)

If not found: warn "spec.md missing Requirement Summary grid", suggest /reinforce.

Phase 2: Parallel Initial Design

2-A: Domain Architect (via Task)

Task(
  subagent_type: "domain-architect",
  description: "Project context-based design",
  prompt: """
    Requirements: {requirements MD content}
    Feature description: {feature description}
    Project structure: {project exploration results}
    Based on the above, provide a design proposal.
  """
)

Expected: design proposal (implementation approach, file structure, dependencies), constraints, tradeoffs.

2-B: Best Practice Advisor (via Task)

Task(
  subagent_type: "best-practice-advisor",
  description: "Best practice-based design",
  prompt: """
    Feature description: {feature description only, no project context}
    Provide an ideal design proposal for implementing this feature.
  """
)

Expected: ideal design (based on best practices), patterns/anti-patterns, considerations.

Phase 3: Debate (Automated, 2 rounds)

Forward the following relay prompt to both agents:

From now on, I will relay another LLM's response.
That LLM was also instructed on the same feature definition.

Other agent's design:
{other_agent_response}

Review from your perspective and provide:
1. Points you agree with
2. Concerns
3. Alternative suggestions (if any)
4. Can we reach consensus: [Agree / Need further discussion]

Round 1: Domain Architect design -> Best Practice Advisor reviews. Round 2: Best Practice review -> Domain Architect revises/rebuts.

Phase 4: User Checkpoint

After 2 rounds, must provide interim report:

Item	Domain Architect	Best Practice Advisor
Core argument	(1-2 sentence)	(1-2 sentence)
Agreed points	(list)	(list)
Disputed points	(list)	(list)

Options (AskQuestion, dynamically generated based on debate results):

Field	Description
title	Reflect debate topic (e.g., "Alert Scheduling Design Debate")
prompt	Include report table summary above
options	4 basic options below, adjust based on situation

• domain: Adopt Domain Architect - Prioritize project reality
• best: Adopt Best Practice - Prioritize ideal design
• continue: 2 more rounds of debate
• custom: User combines on disputed points

Option adjustment: If both agree -> exclude continue. If one clearly superior -> mark as default.

NOTE: Run in Normal mode, not Plan mode. Can both intervene mid-process with AskQuestion + write final file.

Phase 5: Follow-up

Options 1,2,4 -> Phase 6. Option 3 -> Rounds 3-4, auto-terminate after 4. If no consensus: adopt Domain Architect + note Best Practice recommendations.

Phase 5.5: Quality Gate Validation

Validate completeness based on arch_type before writing.

Backend (BE):

Item	Validation Criteria	If Incomplete
DB Schema	All tables have full column definitions	Enter question loop
Code Mapping	All features mapped to file/class/method	Enter question loop
API Spec	All endpoints have Request/Response defined	Enter question loop
Error Policy	Main error scenarios and responses defined	Enter question loop

Frontend (FE):

Item	Validation Criteria	If Incomplete
Component Structure	Page and reusable component hierarchy	Enter question loop
Code Mapping	All features mapped to file/component/hook	Enter question loop
State Management	Global/Server/Local state defined	Enter question loop
Route Definition	Routes and auth guards defined	Enter question loop
API Integration	Backend endpoints connected to hooks	Enter question loop

Question Loop

{"title":"Design Information Supplement Required","questions":[{"id":"missing_info","prompt":"Missing:\n- {list}\n\nHow to proceed?","options":[{"id":"provide","label":"Provide - I will provide the information"},{"id":"infer","label":"Allow inference - AI can estimate"},{"id":"skip","label":"Skip - I will supplement later"}]}]}

• provide -> user input | infer -> AI estimates, note in Assumptions | skip -> mark "[TBD]"

Fail Gate (return to debate if empty)

Backend:

Condition	Minimum Requirement
If API exists	Request/Response spec + at least 2 error responses
If Auth exists	Role/Permission table with at least 1 entry
If Data (DB) exists	Core entity with at least 5 fields defined

Frontend:

Condition	Minimum Requirement
If Pages exist	At least 1 page component with route defined
If Auth exists	Auth guard and protected route defined
If API calls exist	At least 1 hook/service with endpoint connection
If State exists	At least 1 global state store defined

WARNING: If any fails -> re-enter Phase 5.5. When all pass -> Phase 6.

Phase 6: Write Design Document

Write implementation-ready design document based on debate results.

Section	Content	Source
0. Summary	Goal/Non-goals/Metrics	Requirements
1. Scope	In/Out	Debate
1.5. Tech Stack	Language, framework, DB, etc.	Exploration/user
2. Architecture Impact	Components, Data	Domain Architect
3. Code Mapping	Feature -> file/module	Domain Architect + exploration
4. Implementation Plan	Order + reference files	Agreed priority
5. Sequence Diagram	Main flow	Main agent
6. API Specification	Endpoints, Req/Res	Main agent
7. Infra/Ops	Env vars, deployment	Domain Architect
8. Risks & Tradeoffs	Debate conclusion	Both agents
9. Checklist	Error/Auth/Data	Error, auth, data integrity

Tech Stack (Section 1.5)

Item	Examples
Project Structure	Monorepo / Monolith / Custom
BE Path	apps/auth-api / backend / ./
FE Path	apps/auth-web / frontend / ./
Run Command (BE)	uv run uvicorn main:app / npm run dev
Run Command (FE)	npm run dev / yarn dev
Language	Python 3.11 / TypeScript 5.x / Go 1.21 / Java 17
Framework	FastAPI / NestJS / Spring Boot / Gin
DB	PostgreSQL 15 / MySQL 8 / MongoDB 6
ORM	SQLAlchemy / Prisma / TypeORM / GORM / None (Raw SQL)
Package Manager	uv / pip / poetry / npm / yarn / pnpm
3rd-party	Redis, Kafka, S3, Elasticsearch, etc.
Infra	K8s / Docker / AWS / GCP / On-premise

NOTE: BE/FE Paths and Run Commands used by build, test, debug skills.

DB Changes (Section 2)

All tables must have full column definitions.

Anti-patterns (Prohibited):

• "Delete X field from existing table" (change without full schema)
• "Modify existing table" (unclear what columns exist)

Example:

Column	Type	Change	Description
id	INTEGER	Maintain	PK
event_type	VARCHAR(100)	Maintain	Event type
reference_id	VARCHAR(100)	New	Related resource ID
is_read	BOOLEAN	Delete	Moved to separate table

When manual SQL migration: record change types (CREATE TABLE / ALTER TABLE / CREATE INDEX) for build phase.

Code Mapping (Section 3)

Require: full file path, class, method, call location, code to add.

Anti-patterns (Prohibited):

• "Call AlertPublisher from Issue service" (which method?)
• "src/services/*.py (modify)" (no specific file and method)

Example:

#	Feature	File	Class	Method	Action	Impl
1	Quality Create	.../quality_issue/service.py	QualityIssueService	create_issue()	After DB save call _publish_create_alert()	[ ]
2	Quality Resolve	.../quality_issue/service.py	QualityIssueService	resolve_issue()	After status change call _publish_resolve_alert()	[ ]

[ ] = not implemented, [x] = done by build. # = row number, never reused.

Debate Conclusion (Section 8)

## 8. Risks & Tradeoffs (Debate Conclusion)
### Chosen Option
- (adopted design approach)
### Rejected Alternatives
- (unadopted items and reasons)
### Reasoning
- Project constraints: ...
- Best practice adoption: ...
- Future improvement points: ...
### Assumptions
- **Confirmed**: (clearly decided in debate)
- **Estimated**: (assumed due to lack of confirmation)

Save Path

docs/{serviceName}/arch-be.md or docs/{serviceName}/arch-fe.md

Output Template

• Backend: templates/be.md | Frontend: templates/fe.md
• Located in skills/arch/templates/. Read selected template before writing.

Template Reference (Backend - inline)

# Backend Design Doc: {Feature Name}
> Created: {date} | Service: {serviceName} | Spec: docs/{serviceName}/spec.md
## 0. Summary
**Goal**: (1-2 sentences)
**Non-goals**: (excluded)
**Success metrics**: (criteria)
## 1. Scope
**In scope**: (items)
**Out of scope**: (future improvements)

## 1.5. Tech Stack
\`\`\`yaml
tech_stack:
  language: "{lang}"
  framework: "{fw}"
  database: "{db}"
  orm: "{orm}"
  third_party: ["{svc1}", "{svc2}"]
  infra: "{infra}"
\`\`\`
## 2. Architecture Impact
| Service/Module | Responsibility | Change type |
|----------------|----------------|-------------|
| {service} | {role} | new/modify |
### Database Schema
\`\`\`yaml
database_schema:
  new_tables:
    - name: "{table}"
      columns:
        - {name: "id", type: "INTEGER", constraints: ["PK","AUTO_INCREMENT"]}
        - {name: "{col}", type: "{type}", constraints: ["{c1}"], desc: "{desc}"}
        - {name: "created_at", type: "TIMESTAMP", constraints: ["NOT NULL","DEFAULT NOW()"]}
  modified_tables:
    - name: "{table}"
      changes:
        - {column: "{col}", type: "{type}", action: "ADD|DROP|MODIFY", desc: "{desc}"}
  indexes:
    - {table: "{tbl}", columns: ["{c1}","{c2}"], type: "BTREE", unique: false}
migrations:
  - {type: "CREATE_TABLE", table: "{tbl}", desc: "{desc}"}
  - {type: "ALTER_TABLE", table: "{tbl}", desc: "{change}"}
  - {type: "CREATE_INDEX", table: "{tbl}", desc: "{index desc}"}
\`\`\`
## 3. Code Mapping
| # | Feature | File | Class | Method | Action | Impl |
|---|---------|------|-------|--------|--------|------|
| 1 | {feature} | {full path} | {class} | {method} | {action} | [ ] |
## 4. Implementation Plan
**Reference Files**: {path1} ({purpose}), {path2} ({purpose})
1. **{step}** - {tasks}
2. **{step}** - {tasks}
## 5. Sequence Diagram
\`\`\`mermaid
sequenceDiagram
    participant Client
    participant API
    participant Service
    participant DB
    Client->>API: {request}
    API->>Service: {call}
    Service->>DB: {query}
    DB-->>Service: {result}
    Service-->>API: {return}
    API-->>Client: {response}
\`\`\`
## 6. API Specification
\`\`\`yaml
api_endpoints:
  - name: "{api}"
    method: "{GET|POST|PUT|DELETE}"
    path: "/api/v1/{path}"
    auth_required: true
    request:
      headers: [{name: "Authorization", type: "string", required: true}]
      params: [{name: "{param}", type: "{type}", required: true}]
      body: {content_type: "application/json", schema: {field1: "{type}", field2: "{type}"}}
    response:
      success: {status: 200, schema: {success: "boolean", data: "{type}"}}
      errors:
        - {status: 400, code: "INVALID_INPUT", message: "{msg}"}
        - {status: 404, code: "NOT_FOUND", message: "{msg}"}
\`\`\`
## 7. Infra/Ops
| Variable | Description | Default |
|----------|-------------|---------|
| {VAR} | {desc} | {val} |
## 8. Risks & Tradeoffs
**Chosen**: {approach} | **Rejected**: {alternatives and why}
**Reasoning**: constraints: {reason}, best practice: {applied}, future: {later}
**Assumptions**: Confirmed: {decided} | Estimated: {assumed, needs verification}
## 9. Error/Auth/Data Checklist
**Error Handling**

| Situation | Location | Handling | Response |
|---|---|---|---|
| {scenario} | {file/method} | {try-catch/propagation/logging} | {HTTP code, msg} |

**Auth Rules**

| Action | Permission | Location | On Failure |
|---|---|---|---|
| {api} | {role} | {middleware} | {403/401} |

**Data Validation**

| Validation | Timing | On Failure |
|---|---|---|
| {FK existence} | {before save} | {400 + msg} |
| {Duplicate} | {on create} | {409 Conflict} |
| {Range} | {on input} | {422 + details} |
WARNING: If empty, 80% of implementation will be unstable - Must complete

Phase 6.5: Update spec.md Status

After saving: read spec.md, find ## 0. Requirement Summary table. Update Status Draft -> Designed for each Req ID used in Code Mapping.

Example:

| Req ID | Category | Requirement | Priority | Status |
|--------|----------|-------------|----------|--------|
| FR-001 | Auth | Email/password login | High | Designed |  <- Updated
| FR-002 | Auth | Social login | Medium | Designed |  <- Updated

Completion

Design Document Complete Saved to: docs/{serviceName}/arch-be.md or arch-fe.md Updated: spec.md (Status -> Designed) Next: Run /build (recommend Sonnet in new session)

Integration Flow

[spec] -> spec.md -> [arch] -> arch-be/fe.md -> [build] -> Implementation -> (Bug) -> [debug] -> trace.md