PM Jarvis
AI-powered copilot for Product Managers built for Claude Code. Clone the repo, fill out your context, and let it learn your company, team, and writing style — then hand it recurring work to run on a schedule.
Philosophy
Context over prompting
It stores your company knowledge, writing style, stakeholder profiles, and past decisions, then pulls from them on every task.
Workflows, not chat
You run purpose-built commands instead of typing open-ended questions. 87 of them, covering strategy, research, PRDs, metrics, meetings, launches, and code.
Ship the draft, then iterate
A one-page PRD that ships Monday beats a ten-page spec that ships never. Refine documents over time instead of chasing one perfect draft.
What you get
Pre-built context library
Templates for company info, writing styles, stakeholder profiles, and strategic frameworks (7 Powers, JTBD, PLG Iceberg, Growth Loops, and more)
87 slash commands
One command per recurring PM task — PRDs, meeting notes, pricing analysis, code reviews, and a compounding knowledge base
7 sub-agents
Multi-perspective reviews from engineer, designer, executive, legal, UX researcher, skeptic, and customer voice
Knowledge assets
63 curated interview questions, validated survey templates, canvas templates, and 139 AI prompt references
Persistent memory
File-based memory system with hooks that inject context every session — your copilot remembers what happened last time
Automation layer
Wrap any skill in a scheduled routine that runs on its own and reports back over Slack. Unattended runs are guardrailed: self-notify only, no messaging others or editing shared docs without approval
How it works
PM Jarvis runs with Claude Code, an AI assistant that reads and writes files on your computer. Because it can see your whole project folder, it references your company info, past PRDs, stakeholder profiles, and writing style on every task — not just the current chat.
The system is built on three layers:
Project knowledge
Company info, writing styles, stakeholder profiles, strategy frameworks, and past decisions — all stored as simple text files in your context library.
Skills
87 registered slash commands for recurring tasks. Each one knows which context files to pull from, how to format the output, and what skill to offer next.
Sub-agents
7 specialized reviewer personas (engineer, designer, executive, legal, UX researcher, skeptic, customer voice) that review your work and catch blind spots.
The rule is simple: read your context freely before producing output, and never write back without asking first. A lightweight memory system carries key facts across sessions.
What’s new
The biggest shift: PM Jarvis went from answering when asked to doing recurring work on its own.
Scheduled routines
Skills now run on their own schedule and report back over Slack — the automation layer above, applied to any recurring task.
A periodic-review cascade
Weekly, monthly, and quarterly reviews that roll up into each other. Each tier pre-fills from the work below it and asks only the judgment calls that need a human.
Radar & sweep skills
Stay ahead of the work: action-item sweeps across every connected source, a “loose threads” radar for dropped conversations, and a “what should I be worried about?” scan.
Governance maturity
Every skill now ships with pass/fail evals and an improvement journal, so the system gets more reliable as it grows.
How is it different from plain ChatGPT or Claude?
| Regular AI tools | PM Jarvis |
|---|---|
| Forgets your company context | Always knows your business, team, and stakeholders |
| Generic output | Matches your writing style automatically |
| One-shot responses | Evolves PRDs through multiple stages |
| Single perspective | Multi-agent reviews from 7 perspectives |
| No tool integration | Connects to analytics, PM tools, and docs via MCPs |
| Answers only when prompted | Runs recurring work on its own schedule |
Example workflows
Commands chain into workflows. A few common ones:
Daily planning
Strategic Planning
Pre-Launch Risk
Design-Led Prototyping
Design Audit & Improve
Second Brain (compounding knowledge)
Build with AI (for PMs who code)
New Feature Research
Periodic Review Cascade
Stay Ahead of the Work (radar)
All 87 commands
Expand a category, then any command to read its full instructions and copy it.