← スキル一覧に戻る

auto-dev
by ksterx
⭐ 0🍴 0📅 2026年1月22日
SKILL.md
name: auto-dev description: | Autonomous development workflow for code implementation, bug fixes, feature development, and refactoring. Automatically determines task complexity and adjusts workflow accordingly. Use when: implementing features, fixing bugs, refactoring code, or any development task.
Autonomous Development Workflow
You are an autonomous development orchestrator. Analyze tasks, determine complexity, and execute the appropriate workflow with minimal human intervention.
References (read as needed)
| Reference | When to Read |
|---|---|
references/templates/spec.md | When generating spec.md |
references/templates/plan.md | When generating plan.md |
references/templates/tasks.md | When generating tasks.md |
references/agent-prompts.md | When invoking agents |
references/handoff-schema.md | When passing context between agents |
references/gemini-review.md | When performing Gemini review |
references/stacks/typescript.md | TypeScript/React projects (auto-detect) |
Preflight: Stack Detection
Before complexity assessment, detect project stack and read relevant stack reference:
| Detection Criteria | Stack File |
|---|---|
package.json has next in dependencies + tsconfig.json | references/stacks/typescript.md (Next.js mode) |
package.json has react (no next) + tsconfig.json or .tsx files | references/stacks/typescript.md (React-only mode) |
Detection notes:
- Check
dependenciesordevDependenciesentries (not substring match) - Also accept
tsconfig.*.jsonvariants - Skip if project already has established different choices (e.g., Redux, MUI)
Apply stack conventions throughout implementation, respecting existing project choices.
Phase 0: Complexity Assessment
| Level | Criteria | SDD | Gemini |
|---|---|---|---|
| Trivial | 1-2 lines, single file, obvious | No | No |
| Simple | Clear, 1-3 files, existing patterns | No | No |
| Standard | New functionality, multiple files, design needed | Yes | Yes |
| Complex | Large scope, architectural impact, ambiguous | Yes | Yes |
Output:
Complexity: [LEVEL]
Reasoning: [Brief justification]
Workflow by Complexity
Trivial
Implement directly → Verify → Complete
Simple
explorer(summary) → implementer → tests → reviewer → auto-fix → Complete
Standard (with SDD)
EXPLORATION
explorer(detailed) → [Gemini] → validate
SPECIFICATION
architect(spec) → [Gemini] → 👤 APPROVE
architect(plan) → [Gemini] → 👤 APPROVE
architect(tasks) → [Gemini] → 👤 APPROVE
IMPLEMENTATION
implementer → tests → [Gemini] → validate
REVIEW
reviewer(spec-compliance) → [Gemini]
reviewer(standard) → [Gemini] → auto-fix → Complete
Complex (with SDD)
EXPLORATION
explorer(scoped) x2-3 parallel → [Gemini each]
👤 REVIEW: clarify requirements
SPECIFICATION
architect(spec) → [Gemini] → 👤 APPROVE
architect(plan + approach) x2-3 parallel → [Gemini each] → 👤 CHOOSE & APPROVE
architect(tasks) → [Gemini] → 👤 APPROVE
IMPLEMENTATION
implementer(phased) → tests each phase → [Gemini]
REVIEW
reviewer(spec-compliance) → [Gemini]
reviewer(scoped) x3 parallel → [Gemini each] → auto-fix → Complete
SDD Documents
specs/{feature-slug}/
├── spec.md # What (requirements, acceptance criteria)
├── plan.md # How (architecture, components)
└── tasks.md # Steps (phased checklist)
Naming: kebab-case, English. Optional: YYYYMMDD-feature-slug
Agent Modes Quick Reference
| Agent | Modes |
|---|---|
| code-explorer | detailed, summary, scoped:[entry/data/cross-cutting] |
| code-architect | spec, plan, tasks, full, parallel:[minimal/clean/pragmatic] |
| code-implementer | (default) |
| code-reviewer | standard, spec-compliance, bugs, quality, conventions |
For full prompts, read references/agent-prompts.md
Gemini Review (Standard+ only)
After each agent output:
- Call
mcp__gemini__gemini-brainstormwith summary - Evaluate response neutrally and fairly
- Apply valid fixes, ignore subjective/style-only feedback
- Present to human with applied/ignored notes
For full protocol, read references/gemini-review.md
Human Review Points (Standard+)
| Document | Approval Criteria |
|---|---|
| spec.md | Requirements complete, AC clear |
| plan.md | Architecture sound, approach approved |
| tasks.md | Tasks atomic, dependencies correct |
Iteration Budget
| Complexity | Budget |
|---|---|
| Trivial | 0 |
| Simple | 2 |
| Standard | 6 |
| Complex | 10 |
Definition of Done
- All requirements from spec.md implemented
- All acceptance criteria verified
- All tasks in tasks.md checked
- Lint/type-check pass
- Tests pass
- Spec compliance verified
Output Format
## Summary
- **Task**: [Request]
- **Complexity**: [Level]
- **Feature**: [slug]
- **Documents**: specs/{slug}/ ✅
- **Files modified**: [List]
- **Tests**: [Pass/Fail]
スコア
総合スコア
40/100
リポジトリの品質指標に基づく評価
✓SKILL.md
SKILL.mdファイルが含まれている
+20
○LICENSE
ライセンスが設定されている
0/10
○説明文
100文字以上の説明がある
0/10
○人気
GitHub Stars 100以上
0/15
○最近の活動
3ヶ月以内に更新がある
0/10
○フォーク
10回以上フォークされている
0/5
✓Issue管理
オープンIssueが50未満
+5
✓言語
プログラミング言語が設定されている
+5
○タグ
1つ以上のタグが設定されている
0/5
レビュー
💬
レビュー機能は近日公開予定です