The 32 Best AI Agent Skills for Product Managers in 2026

An AI agent skill is a structured, reusable capability packaged in the SKILL.md format that an agent can load, follow, and produce a specific artifact from. That single idea changes how product managers work with AI: instead of re-typing the same prompt every time you need a PRD or a jobs-to-be-done map, you reach for a named skill that produces a consistent result. This guide covers the 32 skills in Product OS — an MIT-licensed, evidence-first operating system for product work maintained by BuildBetter at github.com/buildbetter-app/product-os. Every skill runs on pasted artifacts alone; BuildBetter is an optional acceleration layer for the evidence-heavy ones. Below, we break down all 32 skills across four packs, explain how to choose them by the question you're asking, and show how to install and evaluate them.

What an AI Agent Skill Actually Is (and Why It Beats a Prompt Template)

An agent skill is a portable, version-controlled capability — not a chat message you paste once and lose. Written in the SKILL.md format, a skill defines exactly what an agent should do, what inputs it needs, and what artifact it must produce. The agent loads it, follows it, and hands back a defined output.

The difference from a prompt template comes down to reproducibility. A pasted prompt drifts with every edit and every person who touches it. A skill produces the same exact-state artifact across runs because it lives in a repository, carries a version, and can be reviewed like code.

Skills beat prompt templates on four concrete axes:

• Version-controlled: changes are tracked, reviewable, and reversible.
• Composable: one skill's output feeds the next, so discovery flows into strategy flows into delivery.
• Exact states, not implied completion: a skill produces a defined artifact, not a plausible-looking narrative that sounds finished.
• Portable: a SKILL.md file runs in any compatible agent harness, so there's no lock-in.

Consequential actions require explicit approval — a skill won't quietly mutate a roadmap or send a customer note without a human signing off.

The practical framing is jobs-to-be-done applied to your tooling. When you have a job — define JTBD, write a PRD, run a premortem — you reach for the skill named for that job instead of re-deriving the prompt from scratch. The rest of this article covers 32 focused skills, organized into four packs of eight, plus a Complete pack of all 32.

About Product OS: MIT-Licensed, Vendor-Neutral, No BuildBetter Required

Product OS is an open, evidence-first operating system for product work, published under the MIT license at github.com/buildbetter-app/product-os. It contains exactly 32 focused agent skills in the common SKILL.md format, usable from any compatible agent.

The design is deliberately open. Every workflow has an artifact-only fallback: paste in your interview notes, your feedback exports, or your metrics, and the skill runs. When BuildBetter MCP is connected, evidence-heavy skills — the ones that synthesize research or cluster feedback — can draw on organization Skillsets as an optional acceleration layer. Nothing requires a BuildBetter account to run any skill.

The principles behind Product OS are worth stating plainly:

• Evidence before confidence — claims are grounded in source material, not asserted.
• Outcomes before features — the goal is a moved metric, not a shipped checkbox.
• Decisions before documents — the deliverable is a defensible choice, not a deck.
• Exact states instead of implied completion — an artifact is done or it isn't.
• Progressive disclosure and small focused skills — each skill does one job well.
• Vendor-neutral core with optional acceleration — the open thing is genuinely open.
• Explicit approval for consequential mutations — humans stay in the loop.

The openness matters because it makes Product OS a reference, not a pitch. You can read the skills, fork them, and run them without touching anything BuildBetter sells.

Pack 1 — Discovery: 8 Skills for Understanding the Problem Space

The Discovery pack turns raw customer signal into structured understanding of jobs, journeys, and opportunities. These are the skills you reach for before you've committed to building anything.

The eight Discovery skills are:

• plan-product-discovery — scope a discovery effort with clear questions and methods.
• plan-customer-interviews — design an interview guide that surfaces real jobs, not opinions.
• synthesize-customer-research — turn transcripts and notes into structured findings.
• cluster-feedback-into-jobs — group volume of feedback into coherent job themes.
• define-jobs-to-be-done — articulate the jobs customers are hiring your product for.
• map-customer-journey — chart the end-to-end experience and its friction points.
• analyze-competitors — position alternatives against the jobs you've defined.
• map-product-opportunities — convert findings into a ranked opportunity space.

Reach for plan-customer-interviews and synthesize-customer-research early, when you're still learning the shape of the problem. Move to cluster-feedback-into-jobs and define-jobs-to-be-done once you have enough feedback that manual reading stops scaling.

Synthesis and clustering are where an evidence layer earns its keep. Both skills run fine on pasted artifacts, but they improve when a retrieval layer over conversations and tickets can pull the underlying quotes. If you have hundreds of support tickets, cluster-feedback-into-jobs connected to BuildBetter reads across the whole set instead of the handful you had time to review.

Pack 2 — Strategy: 8 Skills for Deciding Where to Play

The Strategy pack converts validated problems into vision, positioning, ICP, and prioritized bets. This is where discovery findings become defensible choices about where to invest.

The eight Strategy skills are:

• validate-product-problem — confirm a problem is real, frequent, and worth solving.
• define-product-vision — set a durable north star for the product.
• shape-product-strategy — choose how you'll win in your chosen market.
• define-product-positioning — articulate why your product wins for whom.
• define-ideal-customer-profile — specify the accounts worth serving.
• size-market-opportunity — quantify the addressable space.
• prioritize-product-opportunities — rank bets against value and effort.
• size-product-bet — estimate the scope and payoff of a specific investment.

Run validate-product-problem before committing resources — it's the cheapest insurance you can buy against building the wrong thing. Reach for prioritize-product-opportunities and size-product-bet when you're planning a roadmap or defending a decision to leadership.

"Decisions before documents" shows up most clearly in this pack. The output of shape-product-strategy or prioritize-product-opportunities is not a polished slide deck. It's a defensible choice with the rationale attached: this bet over that one, for these reasons, backed by this evidence. A deck can hide a weak decision behind good design. A strategy skill makes the decision the deliverable.

Pack 3 — Delivery: 8 Skills for Building the Right Thing

The Delivery pack moves you from strategy to shipped work with clear requirements, metrics, and verified outcomes. These skills carry a decision through to something customers can use — and prove it worked.

The eight Delivery skills are:

• design-product-experiment — structure a test that produces a clear answer.
• define-product-metrics — pick the measures that tell you if you're winning.
• write-product-requirements — produce a PRD grounded in the validated problem.
• scope-product-release — define what ships and what doesn't.
• plan-outcome-roadmap — sequence work around outcomes, not feature lists.
• run-product-premortem — surface failure modes before launch.
• align-product-stakeholders — get the room agreeing on the same plan.
• verify-product-outcomes — confirm the work actually moved a metric.

Use write-product-requirements and scope-product-release for release planning. Run run-product-premortem before any high-risk launch — it's an hour that saves a quarter. And run verify-product-outcomes to close the loop after shipping.

This pack is where "outcomes before features" and "exact states instead of implied completion" become concrete. verify-product-outcomes forces evidence that a metric moved — not a narrative that it probably did. That distinction separates a trustworthy agent workflow from a plausible-looking one. Anyone can write a launch recap that reads like success. A skill that demands the numbers, and refuses to mark the job done without them, keeps the whole team honest.

Pack 4 — Operate: 8 Skills for Running and Evolving the Product

The Operate pack launches, prices, grows, evaluates fit, and makes governance decisions across a product's life. These are the skills for the long game — after the first ship, when a product needs to be run.

The eight Operate skills are:

• plan-product-launch — coordinate the go-to-market moment.
• design-pricing-packaging — structure how the product is sold and priced.
• design-growth-loop — build a mechanism that compounds usage.
• evaluate-product-market-fit — measure whether the product has found its market.
• run-product-review — hold a recurring, evidence-based check on health.
• write-product-decision — record a choice and its rationale durably.
• manage-product-sunset — retire a product or feature responsibly.
• review-product-plan — audit a plan against its evidence and goals.

Reach for plan-product-launch and design-pricing-packaging at go-to-market. Use evaluate-product-market-fit and manage-product-sunset for portfolio decisions about what to double down on and what to cut. And treat write-product-decision as your durable record — the artifact you return to in six months when someone asks why you chose what you chose.

For teams that want the full operating system, the Complete pack installs all 32 skills together, so Discovery, Strategy, Delivery, and Operate compose end to end.

How to Choose Skills by the Question You're Asking

The fastest way to pick a skill is to name the situation you're in and map it to the pack. Here's a lookup table for the most common product jobs, with BuildBetter as the evidence layer behind the feedback-heavy ones.

Lookup vs. population vs. absence questions

Evidence-heavy skills succeed or fail based on the shape of the question, not the quality of the search. Three shapes matter:

• Lookup question: "What did Acme say about SSO?" A specific fact, answered well by top-k retrieval.
• Population question: "How many customers complained about onboarding this quarter, ranked by severity?" This requires reading the whole corpus, not the top few matches.
• Absence question: "Which accounts never mentioned the feature?" You cannot retrieve evidence that does not exist — absence can't be found by search.

Be fair to vector search: it's the right tool for lookup questions. The failure mode is applying top-k retrieval to population and absence questions, where it structurally undercounts or simply can't surface what was never said.

This is where BuildBetter's AOT (ahead-of-time comprehension) acceleration is relevant. On a benchmark of 6,018 call recordings and 8,533 support conversations, AOT reached 99.0% coverage (95% CI 98.3–99.7%) on population and absence questions, versus 27.9% for hybrid search and 11.3% for keyword search — at $0.03 per question versus $33.55 for a full-corpus scan. So cluster-feedback-into-jobs over 200 tickets returns a complete count, not a sampled one. Details are at buildbetter.ai/aot.

How to Install and Evaluate These Skills

All 32 skills live in the Product OS repository at github.com/buildbetter-app/product-os, MIT licensed and ready to clone. You can install a single pack, a single skill, or the Complete pack of all 32.

To find, install, and evaluate skills with real numbers, use skillrank (github.com/buildbetter-app/skillrank) — an open-source CLI that benchmarks agent skills so you can compare them instead of guessing. It's the honest way to check whether a skill produces the artifact you expect before you wire it into a team workflow.

Engineers on your team may also want the companion coding-skills repository at github.com/buildbetter-app/skills, which ships packs for Claude Code, Codex, Cursor, Copilot, Gemini, Windsurf, and Amazon Q.

To restate the terms: every skill is MIT licensed, every workflow has an artifact-only fallback, and no BuildBetter account is required to run any of them. BuildBetter MCP is an optional acceleration layer for the evidence-heavy skills — nothing more.

Frequently Asked Questions

How many skills are in Product OS?

Exactly 32, organized into four packs of eight — Discovery, Strategy, Delivery, and Operate — plus a Complete pack that installs all 32 together for teams that want the full operating system.

Do I need a BuildBetter account to use Product OS?

No. Product OS is MIT licensed and vendor-neutral, and every workflow has an artifact-only fallback that runs on pasted content. BuildBetter MCP is an optional acceleration layer that only benefits evidence-heavy skills — nothing requires an account to run.

What is the difference between an agent skill and a prompt template?

A skill is version-controlled, composable, produces exact states rather than implied completion, requires explicit approval for consequential actions, and travels across agent harnesses in the SKILL.md format. A prompt template is a one-off you re-paste each time, with no versioning, composition, or portability guarantees.

Which skills should a PM start with?

Begin with Discovery's synthesize-customer-research and define-jobs-to-be-done to turn raw signal into structured understanding, then move to Delivery's write-product-requirements and verify-product-outcomes to convert decisions into shipped, verified work.

What agent does Product OS work with?

Any SKILL.md-compatible agent harness. The skills are vendor-neutral by design, so there is no lock-in to a particular assistant or platform.

Why does the retrieval layer matter for some skills?

Lookup questions are answered well by vector search. Population and absence questions require reading the whole corpus, which is where AOT's ahead-of-time comprehension applies — reaching 99.0% coverage where hybrid search reaches 27.9%.

Make churn optional.

Product OS gives your team a shared language for product work. BuildBetter gives the evidence-heavy skills a complete view of every call, ticket, and Slack thread — so cluster-feedback-into-jobs counts every complaint and verify-product-outcomes proves the metric actually moved. Book a demo and make churn optional.