Download: claude-pm.zip ↓ — direct snapshot from the GitHub repo. Or open the repo: github.com/e3xler/claude-pm

---

1) The problem

You start a new project. You open Claude. You start typing:

„I'm building a small SaaS for X. Stack is Next.js, Supabase, Vercel. Reply in English, keep it informal, no buzzword bingo. For UI use our colours — primary is a dark green, secondary a warm beige. Tone: direct, technical, …"

Three days later you start a new session. You type the same thing again. And again. And again.

That's what a CLAUDE.md is for. But an empty CLAUDE.md isn't a setup either — you need a structure Claude recognises: where's the strategy? where are the personas? where's the brand voice?

This template is that structure.

2) What's in the template

claude-pm/
├── CLAUDE.md                # Central briefing (15 sections, with placeholders)
├── README.md                # How to use the template
├── .claude/
│   ├── settings.json        # Permissions, hooks, model defaults
│   ├── commands/            # 5 slash commands
│   └── agents/              # 3 sub-agents
├── docs/
│   ├── product-strategy.md  # Vision, bet, outcomes
│   ├── personas.md          # Who is the user
│   ├── tech-stack.md        # What we use, why, when we'd switch
│   ├── architecture.md      # System design, data flows
│   ├── brand-voice.md       # How we sound, what we never say
│   ├── brand-colors.md      # Palette, semantic use, WCAG
│   ├── design-system.md     # Tokens, components, patterns
│   ├── security.md          # Threat model, checks, incident response
│   ├── glossary.md          # Project-specific terms
│   ├── decisions/           # ADR template
│   └── research/            # Structure for interviews + insights
├── .env.example
└── .gitignore

Three layers, cleanly separated:

Layer	What it does
CLAUDE.md	Bootstrap layer — Claude reads this before every prompt. Holds language, stack, workflow rules.
.claude/	Automation layer — slash commands like /kickoff or /ship, plus specialised agents like brand-guardian.
docs/	Knowledge layer — the long documents CLAUDE.md links to. Brand, strategy, architecture.

3) Quick start

Step 1: Get the template

# Option A: ZIP snapshot
curl -L https://github.com/e3xler/claude-pm/archive/refs/heads/main.zip -o claude-pm.zip
unzip claude-pm.zip
mv claude-pm-main my-project
cd my-project

# Option B: Git clone (if you use Git anyway)
git clone https://github.com/e3xler/claude-pm.git my-project
cd my-project
rm -rf .git && git init

Step 2: Ask Claude for help

claude

Inside Claude:

/kickoff

The /kickoff command checks what's filled, what's missing, and recommends the next steps — depending on your project phase (discovery, prototype, MVP, production).

Step 3: Fill CLAUDE.md (10–20 min)

CLAUDE.md has {{PLACEHOLDERS}} everywhere. Replace them — or have Claude do it:

Read CLAUDE.md. Ask me the questions you need answered to fill
the placeholders. One at a time, not all at once.

Step 4: Fill the brand docs (before UI work)

When the project gets a UI, these three documents matter most — otherwise Claude builds generic Tailwind templates:

• docs/brand-voice.md — tone, banned phrases, AI patterns we avoid
• docs/brand-colors.md — token tables, WCAG contrasts
• docs/design-system.md — spacing, typography, component inventory

4) The 5 slash commands

Command	What it does
/kickoff	Setup diagnosis and next-step recommendation
/new-feature	Structured feature kick-off: discovery → plan → build → verify
/brand-check	Audit copy or UI against the brand docs
/design-review	Designer-eye QA — hierarchy, spacing, states, "looks AI-generic" smell
/ship	Pre-deploy checklist — tests, build, env, migrations, UI smoke

Commands live as Markdown in .claude/commands/<name>.md. Edit them or add new ones — Claude picks them up automatically.

5) The 3 sub-agents

Sub-agents are specialised Claude personas you spawn for one focused task. In the template:

Agent	For what
product-strategist	Questions scope, frames hypotheses, checks outcome fit. Use when a feature request comes in and you're not sure it's the right lever.
brand-guardian	Reviews copy and UI against brand voice, colors and design system. Use after any major copy or component change.
design-critic	Designer-eye QA. Catches visual inconsistency, weak hierarchy and "AI-generic" patterns — before they ship.

Spawn them via the Agent tool or directly in Claude Code via the sub-agent picker.

6) What CLAUDE.md covers

The 15 sections of the template CLAUDE.md:

1. Project context — name, status, owner, live URLs
2. Audience & use cases — personas, top 3 jobs to be done
3. Tech stack & architecture — layer table with rationale
4. Language & communication — reply language, address, style
5. Code conventions — naming, comments, errors
6. Brand & design — link to the three brand docs
7. Product strategy — North Star, outcomes, what we explicitly DON'T build
8. Workflow & pipeline — branch strategy, commits, PR rules
9. Testing — coverage targets per test type
10. Security — non-negotiables
11. Performance targets — LCP, INP, CLS, bundle budget
12. How Claude should work in this project — proactive permissions, what not without confirm
13. Available custom commands — list
14. Available custom agents — list
15. Important paths — file tree

All with placeholders you replace or delete per project.

7) What you should customise

Required (otherwise the template doesn't work for you)

• CLAUDE.md → replace every {{PLACEHOLDER}} or delete the section
• docs/product-strategy.md → at minimum vision, persona, North Star
• docs/tech-stack.md → what you actually use

Strongly recommended (once UI is involved)

• docs/brand-voice.md → voice attributes, banned phrases
• docs/brand-colors.md → real token values
• docs/design-system.md → pick a style direction, set tokens

Optional (fill iteratively)

• docs/architecture.md — when the system grows
• docs/security.md — before the first production deploy
• docs/glossary.md — when terms become ambiguous
• docs/decisions/ — ADRs for every non-trivial call
• docs/research/ — user interviews and insights

8) How Claude reads the template

On the first prompt Claude Code automatically loads:

1. CLAUDE.md from the project root → into the system context
2. Files under .claude/commands/ → available as slash commands
3. Files under .claude/agents/ → available as spawnable sub-agents

The files under docs/ are not automatically loaded — that would be too much context. Claude reads them on demand, when CLAUDE.md references them (e.g. "For UI work read docs/brand-voice.md").

That's the trick: short CLAUDE.md (cheap in context), deep docs/ (only read when relevant).

9) Best practices

Keep CLAUDE.md short

It's loaded with every prompt. Rule of thumb: under 250 lines. Anything longer belongs in docs/ and gets linked.

Fill brand docs before UI work

Otherwise Claude builds generic Tailwind templates. With brand voice/colors/system you get output that fits the brand — not the AI default.

Extend slash commands

Need the same workflow repeatedly? Drop it as a new slash command under .claude/commands/<name>.md. Examples from daily use:

• /standup — generate a standup update from the latest git log
• /changelog — write CHANGELOG.md from commits since the last tag
• /handoff — write HANDOVER.md before the weekend

Sub-agents for recurring roles

If you often need the same "second opinion" (strategy, brand, design, security), put it as an agent under .claude/agents/<name>.md. Benefit: specialised prompt, scoped tools, no topic switch in the main context.

Memory is alive

If Claude makes the same mistake three times or keeps asking the same question — that's a CLAUDE.md update. Add it now, don't re-explain it next time.

10) Quick reference

Question	Answer
Where's the repo?	github.com/e3xler/claude-pm
How do I download it?	ZIP or git clone
Licence?	Free to use, adapt, share
Does it work without Claude Code?	CLAUDE.md and docs/ yes — slash commands and sub-agents need Claude Code
What's the first command?	/kickoff in Claude Code
How long does setup take?	10–20 min for CLAUDE.md, 1–2 h if you take the brand docs seriously
Can I delete sections?	Yes — everything is modular. Strip what doesn't fit.

11) Next steps

• Download the ZIP or clone the repo
• Rename the folder, re-init .git
• Run claude in the folder, then /kickoff
• Fill CLAUDE.md (or let Claude ask you the questions)
• Fill the brand docs when UI comes in
• Add your own slash commands and agents as patterns emerge