PM-Skills
An open-source product management skill library from Product on Purpose, using 68 composable skills, 12 workflows, and six sub-agent definitions to structure outputs from research and problem framing through PRDs, measurement, and retrospectives, with installation paths for Claude Code and other agents.
Project overview
PM-Skills is a product management Agent Skill library maintained by Product on Purpose. It is not a single generic “write a PRD” prompt. The repository divides product work into 68 self-contained skills, typically combining a method, an output template, and a completed example, then connects them with 12 workflows, 11 commands, and six Claude Code sub-agent definitions. Coverage runs from interview synthesis, market sizing, and problem framing through PRDs, user stories, experiments, instrumentation, release work, and retrospectives. Claude Code plugin, open Skills CLI, and manual clone paths are documented, but reading, writing, networking, and sub-agent behavior always depend on the permissions and capabilities of the host.
Core capabilities
Discovery and problem framing
Interview synthesis, competitive analysis, journey mapping, market sizing, problem statement, hypothesis, and opportunity-tree skills turn source material into traceable product evidence and decision inputs.
Delivery specifications and scope boundaries
deliver-prd, deliver-user-stories, and deliver-acceptance-criteria provide complete steps, templates, and examples while requiring goals, scope, verification, dependencies, and risks.
Experiments, measurement, and retrospectives
Experiment design, instrumentation specs, dashboard requirements, experiment results, OKR grading, survey analysis, and retrospective skills cover the post-launch learning loop.
Multi-skill workflows
Twelve workflows define recommended order and handoffs for feature kickoff, Lean Startup, Triple Diamond, Foundation Sprint, Design Sprint, and related processes.
Skill creation and quality maintenance
Builder, validator, and iterator skills form a create-validate-iterate loop, while the repository includes trigger fixtures, CI contracts, and a sample library for structural and output checks.
Installation and usage
Use the official marketplace for Claude Code. Codex, Cursor, Copilot, Cline, and other Agent Skills hosts can use the Skills CLI or clone the repository and configure host discovery. Confirm install scope and hooks first, then verify skill discovery with a non-sensitive, read-only task.
Let an AI Agent install it
Send this prompt to Codex, Claude Code, or another AI agent that can work with your local environment.
Install PM-Skills from https://github.com/product-on-purpose/pm-skills. First read README.md, QUICKSTART.md, SECURITY.md, PRIVACY.md, LICENSE, .claude-plugin/plugin.json, .codex-plugin/plugin.json, and hooks/README.md. Explain the target host, installation scope, directories that will be written, whether hooks will be registered, and the difference between source manifest 2.33.0 and the latest GitHub Release v2.32.0. Do not install, overwrite existing skills or host configuration, or request any product, customer, or production credentials until I confirm. After confirmation, run only the repository's documented command for that host. If using `npx skills add product-on-purpose/pm-skills`, disclose the Skills CLI anonymous opt-out telemetry first. Verify only with a fictional product topic and deliver-prd template structure, then list changed files. Stop and ask before user-directory writes, project-file changes, networking beyond installation, hook activation, project-memory creation, or access to real business data.- A target host that supports plugins or Agent Skills
- Network access to the public GitHub repository
- A Node.js environment able to run npx when using the Skills CLI
- A defined policy for the skills install directory, project writes, and sensitive product data
/plugin marketplace add product-on-purpose/agent-plugins
/plugin install pm-skills@product-on-purpose- 1Choose the host and install path
Use the marketplace for Claude Code and the Skills CLI or a clone for other Agent Skills hosts. Decide between project and user scope first.
- 2Review versions and permissions
Check the README, manifests, SECURITY, PRIVACY, and hooks. Pin v2.32.0 or a commit SHA when reproducibility matters.
- 3Run the repository's exact command
Use the documented command for the selected host without rewriting the source, providing product data, or adding account credentials.
- 4Verify skill discovery
Invoke deliver-prd or define-hypothesis with a fictional topic and confirm the host loaded the target SKILL.md and template structure.
- 5Open project context gradually
Only after validation, introduce redacted research and project context; separately review project-memory writes, file changes, and hooks.
After installation, confirm the host can discover deliver-prd or define-hypothesis. Use a fictional product topic for the first read-only test and check that the output contains goals, success metrics, scope, dependencies, and risks from the template. Do not provide customer interviews, roadmap details, production credentials, or project write access during the first test. Claude Code users should also inspect hook status; other hosts should not assume command or sub-agent parity.
- Claude Code first adds the product-on-purpose/agent-plugins marketplace and then installs pm-skills@product-on-purpose.
- The Skills CLI supports --skill for subsets, but skills route to other skills and referenced destinations should be checked too.
- The static repository declares no telemetry; the README separately documents anonymous, opt-out telemetry in the Skills CLI.
- For reproducibility, pin the v2.32.0 Release or an explicit commit SHA instead of treating main's 2.33.0 manifest as a released tag.
Use cases
User research synthesis
Turn redacted interviews, surveys, or market material into themes, evidence, confidence, and follow-up validation work.
PRDs and engineering handoff
Produce a structured product specification with goals, metrics, stories, scope, dependencies, risks, and verifiable requirements.
Product experiments and measurement
Plan hypotheses, experiments, instrumentation, dashboards, and outcome reviews while preserving data and methodology limitations.
Team workflow standardization
Use fixed templates, examples, and workflows to align artifact structure across product managers and agent sessions.
Assessment
PM-Skills earns inclusion through an inspectable skill structure, explicit full-lifecycle classification, extensive samples, documented cross-host installation, and recent maintenance—not through its 590 stars. The sampled deliver-prd skill goes beyond document formatting by requiring success metrics, scope boundaries, AI behavior evaluation, and an execution contract for work handed to an executor without authoring context. SECURITY and PRIVACY explain local hooks, project memory, and static-content data boundaries. The main caveats are unequal host capabilities, the need to fact-check natural-language output, and a source manifest that is ahead of the formal Release. This review was read-only; no install or functional test was performed.
Why it may be useful
- All 68 skill directories, templates, and examples are directly inspectable
- Broad coverage across discovery, delivery, measurement, and iteration
- Clear Claude Code and cross-agent installation paths
- Apache-2.0 license with security and privacy documentation
- Recent commits plus CI, fixtures, and maintenance utilities
What to know first
- Natural-language output still requires factual and product-decision review
- Hooks, commands, and sub-agent support differ by host
- Project memory and inputs can contain sensitive business data
- Source manifest 2.33.0 has no matching GitHub Release yet
- Installation and cross-host behavior were not tested
README
PM-Skills
Overview
An open-source product management skill library from Product on Purpose, using 68 composable skills, 12 workflows, and six sub-agent definitions to structure outputs from research and problem framing through PRDs, measurement, and retrospectives, with installation paths for Claude Code and other agents. A curated library of 68 best-practice, plug-and-play product management skills covering the complete product lifecycle, plus templates, workflows, and 200+ sample outputs.
Getting started
- Use the official marketplace for Claude Code. Codex, Cursor, Copilot, Cline, and other Agent Skills hosts can use the Skills CLI or clone the repository and configure host discovery. Confirm install scope and hooks first, then verify skill discovery with a non-sensitive, read-only task.
- The README divides 68 skills into 30 product-phase skills, 11 foundation skills, 12 utility skills, and 15 workshop-method skills.
- A representative skill combines SKILL.md, references/TEMPLATE.md, and references/EXAMPLE.md so method, structure, and sample jointly constrain the output.
- After installation, confirm the host can discover deliver-prd or define-hypothesis. Use a fictional product topic for the first read-only test and check that the output contains goals, success metrics, scope, dependencies, and risks from the template. Do not provide customer interviews, roadmap details, production credentials, or project write access during the first test. Claude Code users should also inspect hook status; other hosts should not assume command or sub-agent parity.
Configuration
/plugin marketplace add product-on-purpose/agent-plugins
/plugin install pm-skills@product-on-purpose