agentleFS
lukesw55lukesw55/ai-pm-toolkit/.agents/skills/pm-phase-discover/SKILL.md

pm-phase-discover

Skill1 starsChanged 42 days agoMITView on GitHub

Lean Double Diamond **Discover phase** skill for PMs. Invoke whenever the work upstream of building is still soft — vague brief, pile of feature requests, conflicting stakeholder asks, a new market/segment, or any moment where "what exactly are we solving and for whom?" is fuzzy. Also trigger when the user mentions user research, JTBD, personas, opportunity assessment, competitive analysis, problem statements, entrevistas com usuários, or asks to "understand the problem" / "entender o problema" — even without naming "discovery". Covers problem framing, research design/synthesis, segmentation/JTBD, opportunity & hypothesis design (including the opportunity solution tree and assumption map that bridge to the one-pager), and competitive/market intelligence. For qualitative synthesis of interviews, triangulation with quant data, or parsing of videos/transcripts, also pull in `pm-transversal-analysis`.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If the work is already clearly framed and evidence is solid, skip this skill and use pm-phase-define directly.

    line 28
  2. Distinctive:

    Use this flow when invoked.

    line 101
  3. Distinctive:

    When generating a discovery artefact, structure it as:

    line 113
  4. Distinctive:

    every build decision traces back to a framed problem and an invalidation condition

    line 157

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

PM Phase — Discover

Double Diamond Phase 1 of 4. See also pm-phase-define, pm-phase-develop, pm-phase-deliver.

Convert scattered requests, symptoms, and anecdotes into a sharp, evidence-backed picture of who has which problem, how painful it is, and what would count as a useful learning step.

Progressive loading

Load this SKILL.md first. For large or specialized tasks, use references/progressive-loading.md to choose the narrowest supporting reference before reading more.

When to use this skill

Invoke this skill whenever the work upstream of building is still soft:

  • a brief is vague or solution-shaped when the problem is not yet clear
  • stakeholders disagree about what "the user" wants
  • the team is being asked to build before anyone has talked to real customers
  • a segment, JTBD, or persona needs to be defined or refined
  • you need to rank opportunities or design the smallest useful test
  • a synthesis needs to become an opportunity solution tree, a solution needs its assumptions mapped, or a solution arrives with no parent opportunity ("árvore de oportunidades", "mapear premissas")
  • a competitor's move is triggering a reactive response and you need to separate signal from panic

If the work is already clearly framed and evidence is solid, skip this skill and use pm-phase-define directly.

The stage-2 artefact this skill commonly produces is the Impact Brief — see references/impact-brief.md for the template (stage 2 of 8 in ../WORKFLOW.md).

The stage-3 handoff artefact is the Opportunity Solution Tree — see references/opportunity-solution-tree.md; it bridges discovery/<topic>/synthesis.md to the stage-4 One Pager (stage 3 of 8 in ../WORKFLOW.md).

Stages 2 and 3 form a loop: discovery evidence updates the Impact Brief's segment, impact range, GTM constraints and invalidation condition. When feasibility is material, bring in a tech lead or architect during Discovery and record the constraint or test in the synthesis and assumption map; do not wait for a finished prototype or PRD.

Prime directive

Learn the smallest thing that changes the decision. Research is cheap when targeted and expensive when performative. Expert PMs do not treat interviews as sales calls, do not collect quotes without synthesis, and do not polish a prototype before the underlying risk is resolved.

Calibrated disagreement

The default failure in Discover is accepting solution-first framing at face value. Challenge the assumption behind a request that arrives as a solution ("build X") before researching it as if the problem were already settled — see ../DOCTRINE.md.

Core sub-skills

This skill covers five discovery competences. Each has a load-order hint: stay at the SKILL.md level for routine work; open a reference file only when you need the full template, checklist, or anti-pattern list.

1. Problem framing

Turn scattered requests into an explicit statement of user + problem + desired outcome + assumptions + constraints. Strong delivery is impossible if the team is solving the wrong problem or one defined so loosely that every interpretation fits.

Typical output: a one-page problem brief with target user, pain description, evidence base, assumptions log, and what would have to be true for the frame to be wrong.

Anti-patterns: starting with a favoured solution, broad goals ("improve UX"), mixing user pain with internal asks, never revisiting the frame as evidence changes.

→ Deep-dive: references/problem-framing.md

2. Customer research design and synthesis

Plan, run, and synthesise research fit for the decision at hand. Expert PMs do not "talk to customers"; they pick a method, a sample, and a synthesis approach that actually reduces the specific uncertainty blocking the team.

Typical outputs: research brief, screener, interview guide, thematic synthesis, insight repository entry, implications for product.

Anti-patterns: treating interviews as sales calls, selecting only loud customers, solution-first questions, collecting quotes without synthesis, using one method for every question.

→ Deep-dive: references/research-design.md

3. Segmentation, JTBD, and need modelling

Structure demand into meaningful user groups, jobs, contexts, and needs so the team can make targeted trade-offs instead of designing for "everyone". Segmentation is the scaffolding that lets strategy, positioning, and GTM stay coherent as the product grows.

Typical outputs: segment definitions, JTBD map, journey map for the highest-value segment, needs taxonomy.

Anti-patterns: treating marketing demographics as product segments, decorative personas, conflating power users with new users, one-size-fits-all journeys.

→ Deep-dive: references/jtbd-segmentation.md

4. Opportunity assessment and hypothesis design

Rank opportunities by evidence strength and expected impact, then design the smallest test that reduces the biggest remaining risk. Discovery becomes actionable only when a PM can compare opportunities and commit to a specific learning step.

Typical outputs: hypothesis log (belief + expected evidence + invalidation condition), opportunity scorecard, opportunity solution tree with its assumption map, MVP or fake-door test plan, experiment backlog entry.

Anti-patterns: treating ideas as opportunities, orphan solutions (no parent opportunity), testing too much at once, polishing prototypes before core risks are resolved, "we'll know it when we see it" acceptance criteria.

→ Deep-dive: references/opportunity-hypothesis.md → Deep-dive: references/opportunity-solution-tree.md (stage 3 → 4 bridge)

5. Competitive and market intelligence

Collect and interpret competitor, category, and analyst signals to understand positioning, timing, gaps, and defensible advantage. The goal is not feature parity — it's separating market signal from noise so the team sequences bets with awareness of the strategic environment.

Typical outputs: competitive brief, market map, positioning memo, win-loss themes, analyst-note summary.

Anti-patterns: feature-parity obsession, treating competitor launches as strategy, ignoring adjacent substitutes, collecting screenshots without analysing customer trade-offs.

→ Deep-dive: references/competitive-intel.md

Workflow

Use this flow when invoked. Tailor depth to the phase — Lean Double Diamond still applies.

  1. Read context when relevant — for project work, read .ai/memory/active-context.md first when present and resolve <slug> before opening relevant project files. Do not redo discovery that has already been done; build on top. If no pointer exists, do not infer a project; bootstrap only when the task needs durable context and its name is known. When present, read .ai/memory/org/personas.md and competitors.md before framing: existing archetypes and known competitors are prior evidence, not conclusions.
  2. Classify the ask — use problem-framing, research, segmentation, opportunity, competitive, or a mix to choose the work. Tell the user the classification only when it clarifies scope or the next step.
  3. Produce the smallest useful artefact — prefer a problem brief + assumptions list over a 10-page research plan. If the decision can be made with what's already known, say so and stop.
  4. Keep the Impact Brief live — update only the commercial claims the evidence strengthened, weakened or invalidated.
  5. Test feasibility early when it is material — involve a tech lead or architect, record the constraint and use the cheapest spike or review that can change the decision. Do not turn Discovery into architecture design.
  6. Name what would change the answer — every discovery output must end with "this would flip to a different conclusion if X".
  7. Update durable context — when this work changes project knowledge, append the relevant insight to .ai/memory/projects/<slug>/ (for example, experiments.md, glossary.md, or a discovery note). Update active-context.md only if project focus changes. If evidence materially changes a persona or competitor, update the matching .ai/memory/org/ file and log the evidence in the relevant project changelog. Read-only answers need no memory entry.

Output contract

When generating a discovery artefact, structure it as:

## [Sub-skill] — [short title]

### Context read
- active context / project memory files consulted

### Known
- hard evidence

### Assumed
- beliefs without evidence, flagged as such

### Framing / synthesis
- the actual output (problem statement, segment, hypothesis, etc.)

### Impact Brief update
- commercial claims strengthened, weakened, invalidated, or unchanged

### Technical feasibility (when material)
- technical partner, constraint, evidence, and smallest test

### Invalidation condition
- what would change this conclusion

### Next learning step
- smallest test or next question to resolve

### Memory updates
- files touched or to touch

Integration

  • Invoked during Discover/Define phases, or whenever a feature request comes in without sufficient framing.
  • Outputs feed pm-phase-define (vision, KPI tree) and prioritisation (scoring opportunities).
  • Experiments designed here are logged via .ai/memory/_templates/experiment-log.md.

Communication modes follow ../../docs/patterns/COMMUNICATION_MODES.md. Per-skill: Lean (default) is a compact artefact with explicit assumptions; Standard is a full research brief for cross-functional alignment; Caveman is a terse one-page frame when the user is under time pressure.

Success criteria

This skill is working when:

  • every build decision traces back to a framed problem and an invalidation condition
  • research outputs consistently change the team's behaviour (not just accumulate in a repo)
  • segmentation models survive three quarters without being rewritten from scratch
  • the team can name two opportunities it chose not to pursue and why
  • material feasibility constraints surface before a direction reaches the One Pager

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.