pmstack

pmstack is an open-source toolkit built by Ryan Alberts, a Staff PM in Agentic AI, that gives Product Managers a structured set of AI-powered commands and skills for the Claude ecosystem. It runs inside the claude.ai web app, the Claude desktop app, the mobile app, and the Claude Code CLI — with no server infrastructure, no telemetry, and no third-party dependencies beyond the user's existing Claude account. The project describes itself as "part learning playground, part tactical skillset" and is explicitly inspired by Gstack.

What It Is

pmstack packages common PM workflows — writing PRDs, running competitive analyses, designing AI feature evals, drafting briefs, and running weekly retrospectives — as markdown-based "skills" or slash commands that Claude reads as instructions. There are two installation paths: non-technical PMs upload skill files to a Claude Project via the claude.ai UI in roughly 30 seconds; technical PMs install via a one-line curl command into ~/.claude/ for use with Claude Code. The same 14 skills are available on both paths, producing identical markdown or YAML artifacts.

Core Capability Areas

pmstack organizes its commands into three logical layers:

• Spec creation: /prd, /competitive, /compare — translate customer signals and market data into structured artifacts
• Measurement: /metrics, /vibe-test, /eval, /run-eval, /transcript-review, /eval-self — implement Anthropic's published eval framework as PM-runnable commands, producing YAML test suites, scored summaries, and diagnosis memos
• Communicate & orchestrate: /brief, /sprint — package work for different audiences or chain multiple commands in sequence

A Routines layer (/eval-drift, /premortem, /weekly, /launch-readiness, /lint) adds recurring operating discipline, including a /loop scheduler that can run weekly drift watches and workspace audits automatically.

The Eval Framework Angle

A central design choice in pmstack is implementing Anthropic's published eval framework for AI features as commands a PM can run without writing engineering scaffolding. The README draws directly on Anthropic's "Demystifying Evals for AI Agents" article and maps pmstack commands to Anthropic's 8-step eval roadmap (documented in docs/anthropic-framework.md). Key concepts surfaced include pass@k vs. pass^k scoring, grader types (code, model, human), transcript review, and drift watching — vocabulary the README presents as separating "an AI PM from a traditional PM."

Audience and Dual Install Paths

The project explicitly targets two personas: a Technical PM who lives in a terminal and wants consistent muscle memory across PM artifacts via Claude Code slash commands, and a Non-technical PM who works in claude.ai web, desktop, or mobile and wants the same toolkit without ever opening a terminal. The README states neither persona is "presumed to be AI-pilled," positioning pmstack as an on-ramp to AI fluency rather than a tool for already-converted practitioners.

Open-Source Status and Lineage

pmstack is licensed under Creative Commons Attribution Share Alike 4.0 International (CC-BY-SA-4.0) and hosted publicly on GitHub. The repository was created in April 2026 and last updated in May 2026. The README credits Gstack (by Garry Tan) and Andrej Karpathy's LLM Wiki as inspirations. The primary language is listed as Python, though the skills themselves are plain markdown files. The project had 1 star and 0 forks at the time of the available data snapshot, indicating it is early-stage and community-building.