A general-purpose AI agent is a blank slate. Product on Purpose fills it in: open-source skills, tools, and libraries that turn an everyday agent into a capable partner for product work, one composable piece at a time. Tools for people who tinker or live in their editors, agents, and file systems live in the sibling org, Prisant Labs.

Version badges on this page update themselves. Everything else is written by hand, including a plain status line for each project.

Status

• 🟢 Stable: 1.0 or later, or a live service
• 🟡 Beta: released, but not yet 1.0
• 🟠 Pre-release: no release yet, so you build from source
• 🟤 Maintenance: works, but no longer developed

Tags

• 🚀 Start here: the best first project in its category
• 🆕 New: public for under 30 days
• 🧪 Experimental: likely to change, and its status line says how

Most of these install as Claude Code plugins. Add the marketplace once, then install any plugin below:

/plugin marketplace add product-on-purpose/agent-plugins

• 🧭 pm-skills · 🟢 Stable · 🚀 Start here
  68 plug-and-play product management skills covering the full product lifecycle

• 📐 product-lifecycle-templates · 🟡 Beta · 🧪 Experimental
  Governed document templates that ship with the research, the guide, and a worked example

• 🤔 thinking-framework-skills · 🟡 Beta · 🚀 Start here
  Evidence-graded thinking methods an agent can run to reason, not just chat

• 🔍 critique-skills · 🟡 Beta
  Critique that cites a published rubric on every finding, and publishes how often it is right

• 🎨 writing-style-catalog · 🟡 Beta · 🧪 Experimental
  Composable writing instructions so AI prose lands in the voice you actually want

• 🤖 agent-skills-toolkit · 🟢 Stable · 🚀 Start here
  A standard and toolkit for grading skill libraries to a Bronze/Silver/Gold bar

• 🧩 agent-plugins · 🟢 Stable
  The plugin marketplace to install from

• 🧰 pm-skills-mcp · 🟤 Maintenance
  The PM catalog as an MCP server

For product managers, and for the agents doing product work alongside them. New here? Start with pm-skills.

🟢 Stable · 🚀 Start here · Claude Code, Codex, Cursor, and more

/plugin install pm-skills@product-on-purpose

Get professional-grade product management work from your agent without teaching it the method first. It hands your agent a curated set of best-practice product workflows to run on demand, end to end.

What's inside: 68 skills, 6 sub-agents, 12 workflows, and 203 sample outputs.

• Stop prompting PM work from scratch. Each skill is a best-practice workflow (PRDs, hypotheses, user stories, sprint facilitation) that you invoke by name instead of re-explaining the method every time.
• Covers the whole lifecycle. From Foundation Sprint and Design Sprint at the fuzzy front end through discovery, delivery, and iteration, with sub-agents and workflow orchestrators that chain skills into full routines.
• Remembers your project, if you let it. Opt-in project memory lets later skills reuse what earlier ones produced, so you stop pasting the same personas and context into every run.
• Quality you can see. 200+ real-world sample outputs set the bar, and CI-enforced contracts keep every skill conformant as the catalog grows.

Recent releases:

• v2.35.0 (Oct 2026): A pilot found that none of three skills beat a plain model with no skill, so the planned headline was dropped. The release fixes 41 broken reference links, and two templates now match their skills.
• v2.34.0 (Oct 2026): Two planned AI skills lost a test against an analyst with no skill, so neither shipped. What survived went into deliver-prd, and samples now mark invented figures as fictional.
• v2.33.0 (Aug 2026): PRDs, ADRs, and instrumentation specs gain sections for AI features, covering model behavior, evaluation, trace privacy, and model choice. Three user-reported fixes also landed, including prioritization that fits small teams.

Status: actively released, with two minor releases in October. Beyond Claude Code, it installs into Cursor, Copilot, Cline, and other agents with npx skills add product-on-purpose/pm-skills.

🟡 Beta · 🧪 Experimental · Claude Code, MCP clients, and more

/plugin install product-lifecycle-templates@product-on-purpose

Stop inventing the structure of every PRD, roadmap, and postmortem from scratch. Where pm-skills runs the method, this decides what the resulting artifact should actually look like, and defends that shape with research rather than taste.

What's inside: 54 template bundles in eleven families, two skills, and an MCP server.

• A bundle, not a blank file. Every document type ships five pieces together: the blank shape you fill in, a companion explaining why it is shaped that way, a fast operator card, a fully worked example, and machine-readable metadata an agent can route on.
• Sized to your context window. Every type offers a lean variant and most also offer full, each with an approximate token count, so an agent pulls the version that fits the job instead of blowing context on a template built for a different scale.
• Reachable from any agent. Two skills fill a template and grade a finished document, and an MCP server lets any MCP-aware agent search the catalog and fetch any of 105 template variants.
• Cited, and gated on staying that way. Shapes are argued from primary sources, and CI enforces the claims: bundle completeness, link health, research logs, manifest freshness, and even the repo's own self-reported counts.

Recent releases:

• v0.24.0 (Oct 2026): Two one-page canvases arrive as the 53rd and 54th bundles. The Lean UX Canvas frames one initiative, and the Lean Canvas states a whole product's business model as beliefs to test.
• v0.23.0 (Oct 2026): A user research report reads out one study and recommends what to do next. It pairs with the user research plan, so a study can be planned and reported from the same library.
• v0.22.0 (Oct 2026): A user research plan opens a new family for research documents. The fifth efficacy run is published too: it was void on its main test, and the evidence still supports no quality claim.

Status: the repository calls itself experimental. All 54 bundles are marked beta, and the project claims none of them is proven. You can also clone the repository, add it with npx skills add product-on-purpose/product-lifecycle-templates, or read every bundle online.

For anyone who wants an agent to reason, review, or write with a method rather than by feel. New here? Start with thinking-framework-skills.

🟡 Beta · 🚀 Start here · Claude Code, Codex, and more

/plugin install thinking-framework-skills@product-on-purpose

Stop letting your agent improvise its reasoning on hard problems. This library is a toolbox of decision-making moves that your agent actually executes, each one graded so you know how far to trust it.

What's inside: 63 frameworks, 4 meta-tools, 9 recipes, and 3 subagents.

• Agents reason better with a method. Canonical thinking tools (premortems, first principles, parallel-perspective review) packaged so an agent runs the actual move, not a vague impression of it.
• Honest about the evidence. Every skill carries a transparent grade, from replicated research down to practitioner heuristic, so you know how far to trust it. No laundered statistics.
• Always ends in an artifact. Each run hands back something usable, a risk register, an option matrix, an argument map, rather than more prose.
• Hand off a whole routine. In Claude Code, a decision stress test or a reasoning audit can run as a subagent and return only the finished brief, so the intermediate work stays out of your conversation.

Recent releases:

• v0.20.0 (Nov 2026): The trust page corrects three claims about its skill-selection eval. An experiment to shorten skill descriptions failed a threshold set in advance, so it did not ship.
• v0.19.0 (Oct 2026): Six skills gain an optional step, and the belief-update routine now asks you to record your confidence before new evidence arrives. Four skills also hand set-aside material to a fresh context.
• v0.18.6 (Oct 2026): The premortem skill now says that, in one study, it lowered confidence in a flawed plan, instead of claiming it reduces overconfidence. The issue tree and pyramid principle evidence pages also report weaker findings.

Status: actively released, with three minor releases since mid-October. Other agents can use npx skills add product-on-purpose/thinking-framework-skills, and the subagents run in Claude Code only. It grades Advanced (Gold) under the agent-skills-toolkit Standard.

🟡 Beta · Claude Code

/plugin install critique-skills@product-on-purpose

Ask a general-purpose model to critique your work, and you get fluent, confident, forgettable commentary. These skills answer to a published rubric instead, and report how often they actually catch what is there.

What's inside: 7 skills, 1 reviewer subagent, 120 criteria, and a scorecard from 851 recorded runs.

• Every finding cites a source. Each skill operationalizes a published external standard (WCAG 2.2, Nielsen's heuristics, Diataxis, the Toulmin model, the Federal Plain Language Guidelines), so every defect it raises carries a permanent criterion ID you can look up and argue with.
• Findings are records, not prose. Output conforms to a frozen JSON Schema, so a critique can be diffed, filtered, tracked across revisions, or handed to another tool rather than read once and lost.
• It publishes its own scorecard. Measured against a corpus with deliberately planted defects, unflattering numbers first: recall, precision, and run-to-run consistency, with the weak axes named rather than buried.

Recent releases:

• v0.2.0 (Oct 2026): A seventh skill, critique-forms, reviews HTML forms for how easy they are to fill in, especially on a phone. A documentation site now shows every skill, criterion, and measured result.
• v0.1.6 (Aug 2026): A critique that could hang while searching your drives for its own scripts now finishes. Every published score also shows its run-to-run spread.
• v0.1.5 (Aug 2026): A script that ships with each skill now does the final tally and refuses to return an invalid report. On a small model, usable reports rose from 2 in 7 runs to 3 in 4.

Status: the repository calls itself pre-release, and its latest release shipped in October. The current benchmark does not reproduce the original Sonnet figures, and the project says so beside them. It grades Convergent (Silver) under the agent-skills-toolkit Standard.

🟡 Beta · 🧪 Experimental · Claude Code, Claude.ai, and Claude Desktop

/plugin install writing-style-catalog@product-on-purpose

Get an agent to write in the voice you want, not the one it defaults to. The catalog turns "make it sound professional" into a precise, reusable instruction you can drop onto any writing task.

What's inside: 98 entries on four axes (Voice, Tone, Style, and Format), 1,176 worked examples, 158 diff-pairs, and 14 recipes.

• Make AI writing stop sounding like AI. Compose precise, reusable instructions from named building blocks instead of retyping "make it sound professional" and hoping.
• Four orthogonal axes. Mix Voice, Tone, Style, and Format independently to dial in exactly the register and shape a piece needs.
• Worked examples, not vibes. Every entry ships with samples that show what it actually produces, so you can see a style before you commit to it.
• Learns your default, then suggests the rest. One skill captures your personal default style, and another recommends a combination for a situation you describe.

Recent releases:

• v0.15.0 (Oct 2026): The marketer voice ships for landing pages and launch copy, which brings the catalog to 98 entries. The adherence evidence page now leads with all three of its results.
• v0.14.0 (Sep 2026): Every entry now carries an honest review status. All 97 are machine-verified, which means they pass every automated check but the maintainer has not yet read each one.
• v0.13.0 (Aug 2026): Diff-pairs now cover all twelve anchor topics instead of five, and each of the 158 pairs carries written commentary.

Status: the repository calls itself experimental. The entry schema is frozen, but the catalog, skills, and docs may still change without notice. For Claude.ai or Claude Desktop, download the ZIP from the latest release.

For people who author skill libraries, and for anyone installing from this one. New here? Start with agent-skills-toolkit.

🟢 Stable · 🚀 Start here · Claude Code, Codex, npm, and GitHub Actions

/plugin install agent-skills-toolkit@product-on-purpose

Find out whether your skill library is actually good, and exactly what to fix next. It defines what a great, multi-agent skill library looks like, then gives you the tooling to prove yours measures up.

What's inside: 27 skills, 7 subagents, and a grader with 35 checks.

• For people building skills, not just using them. A toolkit and a normative Standard for authoring skill libraries that work across Claude Code and Codex from a single source.
• A quality bar you can climb. Grades a whole library against a tiered Bronze/Silver/Gold rubric and returns a burndown of exactly what blocks the next tier.
• Deterministic, not vibes. The grader gives the same answer on your laptop and in CI, because no model is involved, and it returns real exit codes. The repo self-validates at Gold in its own CI as the proof.
• Runs wherever you check. Use it as a plugin, run it once with npx agent-skills-toolkit, or add the GitHub Action so every push gets graded.

Recent releases:

• v1.23.0 (Nov 2026): Evaluation reports gain a health number beside the tier, and every category reads uncalibrated for now. A fourth advisory screen looks for substance, and neither addition can change a grade.
• v1.22.0 (Oct 2026): A new askit-onboard skill guides a new adopter from a survey of their plugin to a roadmap. An advisory layer also reports portability, preload, and security notes beside the tier without changing it.
• v1.21.0 (Oct 2026): gen-manifest --write now refuses to drop or replace hand-written manifest fields that library.json does not supply. A grade also says when a plugin pins a newer Standard than the toolkit ships.

Status: actively released. The npm package and the GitHub Action follow each release directly, and the marketplace pin can trail behind, so compare the badges above.

🟢 Stable · Marketplace · Claude Code

/plugin marketplace add product-on-purpose/agent-plugins

Add one marketplace, and every Product on Purpose plugin becomes a one-line install, including future ones.

What's inside: only the catalog. There's a single marketplace.json, and each plugin keeps its own repository, versions, and issues.

• One front door to everything. The Claude Code marketplace registry for the whole portfolio: add it once and every Product on Purpose plugin is a single command away.
• Pinned, and watched. Each entry pins a release commit, so nothing reaches you until it is re-pinned here, and a daily check flags any pin that falls behind its plugin's latest release.

Recent catalog changes:

• Nov 6, 2026: thinking-framework-skills moved to a release whose trust page corrects three claims about its skill-selection eval. It adds no skills.
• Nov 1, 2026: agent-skills-toolkit moved to a release whose evaluation report adds a health number beside the grade. It cannot change a grade or a badge.
• Oct 30, 2026: product-lifecycle-templates moved to a release with two new bundles, the Lean UX Canvas and the Lean Canvas, its 53rd and 54th.

Status: live, and all six listed plugins install now. You add the marketplace by its repository path, and you install each plugin by the marketplace name, @product-on-purpose.

These projects still work, but they are no longer developed and get security fixes only.

🟤 Maintenance · Any MCP client

npm install -g pm-skills-mcp

Give an agent that speaks the Model Context Protocol the product management catalog as native tools, with no file setup.

What's inside: an MCP server with 59 tools: 40 skills, 11 workflows, and 8 utilities.

Recent releases:

• v2.9.3 (May 2026): A security patch cleared every open dependency advisory, so npm audit now reports zero vulnerabilities. The tools themselves did not change.
• v2.9.2 (May 2026): The server entered maintenance mode with the 40-skill catalog embedded. Feature development is paused until there is demand for it.

Status: in maintenance mode since May 2026, with security patches and critical fixes only. It carries 40 skills while pm-skills now has 68, so new users should install pm-skills instead. Needs Node 18 or later. To ask for development to resume, open a discussion.

Each project carries its own license: Apache-2.0 for most, while writing-style-catalog licenses its code under Apache-2.0 and its content under CC BY 4.0. Issues and pull requests are welcome in each project's own repository.