No phase ownership
The same model validates problems, writes acceptance criteria, and signs off on launch. Corners get cut where nobody is watching.
Agent skill for product management
AI project management skill for agents.
PM Butler is a household staff of eight specialists for Cursor, Claude Code, and Copilot: they challenge assumptions, flag scope creep, and write the documentation you keep putting off.
Works with Cursor, Claude Code, Copilot, and any harness that accepts skills.
pm-butler / sprint review
Watch
Concern
Blocker
The problem
Why can't one generic assistant run a project?
Forty-seven tabs. Twelve half-finished ideas. Zero decision log. Discovery, scope, architecture, QA, and launch need different instincts. Without a shared vocabulary, agents guess and decisions vanish into chat history.
No paper trail
Two hundred messages later, nobody can answer why you built it that way. Post-mortems become archaeology.
No productive tension
Real teams argue about speed vs. quality. A single agreeable assistant lets scope creep and untested launches slide.
The ensemble
Eight personas. One household.
Each member owns an SDLC phase, speaks in a distinct voice, and flags concerns at three severity levels: Watch, Concern, Blocker.
01
Head Butler
Orchestration
02
The Oracle
Discovery
03
The Sculptor
Requirements
04
The Architect
Design
05
The Alchemist
Development
06
The Gatekeeper
QA
07
The Shipper
Launch
08
The Librarian
Documentation
Sample output
What Oracle actually produces.
Structured output you can paste into your repo, not chat filler.
View sample discovery response
Problem statement (draft)
Checkout abandonment spikes on mobile between steps 2 and 3.
Evidence needed
- Current step-2 to step-3 drop-off rate
- Median time on step 2 (baseline)
- 5+ user verbatims, not one Slack message
CONCERN Problem still uses "slow" without a metric. Logged as OQ-014.
Recommendation Stay in discovery until baseline exists, or explicitly accept the risk in DECISIONS.md.
Checkout abandonment spikes on mobile between steps 2 and 3.
Evidence needed
- Current step-2 to step-3 drop-off rate
- Median time on step 2 (baseline)
- 5+ user verbatims, not one Slack message
CONCERN Problem still uses "slow" without a metric. Logged as OQ-014.
Recommendation Stay in discovery until baseline exists, or explicitly accept the risk in DECISIONS.md.
Ready to drop in
Built for how agents actually work.
01
Respects your project context.
Run /pm-butler init once. The skill reads your stack, team size, and constraints before any persona speaks.
> Walking project...
ok PROJECT.md loaded
ok 8 personas routed by SDLC phase
-> Matching your workflow. Not inventing one.
ok PROJECT.md loaded
ok 8 personas routed by SDLC phase
-> Matching your workflow. Not inventing one.
02
Living documentation, automatic.
Project Edition maintains eleven markdown files as you work: decisions, scope, risks, changelog, and more. You do not stop to document; the staff does it while helping you ship.
03
Concern severity that sticks.
Every persona watches for drift. Sculptor tracks scope creep as a percentage. Gatekeeper escalates when blockers pile up. Orange concerns are not optional.
04
Handoffs mirror real projects.
Oracle validates the problem. Sculptor locks requirements. Architect designs the system. Alchemist protects the sprint. Gatekeeper signs off. Shipper launches. Librarian records what actually happened.
05
Alliances and rivalries by design.
Shipper and Gatekeeper will disagree. That tension is a feature. The ensemble produces better outcomes than one voice saying yes to everything.
06
Decision logs you can commit to git.
npx pm-butler docs init scaffolds eleven Project Edition markdown files. npx pm-butler memory init adds structured JSON or SQLite under .claude/pm-butler. Personas update both during commands — your audit trail lives in the repo, not in chat history.
Choose your path
Free plugin or raw packs.
Install the skill from GitHub or npm at no cost. Buy on Gumroad if you want the full markdown files, PDF quick-reference card, bonus solo persona, Notion builder, and extras without a harness.
The Ensemble
Free on GitHub + npm
8 personas across the full SDLC
Phase coverage, team dynamics, coach and review modes, and structured output you can paste into Jira, Notion, or your repo.
- Full ensemble system prompt
- Bonus solo PM Butler persona for lighter workflows
- Quick reference PDF and concern severity framework
- Interactive Notion builder (React)
Project Edition
Free on GitHub + npm
The AI that documents everything you will not
Seven specialists maintain eleven living markdown files. Every interaction produces the answer, the doc update, and any concerns.
- DECISIONS.md, SCOPE.md, RISKS.md, and eight more
- Scope creep index and decision reversal tracking
- Ideal for solo builders and consultants
Complete Household
Gumroad bundle
Ensemble + Project Edition together
Full SDLC coverage plus automatic documentation. Everything in both editions, persona cards, and quick-reference PDF.
- Both system prompts, ready to paste
- All persona cards and cheat sheets
- Best value if you want files, not just the plugin
The language
Commands your agent already understands.
Each maps to one persona or workflow. Pick any to see what it does.
/pm-butler init
Create PROJECT.md with stack, team, constraints, and current SDLC phase.
Get started
How do I install PM Butler?
Three install paths plus decision logging: Claude Code plugin, npm skill installer, Project Edition markdown, and optional persistent memory.
Claude Code plugin
Add the marketplace, install the plugin, then invoke with the plugin namespace.
/plugin marketplace add tedrubin80/pmbutler
/plugin install pmbutler
Invoke: /pmbutler:pm-butler init
Cursor / Claude via npm
Copies the skill to ~/.cursor/skills/pm-butler/ and ~/.claude/skills/pm-butler/.
npx pm-butler skills install
npx pm-butler skills check
Package: npmjs.com/package/pm-butler
npx pm-butler docs init
Scaffolds PROJECT.md + 11 decision-log files. Use --dir docs/pm for a subfolder.
Project-local (Cursor)
Commit the skill into your repo so the whole team gets the same PM vocabulary.
.cursor/skills/pm-butler/
Copy from npmPMButler/skills/pm-butler/, or see npm-packages/v2/project-local/.
Persistent memory
Structured decision and concern log in a git-friendly folder or local SQLite database.
npx pm-butler memory init --enable --path .claude/pm-butler
npx pm-butler memory log --type decision --title "..."
Folder backend: JSON under .claude/pm-butler/memory/. SQLite: --backend sqlite (Node 22+). Toggle in PROJECT.md.
First run
Initialize project context. Creates PROJECT.md from the bundled template.
/pm-butler init
Claude Code: use /pmbutler:pm-butler init instead.
After install, run /pm-butler route (or /pmbutler:pm-butler route in Claude Code). Scaffold decision logs with npx pm-butler docs init and optional npx pm-butler memory init --enable. Commit markdown and memory files to git. Source: GitHub · npm · Raw packs: Gumroad.
FAQ
Common questions about PM Butler.
Quick answers for search, sharing, and first-time installs.
What is PM Butler?
PM Butler is a free AI project management skill for Cursor, Claude Code, and Copilot. It gives your agent eight specialized personas—Head Butler, Oracle, Sculptor, Architect, Alchemist, Gatekeeper, Shipper, and Librarian—each owning a phase from discovery through launch.
How do I keep a record of decisions PM Butler helps make?
Run npx pm-butler docs init for human-readable markdown (DECISIONS.md, SCOPE.md, etc.) and optionally npx pm-butler memory init --enable --path .claude/pm-butler for structured JSON or SQLite memory. Personas update these during discover, scope, and handoff. Commit everything to git.
What is the difference between docs init and persistent memory?
docs init creates eleven Project Edition markdown files for readable decision logs in git. memory init adds a programmable store (folder JSON or SQLite) for structured entries you can query with memory log and memory list. Use both together for the best audit trail.
How do I install PM Butler?
Three options: run npx pm-butler skills install for Cursor or Claude personal skills; add the Claude Code plugin with /plugin marketplace add tedrubin80/pmbutler then /plugin install pmbutler; or copy the skill folder to .cursor/skills/pm-butler/ in your repo for project-local use.
Is PM Butler free?
Yes. The plugin and npm skill installer are free on GitHub and npm. Gumroad sells optional raw markdown packs, PDF quick-reference cards, and bonus content if you want files without a coding harness.
Which AI coding tools work with PM Butler?
PM Butler works with Claude Code (plugin marketplace), Cursor (personal or project-local skills), Claude desktop skills via npm, and any harness that loads Agent Skills or system prompts from the bundled markdown.
What commands does PM Butler support?
Core commands include init, route, discover, scope, design, sprint, review, launch, document, concerns, handoff, and crisis. Run /pm-butler init first to create PROJECT.md, then /pm-butler route to match your current SDLC phase.
About
Order from the chaos.
I build agent skills, prompt packs, and systems for people who think fast and document slow. PM Butler is a household of eight AI personas that cover discovery through launch: they challenge scope creep, flag risks, and write the docs you keep skipping.
Flexible enough for real projects. Structured enough to ship. Generic AI gives answers. PM Butler gives you a staff.
Also on Gumroad
AI Literacy prompt guides
PDF prompt guides for teams and developers. Sold alongside PM Butler on Gumroad.
Clarity pack
Chaos-to-Clarity and Document Decoder guides for teams new to AI-assisted work.
Developer pack
Code Review Partner and Codebase Onboarding Assistant for engineering workflows.
Strategy pack
AI Project Team and project strategy guides for PMs running cross-functional initiatives.