jaykim88/future-architect

Future Architect Skill

Trigger patterns:

• Korean: "생각 정리", "계획 수립", "아이디어 정리", "목표 정리"
• English: "organize my thoughts", "plan clarification", "organize ideas", "structure my plan"

This skill detects user intent and launches the future-architect agent to conduct multi-round thought organization interviews.

---

Execution Logic

Step 1: Detect Intent

Check if the user wants to:

• Create a new plan (free-form input with topics)
• Update an existing plan (mentions "업데이트", "update", or references an existing document)

Step 2: Route to Agent

Case A: New Plan

If the user provides free-form input with topics (bullet points or narrative):

Use Task tool:
  subagent_type: "future-architect"
  description: "Organize thoughts into actionable plan"
  prompt: |
    The user wants to organize their thoughts. Here is their input:

    {user_input}

    Please conduct the 10-step thought organization process:
    - Parse topics from input
    - Conduct topic-by-topic interviews (2-6 rounds per topic)
    - Generate cross-topic relationship analysis
    - Create Mermaid diagram
    - Generate capacity-aware TODO list
    - Save markdown document with agent_id in frontmatter

IMPORTANT: The agent will handle all interview rounds, question generation, and document creation. Do NOT conduct the interview yourself — just launch the agent.

Case B: Update Existing Plan

If the user mentions updating an existing plan:

1. Ask for the document path:

   AskUserQuestion:
     question: "어떤 계획을 업데이트하시겠어요?"
     header: "문서 선택"
     multiSelect: false
     options:
       - "직접 경로 입력" (파일 경로를 입력하겠습니다)
   

2. Read the document using the Read tool

3. Extract agent_id from frontmatter:

   ---
   agent_id: abc123
   created_at: 2026-02-11
   status: active
   ---
   

4. Resume the agent:

   Use Task tool:
     subagent_type: "future-architect"
     description: "Update existing thought plan"
     resume: "{extracted_agent_id}"
     prompt: |
       The user wants to update their existing plan.
   
       Original document path: {document_path}
   
       Ask the user what they want to update:
       - Add new topics
       - Deepen existing topics
       - Update TODOs
       - Regenerate diagram with new relationships
   
       Then proceed with the relevant steps from the 10-step algorithm.
   

IMPORTANT: You must extract and provide the agent_id from the document frontmatter. If the frontmatter doesn't have agent_id, it was created before the agent conversion — treat it as a new plan instead.

---

Error Handling

| Scenario | Action | |----------|--------| | No topics detected in input | Ask user to provide topics in bullet format or narrative | | User input is too vague | Ask clarifying question: "무엇에 대한 생각을 정리하고 싶으신가요?" | | Existing document has no agent_id | Inform user: "이 문서는 이전 버전으로 생성되어 agent_id가 없습니다. 새 계획으로 시작하시겠어요?" | | Document path doesn't exist | Ask user to verify path or create new plan |

---

Examples

Example 1: New Plan (Multi-Topic)

User:

생각 정리해줘

- 커리어 전환
  급하게 취업을 시도하기 보다, 현재까지 vntg에서 했던 것들을 정리...
- 독일행
  독일에서 취업할 수 있는 개발자 포지션은...
- 창업
  서비스를 많이 만들어보고 싶다...

Skill action:

→ Detect: New plan with 3 topics
→ Launch agent with Task tool
→ Agent conducts full 10-step process
→ Document saved with agent_id

Example 2: New Plan (Single Topic)

User: "커리어 전환에 대해 생각 정리하고 싶어"

Skill action:

→ Detect: New plan with 1 topic
→ Launch agent
→ Agent parses single topic, conducts interview
→ Document saved

Example 3: Update Existing Plan

User: "이전 계획을 업데이트하고 싶어"

Skill action:

→ Detect: Update intent
→ Ask for document path
→ User provides: ~/Documents/plans/2026-02-11-커리어-독일-창업-투자.md
→ Read document
→ Extract agent_id from frontmatter
→ Resume agent with agent_id
→ Agent asks what to update, proceeds accordingly

Example 4: Ambiguous Input

User: "생각 정리"

Skill action:

→ Detect: Trigger phrase but no content
→ Ask: "무엇에 대한 생각을 정리하고 싶으신가요? 자유롭게 말씀해주세요."
→ Wait for user response
→ Launch agent with user's clarified input

---

Configuration

The agent reads configuration from: ~/.claude/skills/future-architect/config.yaml

Default config:

output_dir: "~/Documents/plans"
auto_open: true
diagram_direction: "TD"
priority_system: "3-tier"
filename_format: "{date}-{topic1}-{topic2}-....md"  # Must run `date '+%Y-%m-%d'` to get exact date. Never estimate.

If config doesn't exist, the agent uses these defaults.

---

Related Files

• Agent: plugins/future-architect/agents/future-architect.md (actual logic)
• Config: ~/.claude/skills/future-architect/config.yaml (settings)
• Output: ~/Documents/plans/ (default save location)

---

Important Notes

• This Skill is just a trigger and router — all logic is in the agent
• The agent maintains state across multiple turns (multi-round interviews)
• The agent saves agent_id in document frontmatter for resume functionality
• To update a plan, the user must provide the document path (Skill extracts agent_id)

---

Architecture

User input
    ↓
future-architect Skill (this file)
    ↓
Detect: New plan or Update?
    ↓
Launch future-architect Agent
    ↓
Agent executes 10-step algorithm
    ↓
Document saved with agent_id