agentleFS
lukesw55lukesw55/ai-pm-toolkit/.agents/skills/pm-product-sense/SKILL.md

pm-product-sense

Skill1 starsChanged 42 days agoMITView on GitHub

Cross-phase skill for **product sense** — the structured judgement that turns ambiguity into a defensible product decision. Two modes. **BUILD** walks a decision through six steps (clarifying questions → strategy/goal → user types → pain points → solutions → MVP), each step constraining the next. **EVALUATE** acts as a critical interviewer, scoring a one-pager/PRD/pitch on five dimensions (user empathy, structured thinking, product taste, strategic awareness, communication) with 1-5 anchors and a proceed/sharpen/back-to-discovery verdict. Adapted from Exponent's Product Sense Interview guide (tryexponent.com/blog/product-sense-interview). Invoke BUILD for "how would you approach...", an ambiguous product decision, or an interview-style product-sense question; invoke EVALUATE to pressure-test a one-pager or PRD before it advances. Runs as a **mandatory non-blocking shadow evaluation** at stages 4 and 6 — see `skills/WORKFLOW.md`.

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Ask clarifying questions before assuming scope — don't silently pick an interpretation.

    line 60
  2. Distinctive:

    Read the artefact once for what it's actually claiming, not for polish.

    line 69

≈ “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 Product Sense — structured judgement under ambiguity

Transversal, cross-phase skill. BUILD is most useful in Discover/Define when a decision is genuinely open; EVALUATE is most useful right before a One Pager (stage 4) or PRD (stage 6) advances, as a second, more adversarial read before the formal gate.

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 BUILD when:

  • the ask is genuinely ambiguous ("how would you improve X", "what should we build for Y") with no problem statement yet handed to you
  • a product decision needs to be reasoned through from first principles rather than assembled from an existing brief
  • practising or producing an interview-style product-sense answer

Invoke EVALUATE when:

  • a One Pager or PRD is about to advance and needs a critical, structured second read
  • a pitch or recommendation needs pressure-testing before it goes to stakeholders
  • the shadow evaluation at stage 4 or stage 6 is required (see skills/WORKFLOW.md — non-blocking, but running it is not optional)

Skip BUILD when the problem is already validated and framed — use pm-phase-define or pm-phase-develop directly. Skip EVALUATE for artefacts that aren't decision-shaped (status updates, release notes — those go through pm-transversal-comms).

Prime directive

Product sense is judgement made checkable, not a vibe. The six BUILD steps and the five EVALUATE dimensions exist so that "this feels right" turns into a decision someone else can audit — what was assumed, what was ruled out and why, and what evidence would change the call.

Core sub-skills

1. BUILD — six-step decision framework

Clarifying questions → strategy/goal → user types → pain points → solutions → MVP. Each step constrains the next: skipping straight to solutions without naming user types and pain points first is the single most common product-sense failure.

Outputs: a structured decision walk-through, mappable onto a One Pager (pm-phase-define/references/one-pager.md).

Anti-patterns: jumping to a solution before naming the user and the pain point it addresses, picking one user type "out of the gate" instead of enumerating candidates first, no explicit vision/goal statement, features with no MVP cut.

→ Deep-dive: references/six-step-framework.md

2. EVALUATE — five-dimension rubric

User empathy, structured thinking, product taste, strategic awareness, communication — each scored 1-5 against explicit anchors. The lowest-scoring dimension is reported first, not buried; a score of ≤2 on any dimension limits the overall verdict regardless of how the other four score.

Outputs: a scored evaluation with verdict (proceed / sharpen / back-to-discovery), the lowest dimension surfaced first, and what would move the score.

Anti-patterns: averaging away a real weakness, scoring generously to avoid a hard conversation, evaluating polish (writing quality) instead of the underlying judgement.

→ Deep-dive: references/evaluation-rubric.md

Workflow

BUILD:

  1. Ask clarifying questions before assuming scope — don't silently pick an interpretation.
  2. State the strategy/goal this decision serves, explicitly.
  3. Enumerate user types (plural) before picking which to focus on.
  4. Surface pain points per user type, ranked by severity — not a flat list.
  5. Generate solution candidates tied back to a specific pain point; reject weak ones with a stated reason, not silently.
  6. Cut to an MVP against the stated goal — what's in, what's explicitly out, how success is measured.

EVALUATE:

  1. Read the artefact once for what it's actually claiming, not for polish.
  2. Score all five dimensions independently against the anchors in references/evaluation-rubric.md.
  3. Report the lowest score first, with the specific gap that produced it.
  4. Apply the ≤2-limits-verdict rule before compositing a verdict.
  5. State what evidence or revision would move the weakest dimension.

Output contract

## BUILD — [decision]
### Clarifying questions asked
### Strategy / goal
### User types considered (+ which is targeted, and why)
### Pain points (ranked)
### Solution candidates (+ rejected, with reason)
### MVP (in / out / success metric)

## EVALUATE — [artefact]
### Lowest-scoring dimension (reported first)
### Scores — user empathy / structured thinking / product taste / strategic awareness / communication (1-5 + evidence)
### Verdict — proceed / sharpen / back-to-discovery
### What would move the weakest score

Integration

  • Stage 4 (One Pager) and stage 6 (PRD): mandatory non-blocking shadow evaluation. Running EVALUATE is required before advancing; its result does not block the formal gate, which stays authoritative. See skills/WORKFLOW.md for the exact wording and references/evaluation-rubric.md for graduation criteria toward a future formal gate.
  • pm-phase-define/references/one-pager.md: BUILD's six steps map onto the One Pager's five-unlocks quality bar (see references/six-step-framework.md for the mapping).
  • pm-phase-develop/references/prd-writing.md: EVALUATE reads a PRD the same way it reads a One Pager — for judgement, not formatting.
  • Doctrine: EVALUATE is where calibrated disagreement is the whole job — see ../DOCTRINE.md. A rubric that only ever scores high isn't evaluating.

Communication modes follow ../../docs/patterns/COMMUNICATION_MODES.md. Per-skill: Lean (default) is the BUILD output contract or the EVALUATE scorecard as written above; Standard adds the reasoning behind each score/step; Caveman is the verdict/MVP line only.

Success criteria

  • BUILD answers name a user type and a pain point before any solution appears
  • BUILD answers include at least one rejected solution with a stated reason
  • EVALUATE scores lead with the lowest dimension, not the highest
  • EVALUATE never produces a "proceed" verdict alongside an unaddressed ≤2 score
  • shadow evaluations run at stage 4 and stage 6 without blocking advancement

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.