Product discovery skills
Product Discovery Skills-md for AI Agents
Product discovery skills-md are SKILL.md files that teach Claude, Codex and other AI agents to learn what customers actually need before a line of code is written, using honest interview methods and real jobs-to-be-done frameworks rather than a plausible guess. Install one and the agent asks about behaviour in the past, not opinions about the future, and traces every finding back to a real quote.
6 of 6 skills published, version 1.0.0.
Where to start: begin with customer-interviews to plan, run and synthesise conversations that surface how people behave, then jobs-to-be-done to model why they buy, switch or churn and where the unmet need sits. Use continuous-discovery-habits to turn that into a weekly rhythm, assumption-mapping to rank the riskiest beliefs behind an idea, and concept-testing to check demand before building. Once there is a design, usability-testing shows whether people can actually use it. All hand off to opportunity-solution-trees in the roadmapping and prioritisation set, which picks the problem worth solving.
Customer interviews
Plans, runs and synthesises customer interviews that stay in the past and in specifics, so an agent hears how people actually behave rather than collecting opinions. Covers The Mom Test rules, Teresa Torres’s story-based opening, a behavioural screener, saturation instead of an arbitrary count, one snapshot per interview, and synthesis that separates pain from pull and never calls a single mention a pattern.
Jobs to be done
Models the job a customer hires a product to do, choosing the right school for the question: Christensen and Moesta’s forces and switch interviews for why people buy or churn, and Ulwick’s outcome-driven innovation for measuring unmet need. Carries the four forces with the push-plus-pull rule, switch-interview recruiting, three solution-free statement formats, and the opportunity algorithm with its score bands.
Assumption mapping
Surfaces every belief a product idea depends on, sorts them into the five risk types (desirability, viability, feasibility, usability and ethical), and ranks them on a 2x2 of importance against evidence, so the team tests the one or two beliefs that could actually sink the idea first. Carries the story-map surfacing, the "We believe that..." hypothesis format, an evidence ladder where only behaviour counts, the four quadrant actions, and a map from each risk type to the cheapest sufficient test.
Continuous discovery habits
Sets up and runs a weekly product discovery rhythm instead of a one-off research sprint, following Teresa Torres's continuous discovery and Marty Cagan's cadence. Carries a week-by-week runbook: one product outcome, the trio or a fallback when there is no trio, automated recruiting with fallbacks for low-traffic and B2B teams, an interview snapshot each, a current opportunity solution tree, a weekly assumption test, and a decision step, measured by whether the cadence actually changes decisions.
Concept testing
Designs and runs the cheapest experiment that tests whether people want a product idea and will act on it, before building, drawing on Testing Business Ideas and lean validation. Carries the sequence (problem interview, landing page or fake door, pre-sale, then concierge or wizard of oz), the rule to weight behaviour over stated intent, pre-registered pass thresholds with real benchmarks, sample and stop rules, and an ethics guardrail for fake doors and wizard-of-oz tests.
Usability testing
Plans, runs and reads a usability test that shows whether people can use an existing design, not whether they want it. Carries method and sample set before running (about five per group for a qualitative round), scenario tasks that do not lead, a facilitator who does not rescue, behaviour and perception measured with task success, SEQ (about 5.5) and SUS (about 68), Nielsen 0 to 4 severity rated by more than one person with frequency as a rate, accessibility with the WCAG caveat, and consent and data guardrails.
No skill matches that search. Clear it or choose All.
What product discovery skills do for an AI agent
A general model can draft interview questions or name the four forces, but left alone it does the two things that discredit discovery: it asks leading, future-tense questions that return comfortable opinions, and it writes a tidy job statement from the team’s imagination with no customer behind it. These skills stop that. Customer-interviews keeps every question in the past and in specifics, and makes the agent synthesise only what people did. Jobs-to-be-done picks the right school for the question and keeps each statement free of solutions. Each hands back findings a product manager can defend, traced to real quotes, not a confident summary.
Where to start
Begin with customer-interviews. Learning how people behave comes first, and the skill writes a behavioural screener, keeps the guide honest with The Mom Test rules, and runs to saturation rather than to an arbitrary count. Then move to jobs-to-be-done to turn those conversations into a model: the four forces behind a switch when the question is why someone bought or churned, or Ulwick’s outcome scores when the question is where the market’s biggest unmet need sits. The two compose: one gathers the evidence, the other frames it. Both feed opportunity-solution-trees, which chooses which problem to pursue.
What the agent needs from you
Give the agent the decision the research will inform and the segment, described tightly enough to screen for one person. For interviews, say how you can reach people and what you can offer for their time. For a jobs model, say whether the question is about a specific purchase or about the market as a whole, because that choice decides the method. Share any real evidence you already have: past interview notes, support tickets or survey responses. Where you have none, the skill says what to gather first rather than inventing a quote or a number.
Frequently asked questions
Which product discovery skill should I start with?
Start with customer-interviews. It plans and runs the conversations that everything else rests on. Once you have real interviews, jobs-to-be-done turns them into a model of why customers buy or churn, or a scored map of unmet needs. Both then feed opportunity-solution-trees, which picks the problem worth solving.
Do these skills run the interviews for me?
No. They structure the work: write the screener and guide, coach the conversation, and synthesise the notes you bring back. A human still talks to the customers and still owns the decision. The skills default to drafting anything that reaches a real person, such as a recruiting email, for you to send.
What is the difference between customer interviews and jobs to be done?
Customer-interviews is about gathering evidence well: honest questions, recruiting, and synthesis. Jobs-to-be-done is about making sense of it: modelling the job, the four forces of a switch, or the outcome scores that show where need is unmet. You use the first to collect, the second to frame.
How many skills are in this set?
All six are published: customer-interviews, jobs-to-be-done, continuous-discovery-habits, assumption-mapping, concept-testing and usability-testing. Download the set as one folder from the button above, or take any skill on its own.
Sources and more product management skills
Each skill lists its own sources with the date it was last checked, from the methods’ originators, Rob Fitzpatrick’s The Mom Test and Teresa Torres on interviewing, and Clayton Christensen, Bob Moesta and Tony Ulwick on jobs to be done, through to the practitioners who have run them at scale. Open either skill to read them in full. For the rest of the toolkit, see the Product Management category.
