Skip to content

CLAUDE.md

Latest commit

 

History

History
209 lines (166 loc) · 7.85 KB

CLAUDE.md

File metadata and controls

209 lines (166 loc) · 7.85 KB

PM Knowledge Base — CLAUDE.md

Project memory. Read this at the start of every session.

Last updated: 2026-04-09 — v2 template review after 88 lessons

What this project is

Two parts:

  1. KB (Knowledge Base) — 214 enriched markdown files from BrightChamps production systems. Source of truth for technical architecture, PRDs, competitive research, performance reviews. Lives in: technical-architecture/, product-prd/, research-competitive/, performance-reviews/ Rule: NEVER modify KB files. Read-only reference only.

  2. Concept Library — A PM learning curriculum. 9 modules, 88 lessons. Each lesson covers ONE universal concept. Written for 10 distinct PM profile types. Lives in: concept-library/ Architecture: v2 (3-level depth system — see below)

Curriculum architecture — v2

Every lesson has THREE self-contained depth levels. The dashboard shows one level at a time. Reader selects their level.

Level 1: Foundation

  • Who: non-technical business PM, aspiring PM, designer PM, MBA PM (tech modules)
  • Assumes: nothing. No prereqs.
  • Sections: F1 (world before), F2 (what it is + analogy), F3 (when PM encounters this)
  • Length: 600–900 words
  • Rules:
    • Every technical term defined on first use. No exceptions.
    • Exactly one physical everyday analogy before any mechanism.
    • Story first. Business impact before technical detail.
    • No jargon, no shortcuts.

Level 2: Working Knowledge

  • Who: growth PM, consumer startup PM, B2B enterprise PM, 2+ years experience
  • Assumes: Foundation (or existing PM context)
  • Sections: W1 (how it works), W2 (decisions it forces), W3 (engineer questions), W4 (real examples)
  • Length: 1200–2000 words (no cap — as long as the concept requires)
  • Rules:
    • W1: numbered steps, mechanism first. As detailed as needed to build a real mental model.
      • If W1 exceeds ~400 words: include at least one comparison table or summary box.
      • Long W1 prose with no visual break = D1 critique every time. Don't do it.
    • W2: 3–5 real tradeoffs. Each must end with a recommendation that includes a specific threshold, number, or condition — not just "it depends on your context." That is a fail.
    • W3: 6–8 specific questions. Each includes what the answer reveals.
    • W4: 3–5 named companies. Format: Company | What they did | Hard metric (number, %, timeframe). At least 3 of 5 examples must have a hard number. Narrative-only examples don't pass QA.

Level 3: Strategic Depth

  • Who: ex-engineer PM, senior PM, head of product, AI-native PM
  • Assumes: Working Knowledge
  • Sections: S1 (what breaks), S2 (system connections), S3 (what senior PMs debate)
  • Length: 600–1000 words
  • Rules:
    • No hand-holding. Debates, not definitions.
    • S1: failure modes with mechanism and PM prevention role.
    • S2: explicit links to other curriculum concepts, how understanding compounds.
    • S3: genuine intellectual tension. What changed in last 2 years. What AI is doing to this concept. S3 must include at least one reference to a shift in the last 2 years — if nothing changed, say why.

Frontmatter schema — v2

Every lesson must use this exact frontmatter:

---
lesson: [Concept Name]
module: [number — name]
tags: [tech | product | business]
difficulty: [foundation | working | strategic]  # default starting level
prereqs:
  - [lesson-id] — [one line why]
writer: [staff-engineer-pm | senior-pm | cfo-finance | gtm-lead]
qa_panel: [comma-separated personas]
kb_sources:
  - [path/to/kb/file.md]
profiles:
  foundation: [list of PM profiles]
  working: [list of PM profiles]
  strategic: [list of PM profiles]
status: draft | ready
last_qa: [YYYY-MM-DD]
---

Writer personas

persona modules voice
staff-engineer-pm 01, 02, 03, 04, 09 Direct. Gives opinions. Names failure modes before happy path.
senior-pm 05, 06 Conversational. Story-driven. Opinionated frameworks.
cfo-finance 07 Precise numbers. Mechanism before metric. Connects to P&L.
gtm-lead 08 User-first. Distribution-first. Opinionated on sequencing.

Full persona definitions: .claude/skills/writer-personas.md

QA panel assignment

QA panel comes from _CONCEPT_INDEX.md qa_panel column — not from tags alone.

Cross-pairing rule (always enforced):

  • staff-engineer-pm writer → Junior PM Reader always in panel
  • senior-pm writer → Staff Engineer always in panel (keeps tech claims honest)
  • cfo-finance writer → GTM Lead always in panel (catches when finance ignores distribution)
  • gtm-lead writer → CFO/Finance Lead always in panel (catches when growth ignores margin)

Full QA rules: .claude/skills/lesson-qa.md

Lesson section reference

ID Name Level Required
F1 The world before this existed Foundation Yes
F2 What it is — definition + analogy Foundation Yes
F3 When you'll encounter this as a PM Foundation Yes
W1 How it actually works Working Yes
W2 The decisions this forces Working Yes
W3 Questions to ask your engineer Working Yes
W4 Real product examples Working Yes
S1 What breaks and why Strategic Yes
S2 How this connects to the bigger system Strategic Yes
S3 What senior PMs debate Strategic Yes

Section heading format (canonical)

Always use em-dash style: ## F1 — [Custom title], ## W1 — [Custom title], etc. The analogy belongs inside F2, not as a separate F3. F3 is always "When you'll encounter this as a PM."

Known technical debt (early v1→v2 migrations)

These 5 lessons use correct ═══ level markers but bare section names (no F/W/S prefix). They render correctly in the dashboard. Lower priority for rewrite unless content is also stale.

  • 01-apis-and-integration/api-authentication.md
  • 01-apis-and-integration/rate-limiting-and-throttling.md
  • 01-apis-and-integration/webhooks-vs-polling.md
  • 01-apis-and-integration/what-is-an-api.md
  • 03-infrastructure-and-devops/cicd-pipelines.md

Folder structure

pm-knowledge-base/
├── CLAUDE.md                 ← you are here
├── _REGISTRY.md              ← KB index (read-only)
├── tasks.md                  ← update every session
├── decisions.md              ← architectural decisions log
├── memory.md                 ← long-term patterns
│
├── technical-architecture/   ← KB (READ ONLY)
├── product-prd/              ← KB (READ ONLY)
├── research-competitive/     ← KB (READ ONLY)
├── performance-reviews/      ← KB (READ ONLY)
├── assets/                   ← KB (READ ONLY)
│
├── concept-library/
│   ├── _CONCEPT_INDEX.md     ← lesson manifest (pipeline controller)
│   ├── 01-apis-and-integration/
│   ├── 02-databases-and-data/
│   ├── 03-infrastructure-and-devops/
│   ├── 04-ai-and-ml-systems/
│   ├── 05-product-fundamentals/
│   ├── 06-metrics-and-analytics/
│   ├── 07-business-and-monetization/
│   ├── 08-gtm-and-growth/
│   └── 09-security-and-scale/
│
└── dashboard/
    └── index.html            ← local study dashboard

Active rules

  • Never modify files outside concept-library/, dashboard/, tasks.md, decisions.md, memory.md, CLAUDE.md, .claude/
  • Never delete any KB file or _REGISTRY.md
  • Always update tasks.md at the end of every session
  • When writing lessons: pull real examples from KB files, never invent
  • All lessons must follow v2 architecture. No v1 lessons (8 sections, no levels) are acceptable.
  • QA is mandatory before any lesson is marked ready
  • _CONCEPT_INDEX.md is the pipeline controller — it drives all lesson writes

KB notebook IDs (for MCP calls)

NB1: 2c11acfe-fafc-4bb8-8381-c92ca2cb4353 NB5: 256e3b28-eea5-40b6-aebb-dd4833c8b870

Owner

Suneet Jagdev — Senior PM, BrightChamps, Bengaluru