Skip to content

CUSTOMIZATION.md

Latest commit

 

History

History
353 lines (259 loc) · 11.4 KB

CUSTOMIZATION.md

File metadata and controls

353 lines (259 loc) · 11.4 KB
Raw

Customization Guide

PM Coach is designed to be extended and personalized. This guide covers the most common customizations.

Table of Contents

Add a New Learning Track

Each track lives in learning/tracks/<track-id>/ and needs three files.

1. Create the Directory

mkdir -p learning/tracks/my-new-track

2. Create track.json

{
  "track_id": "my-new-track",
  "display_name": "My New Track",
  "pillar": "pm-craft",
  "pillar_emoji": "🛠️",
  "pillar_tier": "always-on",
  "created": "2025-01-15",
  "last_updated": "2025-01-15",
  "topic_type": "framework-application",
  "session_count": 0,
  "mastery": {
    "overall": 30,
    "dimensions": {
      "core_concepts": { "score": 30, "trend": "flat", "notes": "" },
      "practical_application": { "score": 30, "trend": "flat", "notes": "" },
      "current_research": { "score": 30, "trend": "flat", "notes": "" }
    }
  },
  "conversational_readiness": {
    "overall": 20,
    "dimensions": {
      "vocabulary_fluency": { "score": 20, "trend": "flat", "last_tested": null },
      "conceptual_depth": { "score": 20, "trend": "flat", "last_tested": null },
      "current_awareness": { "score": 20, "trend": "flat", "last_tested": null, "staleness_days": 0 },
      "opinion_formation": { "score": 20, "trend": "flat", "last_tested": null }
    },
    "target_persona": "principal-engineer",
    "expert_test_history": []
  },
  "spaced_repetition": {
    "current_interval_days": 2,
    "next_review_date": "2025-01-15",
    "consecutive_successes": 0,
    "last_review_quality": null,
    "flagged_for_immediate_review": false
  },
  "pedagogy": {
    "approach": "framework-application",
    "current_phase": "Learn",
    "phase_progression": ["Learn", "Apply (Guided)", "Apply (Solo)", "Transfer", "Adapt"],
    "active_techniques": ["worked-example"],
    "techniques_tried": []
  }
}

Key fields to customize:

  • track_id — URL-safe slug (matches directory name)
  • display_name — Human-readable name
  • pillar — Must match an existing pillar slug (or create a new one)
  • topic_type — Maps to a pedagogy approach in approaches.md
  • pedagogy.approach — Which of the 10 approaches to use
  • spaced_repetition.next_review_date — Set to today so it's due immediately

3. Create Empty history.jsonl

touch learning/tracks/my-new-track/history.jsonl

4. Create lesson-plan.md

# Lesson Plan: My New Track

## Current Focus
Introduction to core concepts.

## Key Topics
- Topic A
- Topic B
- Topic C

## Exercises
1. Walk through a worked example of [concept]
2. Apply the framework to a familiar product
3. Compare two approaches and discuss tradeoffs

## Resources
- [Link to relevant material]

5. Regenerate the Dashboard

./scripts/update-dashboard.sh

See examples/custom-track-example/ for a complete working example.

Create Custom Expert Personas

Expert personas live in learning/pedagogy/expert-personas.md. To add a new one:

Template

### N. 🎭 Your Pillar — "Role Title at Company"

**Role:** Description of who this persona is and their expertise.

**Conversation opener:**
- "Opening line that tests a specific aspect..."
- "Alternative opener for a different angle..."

**Evaluation emphasis:**
- **Accuracy** and **Depth** are critical because...
- **Current Awareness** is tested through...
- **Opinion Formation** matters because...

**Example probing questions:**
- "Question that tests depth..."
- "Question that tests current awareness..."
- "Question that challenges opinions..."

Tips

  • Each persona should have a distinct "personality" — a skeptical engineer tests differently than a visionary CPO
  • Tie the persona to a specific pillar so the coach knows when to deploy it
  • Include 2-3 conversation openers so tests don't feel repetitive
  • Be specific about what each persona probes hardest

Modify Pedagogy Approaches

Teaching approaches live in learning/pedagogy/approaches.md. Each approach defines:

  1. Phase Progression — The 5 stages of learning
  2. Techniques — Specific teaching methods per phase
  3. Mastery Signals — How to know the learner has achieved the phase
  4. Phase-to-Dimension Mapping — Which mastery dimensions gate each phase transition

Adding a New Technique

Find the relevant approach section and add a row to the Techniques table:

| `my-new-technique` | Description of what this does | Best Phase |

Adding a New Approach

Add a new numbered section (##) following the existing pattern. You'll also need to add a corresponding Phase → Focus Dimension Mapping section at the bottom of the file.

Caution

The reflection engine reads approaches.md programmatically. Maintain the existing Markdown table formats and section headers to ensure compatibility.

Populate the Question Bank

The question bank in data/question-bank/ starts empty — you add your own questions.

Question Sources

Build your question bank from:

  • Your own interview experiences — After each real interview, add the questions you received
  • Job postings — Extract skill requirements and turn them into practice questions
  • Interview prep communities — Public forums, study groups, peer practice
  • Framework application — Create questions that exercise specific frameworks (CIRCLES, RICE, etc.)

Adding Questions

  1. Copy data/question-bank/template.json to a new file (e.g., product-sense.json)
  2. Fill in the questions array with your questions
  3. Tag each question with:
    • category and subcategory for organization
    • company_tags for company-specific prep
    • framework_tags to link to relevant frameworks
    • difficulty for session calibration

During Coaching Sessions

You can also tell the coach: "Add this question to the bank: How would you design a notification system for a healthcare app?"

The coach will create or update the appropriate file.

Pro Tips

  • After real interviews, immediately add questions while they're fresh
  • Tag questions by company to create targeted mock interview sessions
  • Rate difficulty honestly — the coach uses it to calibrate session intensity
  • Add follow-up questions you've encountered for more realistic drills

Extend the Knowledge Base

Adding Knowledge Sources

Drop articles, papers, and notes as .md or .txt files into the configured knowledge directories (default: knowledge/).

Then run:

./scripts/feed-knowledge-queue.sh

The pipeline will:

  1. Detect new files
  2. Score them for relevance (if an LLM CLI is available)
  3. Add them to learning/knowledge-queue.jsonl
  4. The next reflection cycle integrates them into track lesson plans

Recommended Knowledge Sources

Podcasts: Lenny's Podcast by Lenny Rachitsky is an outstanding source of PM knowledge. We recommend distilling episode notes organized by topic (product sense, metrics, growth, strategy, hiring, etc.) and adding them to knowledge/. The reflection engine will score and route the insights to the relevant learning tracks automatically.

Articles: Blog posts from product leaders, company engineering blogs, and industry analysis pieces.

Research papers: For technical tracks, add markdown summaries of relevant AI/ML papers. Include the paper title, key contributions, and your notes on PM implications.

Your own notes: After meetings, conferences, or conversations — capture insights and add them to the pipeline.

Configuring Knowledge Directories

Edit config.json to add multiple knowledge source directories:

{
  "paths": {
    "knowledge_sources": [
      "./knowledge",
      "/path/to/my/pm-notes",
      "/path/to/podcast-notes"
    ]
  }
}

Manual Topic Addition

To add a specific topic for the coach to teach you:

./scripts/add-topic.sh "transformer attention mechanisms"
./scripts/add-topic.sh "RICE prioritization" --tracks pm-frameworks

Add PM Framework References

Framework documents live in data/frameworks/. To add a new one:

  1. Create data/frameworks/your-framework.md
  2. Follow the existing structure:
    • Title and purpose
    • Core concepts with explanations
    • Worked example
    • "How to Use in Interviews" section
    • "When NOT to Use" section

The coach references these during framework teaching sessions. Good framework docs are 80-150 lines with clear headers and practical examples.

Change the LLM Model

For Reflections

Set the REFLECTION_MODEL environment variable:

REFLECTION_MODEL=claude-opus-4-5 ./scripts/run-reflection.sh

Or edit config.json:

{
  "model": {
    "default": "claude-sonnet-4-5",
    "deep": "claude-opus-4-5"
  }
}

For Coaching Sessions

This depends on your platform:

  • Claude Code: claude --model claude-opus-4-5
  • Cursor: Change model in Cursor settings
  • Windsurf: Change model in Windsurf settings

Recommended Models

Use Case Recommended Model Why
Coaching sessions Claude Sonnet 4.5+ Good balance of speed and depth
Reflections Claude Sonnet 4.5+ Handles complex multi-file analysis
Deep analysis Claude Opus 4.5+ Best for nuanced technique evaluation

Any model with strong instruction-following and file access should work. The system has been primarily tested with Claude models.

Adapt for Non-PM Domains

PM Coach's architecture is domain-agnostic. The PM-specific content lives entirely in:

  • Track definitions (learning/tracks/*/track.json)
  • Pedagogy approaches (learning/pedagogy/approaches.md)
  • Expert personas (learning/pedagogy/expert-personas.md)
  • Framework library (data/frameworks/)
  • System prompt (CLAUDE.md)

To adapt for another domain (e.g., Software Engineering, Data Science, Design):

  1. Replace tracks — Create tracks for your domain's skill areas
  2. Update pedagogy — Modify or create approaches suited to your domain's learning patterns
  3. Create expert personas — Design personas relevant to your field (e.g., "Senior Staff Engineer" for SWE, "Design Director" for Design)
  4. Replace frameworks — Add framework references for your domain
  5. Update CLAUDE.md — Modify the system prompt to reflect the new domain, coaching style, and scoring rubrics
  6. Update config.json — Point knowledge sources at relevant materials

The spaced repetition algorithm, session logging, reflection loop, and dashboard scripts work for any domain without modification.

Example Adaptations

  • Software Engineering Coach — Tracks: system design, algorithms, code review, architecture. Personas: Staff Engineer, Tech Lead, Hiring Manager.
  • Data Science Coach — Tracks: statistics, ML algorithms, experiment design, communication. Personas: Principal Data Scientist, VP of Analytics.
  • Design Coach — Tracks: visual design, UX research, interaction design, design systems. Personas: Design Director, UX Researcher, Product Partner.