Initialize Self-Learning Resource Project

Interactive interview-based project initialization for learning resources.

---

Phase 1: Project Interview

<communication_rules> CRITICAL: The interviewer agent communicates DIRECTLY with the user.

• Do NOT add any introduction before launching the interviewer
• Do NOT add transition messages like "Starting interview..." or "Let me start the interview"
• Do NOT summarize or rephrase the agent's output after it completes
• Let the conversation flow naturally as if the interviewer IS the main assistant
• The user should feel they are talking to ONE entity, not being handed off to another

When the interview completes, silently proceed to Phase 2 without announcing completion. </communication_rules>

Launch Project Interviewer

Task(
  subagent_type="dokhak:project-interviewer",
  model="opus",
  description="Conduct project initialization interview",
  prompt="""
    <role>
    You are a learning experience designer having a direct conversation with the learner.
    You ARE the main assistant - not a separate tool or delegated agent.
    </role>

    <task>
    Conduct an interview for project initialization following the project-interview skill.
    </task>

    <instructions>
    1. Start immediately with your multilingual greeting - no preamble
    2. Have a natural conversation to understand the learner
    3. Generate persona.md and interview-data.md
    4. Return results in project_metadata XML format
    </instructions>

    <output_format>
    Use persona-template.md from project-scaffolder skill when creating persona.md.
    </output_format>
  """
)

Interview Goals

Through natural conversation, understand:

What to Learn	How to Discover
Topic	Open question: "What would you like to learn?"
Domain	Infer from topic, confirm if ambiguous
Motivation	Ask why they want to learn this
Level	Ask about background and experience
Preferences	Discuss desired style and volume
Environment	For technology: ask about OS/tools

Store the returned project_metadata XML for subsequent phases.

---

Phase 2-6: Background Processing

<background_processing_rules> After the interview completes, execute Phases 2-6 silently in sequence.

• Do NOT announce each phase to the user (e.g., "Now collecting research...")
• Do NOT show progress updates between phases
• Process all phases quietly, then show only the final Completion Report
• If any phase fails, note it in the report rather than interrupting with error messages </background_processing_rules>

Phase 2: Research Collection

Use the research-collector subagent with domain-adaptive strategy. Research results are saved to .research/init/ directory.

Task(
  subagent_type="dokhak:research-collector",
  model="haiku",
  description="Research {topic} resources",
  prompt="""
    <research_request>
      <topic>{topic}</topic>
      <domain>{domain}</domain>
      <audience_level>{audience}</audience_level>
    </research_request>

    Conduct research appropriate for this domain using the domain-profiles skill.
    Save results to .research/init/ following research-storage skill conventions.
    Return confirmation message only (not the research content).
  """
)

Agent returns confirmation like:

research_saved:.research/init/
sources:12
concepts:8

IMPORTANT: Do NOT store the research content. The confirmation message is sufficient.

---

Phase 3: Create plan.md

Use the structure-designer subagent to create plan.md. Pass file PATHS only - structure-designer reads files directly.

Task(
  subagent_type="dokhak:structure-designer",
  model="opus",
  description="Create plan.md structure",
  prompt="""
    <structure_request>
      <topic>{topic}</topic>
      <domain>{domain}</domain>
      <volume>{pages} pages</volume>
      <language>{lang}</language>
      <audience>{audience}</audience>
      <environment>{environment}</environment>
    </structure_request>

    <research_files>
      <summary_path>.research/init/summary.md</summary_path>
      <sources_path>.research/init/sources.md</sources_path>
    </research_files>

    Read the research files above using Read tool.
    Create plan.md using the project-scaffolder skill as reference.
  """
)

IMPORTANT: Do NOT Read the research files in this command. The structure-designer reads them directly in its isolated context.

---

Phase 4: Create task.md

Use the structure-designer subagent to create task.md with session distribution.

Task(
  subagent_type="dokhak:structure-designer",
  model="opus",
  description="Create task.md with sessions",
  prompt="""
    Read plan.md and create task.md using the project-scaffolder skill as reference.
    Split into sessions and mark session boundaries with HTML comments.
  """
)

---

Phase 5: Create project-context.md

Create project-context.md with research results and domain-specific information.

Using project-context-template.md from project-scaffolder skill as reference,

Create project-context.md with:
- Research results from Phase 2 (authoritative sources)
- Domain-specific context from {domain_specific} XML
- Environment-specific notes (if technology)
- Version information (if applicable)
- Learning path recommendations
- Prerequisites summary

---

Phase 6: Create CLAUDE.md

Create CLAUDE.md using claude-md-template.md from project-scaffolder skill as reference:

• Project overview (topic, domain, audience)
• Key skills and agents references (/write, researcher, writer)
• Document structure guidelines
• Writing persona summary (from persona.md)
• Domain-specific guidelines (from domain-profiles)
• Quality standards

---

Completion Report

After all phases complete, provide a summary:

=== Project Initialization Complete ===

Topic: {topic}
Domain: {domain}
Scope: {pages} pages in {lang}
Audience: {audience}
Environment: {environment} (if applicable)

Files Created:
  ✓ persona.md - Writer/Reader personas with {domain} guidelines
  ✓ interview-data.md - Interview transcript and insights
  ✓ plan.md - X Parts, Y Chapters, Z Sections
  ✓ task.md - N sessions
  ✓ project-context.md - Research and context
  ✓ CLAUDE.md - Project guidelines

Next Steps:
  1. Review plan.md structure
  2. Review persona.md for accuracy
  3. Run /write to start writing first section
  4. Use /status to track progress

---

Variable Mapping

Interview collects and maps the following variables:

Variable	Source	Notes
topic	Conversation	What the learner wants to learn
domain	Inferred + confirmed	technology, history, science, arts, general
pages	Conversation	50 (small), 100 (medium), 200 (large)
lang	Detected from conversation	ko, en, etc.
audience	Inferred from background	beginner, intermediate, advanced
environment	Conversation (tech only)	linux, macos, windows, cross-platform
motivation	Conversation	Why they want to learn
preferences	Conversation	Learning style preferences

---

Error Recovery

• If research yields limited results: Proceed with available information, note gaps
• If page allocation doesn't sum correctly: Adjust proportionally
• If any file creation fails: Report specific error, continue with others
• If persona.md already exists: Ask user whether to overwrite or merge