nanopm
A PM skill pack for AI coding agents. Runs the PM workflow end-to-end — company + product context, the three external signals, the planning cycle (challenge, strategy, roadmap, PRD), and the day-to-day ops (an adversarial challenge, a jam with Nano, the standup, the weekly update) — inside the agent you already use. Keeps typed state across sessions. Won't write a PRD until you name what would prove your bet wrong.
Works with Claude Code, Mistral Vibe, and OpenAI Codex. One install command picks up whichever you have. An optional macOS viewer browses the artifacts and re-runs skills on the Claude Code backend, for the moments you'd rather not be in a terminal.
Built on the SKILL.md standard from gstack. nanopm is the PM layer.
Why this exists
Building the wrong thing fast is the failure mode AI coding agents make easy. They'll ship whatever you describe — they don't ask if it's the right thing.
The PM work that catches that (challenge, strategy, falsification, roadmap, retro) usually lives in tools that don't know your code, lose context between sessions, and don't talk to each other. ChatGPT for the strategy. Notion for the PRD. Linear for the tickets. Nothing compounds.
nanopm runs the whole cycle where you already work, with two structural properties:
- Every artifact lands in typed, schema-validated state. Each skill writes to
~/.nanopm/projects/{slug}/{type}.jsonl. The next skill reads from there. No grep on markdown, no silent corruption, no tail -1 | cut -d'|' parsing.
- Adversarial gates on the claims that matter. Every strategy needs a falsifiable bet. Every NOW roadmap item needs a measurable outcome with a timeframe. Every PRD needs a paragraph stating what would prove it wrong. Each gate is a subagent against a rubric plus a state validator. A skill cannot complete unless a well-formed record actually lands.
One command runs the full pipeline: /pm-run. Or invoke any skill standalone.
Example
You: /pm-challenge-me
nanopm: Q1: What are you building? (one sentence)
You: An AI agent harness for financial advisors — automates
client reporting, portfolio commentary, and compliance checks.
nanopm: Q2: Who is it actually for?
You: Independent RIAs managing $50M–$500M books, no ops team.
...11 questions later...
nanopm: CHALLENGES.md written.
Biggest gap: you're solving the reporting problem, but the
job your users are actually hired to do is winning and
keeping clients. Faster reports don't help if advisors
can't explain the AI's output to a nervous client.
Question you're avoiding: would a compliance officer
sign off on AI-generated commentary going out unsupervised?
Recommended next: /pm-discovery
Who it's for
- Solo founder running PM yourself. Audit your own product before you keep building.
- Engineer without a PM partner. Strategy, roadmap and PRDs in your editor, not in a separate tab.
- Small team without a dedicated PM seat. A repeatable cycle that doesn't depend on a hire.
If you already have a polished product process and a PM you trust, you don't need this.
Install
# Auto-detect installed agents (Claude Code, Mistral Vibe, OpenAI Codex)
curl -fsSL https://raw.githubusercontent.com/nmrtn/nanopm/main/setup | bash
# Target a specific host
curl -fsSL https://raw.githubusercontent.com/nmrtn/nanopm/main/setup | bash -s -- --host=claude
curl -fsSL https://raw.githubusercontent.com/nmrtn/nanopm/main/setup | bash -s -- --host=vibe
curl -fsSL https://raw.githubusercontent.com/nmrtn/nanopm/main/setup | bash -s -- --host=codex
# Install to all hosts at once
curl -fsSL https://raw.githubusercontent.com/nmrtn/nanopm/main/setup | bash -s -- --host=all
| Host | Skills install to | Invocation |
|---|
| Claude Code | ~/.claude/skills/ | /pm-* commands |
| Mistral Vibe | ~/.vibe/skills/ | /pm-* commands |
| OpenAI Codex | ~/.codex/skills/ | /pm-* commands |
Requirements: One of: Claude Code, Mistral Vibe, or OpenAI Codex. python3 (standard on macOS/Linux).
Claude Code plugin (optional, Claude Code only)
Claude Code users can install nanopm as a native plugin instead of the curl | bash script:
/plugin marketplace add nmrtn/nanopm
/plugin install nanopm
