The PM's Claude Code Playbook

Free playbook by ProductBuilt — the complete, reusable system for running product management with Claude Code: workspace structure, CLAUDE.md, agents, slash commands, Notion export, security, and the weekly operating rhythm. Built by a PM with 15 years in regulated fintech. → productbuilt.io

The AI drafts. You decide.

Last verified against Claude Code documentation: July 2026. The tool moves fast — if a detail differs on your machine, the structure still holds; check code.claude.com/docs for the current syntax.

Who this is for

Product managers and product owners who want Claude Code — Anthropic's agentic tool — as their product operations engine: competitive research, PRDs, discovery synthesis, prototyping, data analysis, and automated exports to Notion.

You don't need to be an engineer. You need a terminal, an IDE, and one afternoon to set up a structure you'll reuse for years.

Part 0 — The complete PM workspace, visualized

This is the structure this whole playbook builds. Set it up once, copy it for every product you ever work on.

flowchart TD
    A[📁 my-product-workspace] --> B[📄 CLAUDE.md<br/>context & rules — the brain]
    A --> C[📁 .claude/]
    C --> C1[📁 agents/<br/>your specialized team]
    C1 --> C1a[competitor-analyst.md]
    C1 --> C1b[prd-writer.md]
    C1 --> C1c[meeting-synthesizer.md]
    C1 --> C1d[hypothesis-tracker.md]
    C1 --> C1e[data-analyst.md]
    C1 --> C1f[roadmap-challenger.md]
    C --> C2[📁 skills/<br/>your shortcuts: /weekly-update, /new-prd]
    C --> C3[📄 settings.json<br/>permissions & guardrails]
    A --> D[📄 .mcp.json<br/>connection to Notion & tools]
    A --> E[📁 research/]
    E --> E1[📁 meetings/ — raw notes, transcripts]
    E --> E2[📁 interviews/ — user research]
    E --> E3[📁 competitors/ — snapshots by month]
    A --> F[📁 prds/ + TEMPLATE.md]
    A --> G[📁 data/ — CSV & analytics exports]
    A --> H[📁 templates/ — email, update, one-pager]
    A --> I[📁 decisions/ — decision log]
    A --> J[📁 prototypes/ — HTML mockups]
    B -.reads at every session start.-> K((Claude Code))
    C1 -.invoked by task.-> K
    D -.grants tool access.-> K
    K ==exports on your command==> L[(🗂️ Notion<br/>Research / PRDs / Hypotheses / Decisions)]

The same structure as an annotated tree — copy this into your terminal notes:

my-product-workspace/
│
├── CLAUDE.md                  ← THE BRAIN. Context + rules. Read at every session start.
│
├── .claude/
│   ├── agents/                ← THE TEAM. One .md file per specialist.
│   │   ├── competitor-analyst.md
│   │   ├── prd-writer.md
│   │   ├── meeting-synthesizer.md
│   │   ├── hypothesis-tracker.md
│   │   ├── data-analyst.md
│   │   └── roadmap-challenger.md
│   ├── skills/                ← THE SHORTCUTS. Reusable procedures as /commands.
│   │   ├── weekly-update/SKILL.md
│   │   ├── new-prd/SKILL.md
│   │   └── synthesize-meeting/SKILL.md
│   └── settings.json          ← THE GUARDRAILS. What Claude may/may not do alone.
│
├── .mcp.json                  ← THE CONNECTIONS. Notion (and other tools).
│
├── research/                  ← INPUTS. Everything raw goes here.
│   ├── meetings/              ← drop transcripts & notes, dated: 2026-07-13-steerco.txt
│   ├── interviews/            ← user research transcripts
│   └── competitors/           ← monthly snapshots: 2026-07-landscape.md
│
├── prds/                      ← OUTPUTS. Specs & one-pagers.
│   └── TEMPLATE.md            ← your PRD structure (provided below)
│
├── data/                      ← NUMBERS. CSV exports from analytics/CRM.
│
├── templates/                 ← FORMATS. Email structure, exec update, one-pager.
│   ├── email-structure.md
│   └── exec-update.md
│
├── decisions/                 ← MEMORY. One file per significant decision.
│   └── 2026-07-10-pricing-model.md
│
└── prototypes/                ← ARTIFACTS. HTML mockups Claude builds for you.

Why each folder exists (the logic, not just the layout):

Folder Role Rule
CLAUDE.md What Claude knows & how it behaves Under 500 tokens. Stable. Never put agents here.
.claude/agents/ Specialists with one job each One file = one job. Whitelist tools explicitly.
.claude/commands/ Prompts you repeat weekly If you've typed it 3 times, make it a command.
.claude/settings.json Permissions Claude asks before destructive actions.
.mcp.json External tools (Notion) Project-level, versioned with the workspace.
research/ Raw inputs, dated Date-prefix every file: YYYY-MM-DD-name.
prds/ Finished documents Generated → reviewed by you → exported.
data/ CSV/analytics exports Claude queries these with pandas, no SQL needed.
templates/ Your document formats Agents reference these by path.
decisions/ Decision log The file future-you will thank you for.
prototypes/ Clickable HTML mockups Test with users before speccing for eng.

Setup in 30 minutes — the checklist

Do this once, in order. Every part below explains its step in detail.