ADR Write Command

Write ADR documents based on discovered candidates from adr-candidates.md.

User Input

text
1$ARGUMENTS

You MUST consider the user input before proceeding.

Overview

Generate complete ADR documents by researching GitHub commits/PRs and applying the ADR template.

Multi-Agent Collaboration Pipeline:

┌─────────────────────────────────────────────────────────────────┐
│                    ADR Writing Pipeline                          │
├─────────────────────────────────────────────────────────────────┤
│  Phase 1: specvital-specialist                                   │
│    → Search related commits/PRs via GitHub API                   │
│    → Extract decision context and rationale                      │
│    → Identify options that were considered                       │
├─────────────────────────────────────────────────────────────────┤
│  Phase 2: product-strategist                                     │
│    → Analyze decision trade-offs                                 │
│    → Evaluate consequences (positive/negative)                   │
│    → Structure options comparison                                │
├─────────────────────────────────────────────────────────────────┤
│  Phase 3: Write ADR Documents                                    │
│    → Generate English version (docs/en/adr/)                     │
│    → Generate Korean version (docs/ko/adr/)                      │
│    → Update all required index files                             │
│    → Update sidebar configuration                                │
└─────────────────────────────────────────────────────────────────┘

Input Interpretation

Numeric Reference

Reference candidates by number from adr-candidates.md:

bash
1/adr-write 1          # Platform Level candidate #1
2/adr-write core/1     # Core repository candidate #1
3/adr-write worker/1   # Worker repository candidate #1

Natural Language

Describe what ADR to write:

bash
1/adr-write billing and quota system
2/adr-write write the AI spec document generation ADR
3/adr-write domain hints extraction for core repository

Parsing Logic

1. Numeric only (e.g., 1, 2): Match Platform Level candidate by number
2. Prefixed numeric (e.g., core/1, web/2): Match repository-level candidate
3. Natural language: Search candidate titles and descriptions for best match

Execution Flow

Phase 0: Load Candidate Information

Read adr-candidates.md and identify the target ADR:

1. Parse user input to identify candidate
2. Extract from adr-candidates.md:
   - Title
   - Target location (e.g., /en/adr/13-xxx.md)
   - Related repositories
   - Complexity and priority
3. Determine ADR number and file paths

File Path Determination:

Phase 1: Context Research (specvital-specialist)

Task tool invocation:

Task(subagent_type="specvital-specialist", prompt="""
Research GitHub context for ADR: {candidate_title}

**Target Repositories**: {related_repos}

**Research Tasks**:

1. **Search Related Commits**
   - Use gh CLI or GitHub MCP to search commits
   - Keywords: {extracted_keywords}
   - Timeframe: Last 90 days
   - Command: `gh search commits "{keyword}" --repo specvital/{repo} --limit 20`

2. **Search Related PRs**
   - Find PRs that introduced or modified this feature
   - Command: `gh pr list --repo specvital/{repo} --search "{keyword}" --state all --limit 10`

3. **Extract Decision Context**
   For each relevant commit/PR:
   - What problem was being solved?
   - What approach was taken?
   - Any alternatives mentioned in discussions?

4. **Output Format**

Relevant Commits

• {commit_sha}: {message} - {date}

Relevant PRs

• #{pr_number}: {title} - {state}

Context Summary

• Problem: ...
• Decision: ...
• Alternatives mentioned: ...

""")

Phase 2: Decision Analysis (product-strategist)

Task tool invocation:

Task(subagent_type="product-strategist", prompt="""
Analyze the architectural decision for ADR: {candidate_title}

**Input**: Context from Phase 1

**Analysis Tasks**:

1. **Structure the Context**
   - What was the problem/need?
   - Why was a decision required?
   - What constraints existed?

2. **Identify Options**
   Based on commits/PRs and common patterns:
   - Option A (selected): What was chosen and why
   - Option B: What alternatives existed
   - Option C: Any other approaches considered

3. **Evaluate Consequences**
   For the selected option:
   - Positive outcomes (benefits, improvements)
   - Negative outcomes (trade-offs, limitations)
   - Technical implications

4. **Output Format**
   Structured content ready for ADR template sections:
   - Context section content
   - Decision section content
   - Options Considered section content
   - Consequences section content
""")

Phase 3: Write ADR Documents

Generate both English and Korean versions following the template.

ADR Template Structure:

markdown
1---
2title: { Title }
3description: ADR for {brief description}
4---
5
6# ADR-{num}: {Title}
7
8> [Korean Version](/ko/adr/{path}) OR 🇰🇷 [한국어 버전](/ko/adr/{path})
9
10| Date       | Author    | Repos            |
11| ---------- | --------- | ---------------- |
12| YYYY-MM-DD | @{author} | {repo1}, {repo2} |
13
14## Context
15
16{Problem situation and why a decision was needed}
17
18## Decision
19
20{The decision made and its rationale}
21
22## Options Considered
23
24### Option A: {Name} (Selected)
25
26{Description, pros, cons}
27
28### Option B: {Name}
29
30{Description, pros, cons}
31
32## Consequences
33
34**Positive:**
35
36- {Benefit 1}
37- {Benefit 2}
38
39**Negative:**
40
41- {Trade-off 1}
42- {Trade-off 2}
43
44## References
45
46- [Related ADR](/en/adr/XX-related.md)
47- [Related PR](https://github.com/specvital/{repo}/pull/{num})

Post-Write Updates (CRITICAL)

After writing ADR documents, you MUST update all required files per CLAUDE.md rules.

For Platform Level ADR (e.g., adr/13-xxx.md)

Update 6 files:

☐ docs/en/index.md           ← Homepage ADR links section
☐ docs/ko/index.md           ← Homepage ADR links section
☐ docs/en/adr/index.md       ← ADR overview
☐ docs/ko/adr/index.md       ← ADR overview
☐ docs/.vitepress/config.mts ← Sidebar navigation

For Repository Level ADR (e.g., adr/core/16-xxx.md)

Update 6 files:

☐ docs/en/index.md                 ← Homepage ADR links section
☐ docs/ko/index.md                 ← Homepage ADR links section
☐ docs/en/adr/index.md             ← ADR overview (repository section)
☐ docs/ko/adr/index.md             ← ADR overview (repository section)
☐ docs/en/adr/{category}/index.md  ← Category index
☐ docs/ko/adr/{category}/index.md  ← Category index
☐ docs/.vitepress/config.mts       ← Sidebar navigation

Language Rules

English Version (/en/)

• Write entirely in English
• Use technical terminology appropriately
• Clear, concise prose

Korean Version (/ko/)

• Write entirely in Korean
• MUST use noun-ending style (명사형 종결)
• NEVER use verb-ending style (동사 종결어미 금지)

Korean Style Examples:

Key Rules

Must Do

• Load candidate info from adr-candidates.md
• Research actual GitHub commits/PRs for context
• Follow ADR template structure exactly
• Generate both English and Korean versions
• Update ALL required index files (check CLAUDE.md for complete list)
• Update sidebar in config.mts
• Use noun-ending style for Korean

Must Not Do

• Skip GitHub research (must base on actual commits)
• Write only one language version
• Forget to update index files
• Use verb endings in Korean documents
• Guess content without evidence from commits/PRs

Output Summary

After completion, report:

## ADR Written

| Item | Value |
|------|-------|
| Title | {ADR Title} |
| Number | ADR-{num} |
| English | docs/en/adr/{path} |
| Korean | docs/ko/adr/{path} |

## Files Updated

- [x] docs/en/adr/{path} (created)
- [x] docs/ko/adr/{path} (created)
- [x] docs/en/index.md
- [x] docs/ko/index.md
- [x] docs/en/adr/index.md
- [x] docs/ko/adr/index.md
- [x] docs/.vitepress/config.mts

## Source References

- {commit_sha}: {message}
- PR #{num}: {title}

Usage Examples

bash
1# Write Platform Level candidate #1 (Billing and Quota)
2/adr-write 1
3
4# Write Core repository candidate #1
5/adr-write core/1
6
7# Natural language - matches "AI-Based Spec Document Generation Pipeline"
8/adr-write AI spec generation
9
10# Natural language with context
11/adr-write write the billing ADR, focus on quota tracking aspects

Related Commands

• /adr-discover: Discover ADR candidates
• /document-business-rule: Business rule documentation
• /commit: Commit message generation

Related Agents

• specvital-specialist: GitHub context research
• product-strategist: Decision analysis and trade-offs
• prompt-engineer: Document quality optimization
• claude-code-specialist: Command and workflow optimization

Execution

Now execute the workflow:

1. Parse $ARGUMENTS: Identify target candidate from adr-candidates.md
2. Phase 1: Invoke Task(specvital-specialist) for GitHub research
3. Phase 2: Invoke Task(product-strategist) for decision analysis
4. Phase 3: Write ADR documents (English + Korean)
5. Update Index Files: Per CLAUDE.md requirements
6. Report: Summary of created/updated files