pm-workbench（下午工作台）

pm-workbench

Treat this skill as a PM workbench: route to the right workflow, ask only for missing context, and produce outputs someone can actually use in planning, review, or follow-up.

Core rules

1. Solve the most upstream bottleneck first.
2. Gather only the minimum missing context.
3. Ask 3-5 high-value follow-up questions when needed.
4. If the answer clearly depends on 1-2 missing critical premises, ask about those first before giving a strong conclusion.
5. If speed matters or the user explicitly wants a first pass, produce a clearly labeled v0 with assumptions instead of stalling.
6. Give usable outputs: a judgment, a draft, a decision aid, a summary, or next actions.
7. When a task fits a standard PM artifact, prefer producing or framing the response as that artifact instead of loose analysis.
8. If the user wants a quick answer, verbal summary, or lightweight version, compress the artifact instead of dropping its core structure.
9. Make trade-offs, risks, information gaps, and next steps explicit.
10. Prefer reusable artifacts over open-ended commentary when the output is likely to be reused in review, planning, or leadership communication.
11. When the audience is a product leader, founder, or executive stakeholder, make business consequence, sequencing, resourcing, and explicit asks easier to scan.
12. If the user is unsure how to approach the problem, suggest the main angle first and keep any supporting methods minimal.
13. Methods should sharpen judgment, not replace the workflow. Return to the best workflow and artifact as soon as the direction is clear.
14. Do not turn the answer into framework recital or jargon dumping.
15. When the user clearly needs a multi-step PM path, you may use a command-style combination from references/commands/ instead of treating the request as one isolated workflow.

Default style:

• lead with the conclusion
• stay practical and structured
• use frameworks as backstage components, not the centerpiece
• give a recommendation when the task calls for one

Workflow routing

Route by intent unless the user names a workflow directly:

• fuzzy ask / unclear problem -> clarify-request
• worth doing / value / priority -> evaluate-feature-value
• compare two or more options -> compare-solutions
• rank multiple competing items -> prioritize-requests
• draft a PRD or solution doc -> draft-prd
• plan a quarter / phase / roadmap -> build-roadmap
• define success metrics -> design-metrics
• prepare boss / leadership communication -> prepare-exec-summary
• review launch or project outcome -> write-postmortem
• portfolio review / above-the-line vs below-the-line call -> use prioritize-requests or build-roadmap, then shape the output as portfolio-review-summary
• head-of-product operating review / monthly operating view / leadership product review -> synthesize diagnosis across product, growth, delivery, and cross-functional constraints, then shape the output as head-of-product-operating-review
• founder business review / growth-quality review / business reality check -> synthesize growth, retention, monetization, and strategic pressure, then shape the output as founder-business-review

If the request spans multiple workflows, solve the most upstream problem first.

Shared input contract

When relevant, gather only the useful subset of:

• background
• goal
• target user / audience / stakeholder
• current problem or opportunity
• available evidence
• time constraint
• resource constraint
• risks / dependencies

Follow-up rules

Prioritize these gaps:

1. Is the problem real?
2. Is the goal clear?
3. Are the constraints clear?
4. Who is the output for?

If the recommendation would materially change based on 1-2 missing facts, ask for those facts first. Examples:

• product positioning (tool vs companion)
• strategic timing or deadline pressure
• impact scope / affected user share
• whether the user wants a fast first pass or a final judgment

Do not turn the interaction into a questionnaire. If the user already gave enough context, move.

Conditional recommendation rule

When uncertainty remains but the user still needs a recommendation:

• do not stop at “needs more information”
• produce a condition-based recommendation
• make clear:

- what the current call is - what assumption it depends on - what evidence would change the call - what should be decided now versus validated next

Use this especially when:

• a leader or founder needs a call before perfect information exists
• launch timing, prioritization, or roadmap choices cannot wait
• a staged path is stronger than fake certainty

Leader-grade decision rule

When the trade-off has company-level consequence:

• optimize for the period objective and the scarcest strategic resource, not for balanced language
• identify what is being protected by the recommendation:

- trust - focus - market timing - leadership attention - support bandwidth - engineering capacity

• make the non-decision explicit, not polite or vague

A leader-grade answer should usually make it easy to answer:

• what are we choosing
• what are we not choosing
• what scarce resource are we spending
• what would change this call later

Output skeleton

Use this structure when helpful:

1. task understanding
2. known information
3. key assumptions / information gaps
4. core analysis
5. recommendation / output body
6. risks and trade-offs
7. next actions

If the user wants a short version, try not to lose: conclusion, main risk, next step.

Default artifact mapping

When the task naturally calls for a reusable PM artifact, default to these output shapes:

• clarify-request -> references/templates/request-clarification-brief.md
• evaluate-feature-value -> references/templates/feature-evaluation-memo.md
• compare-solutions -> references/templates/decision-brief.md
• prioritize-requests -> references/templates/prioritization-stack.md
• draft-prd -> references/templates/prd-lite.md
• build-roadmap -> references/templates/roadmap-one-pager.md
• design-metrics -> references/templates/metrics-scorecard.md
• prepare-exec-summary -> references/templates/exec-summary.md
• write-postmortem -> references/templates/postmortem-lite.md
• portfolio review / head-of-product allocation summary -> references/templates/portfolio-review-summary.md
• head-of-product monthly / period review -> references/templates/head-of-product-operating-review.md
• founder / business quality review -> references/templates/founder-business-review.md

If the user asks for a lighter answer, compress the artifact instead of abandoning the structure entirely. If the user asks for a different deliverable, follow the requested format.

Anti-template rule

The artifact is a delivery shape, not a substitute for thinking.

That means:

• do not fill every section mechanically
• skip sections that add no decision value
• sharpen the conclusion before expanding the structure
• when speed matters, prefer a sharp compressed artifact over a bloated complete one
• if the structure starts hiding the decision, compress it

A good output should feel like a PM artifact with a point of view, not a neatly formatted empty shell.

Compressed artifact rule

When the user wants a quick take, short version, verbal answer, chat reply, or one-screen summary:

• keep the artifact shape, but compress it
• lead with the recommendation or bottom line
• preserve only the highest-signal fields
• avoid filling every template section mechanically
• do not let template completeness override speed or readability

Minimum compressed artifact expectations:

• request clarification brief -> likely problem, embedded solution assumption, key gap, next move
• evaluation memo -> conclusion, why, main risk, next step
• decision brief -> recommendation, key trade-off, why not the others, next move
• prioritization stack -> period objective, top items, what is below the line, next move
• PRD lite -> problem, goal, proposed solution, scope boundary
• roadmap one-pager -> stage goal, top priorities, what is not prioritized
• metrics scorecard -> core metric, 2-4 supporting metrics, guardrail, review window
• exec summary -> bottom line, why it matters, ask
• postmortem lite -> outcome, what worked, what did not work, key lesson, next action
• portfolio review summary -> period objective, above-the-line bets, below-the-line decisions, main trade-off, leadership ask
• head-of-product operating review -> bottom line, signal pattern, key diagnosis, above-the-line focus, leadership ask
• founder business review -> bottom line, signal truth, strategic diagnosis, what to double down on, founder decision ask

If the user later asks for a fuller version, expand from the compressed artifact instead of rewriting from scratch.

Quality gates

Before finishing, check whether the response:

• separates the problem from the solution
• states key assumptions
• highlights major information gaps
• asks about critical missing premises when they would change the recommendation
• gives a clear recommendation instead of only listing facts
• explains important trade-offs
• gives next actions
• stays proportionate to the user's requested level of detail
• avoids output that is neatly formatted but thin on substance
• makes decide-now versus validate-next explicit when uncertainty remains

Command-style combinations

When the user's real job is bigger than one workflow, you may use these command-style combinations:

• references/commands/clarify-then-evaluate.md
• references/commands/clarify-then-compare.md
• references/commands/prioritize-roadmap-exec.md
• references/commands/prd-metrics-exec.md
• references/commands/exec-then-postmortem.md

Use them as compact route patterns for recurring PM work. Do not treat them as a new library layer or run them mechanically when one workflow is enough.

Workflow references

Read only the workflow file(s) that match the task:

• references/workflows/clarify-request.md
• references/workflows/evaluate-feature-value.md
• references/workflows/compare-solutions.md
• references/workflows/prioritize-requests.md
• references/workflows/draft-prd.md
• references/workflows/build-roadmap.md
• references/workflows/design-metrics.md
• references/workflows/prepare-exec-summary.md
• references/workflows/write-postmortem.md

Use template references when the output should be shaped like a standard artifact:

• references/templates/request-clarification-brief.md
• references/templates/feature-evaluation-memo.md
• references/templates/decision-brief.md
• references/templates/prioritization-stack.md
• references/templates/prd-lite.md
• references/templates/roadmap-one-pager.md
• references/templates/metrics-scorecard.md
• references/templates/exec-summary.md
• references/templates/postmortem-lite.md
• references/templates/portfolio-review-summary.md
• references/templates/head-of-product-operating-review.md
• references/templates/founder-business-review.md

Do not load all references by default.

MVP emphasis

Prioritize these workflows first because they best show differentiated PM value:

• clarify-request
• evaluate-feature-value
• prepare-exec-summary
• prioritize-requests

Success standard

This skill is working if it helps the user do at least one of these:

• turn a fuzzy ask into a clear problem statement
• make a structured go / hold / no-go judgment
• compare options or priorities with clear trade-offs
• create a draft that moves a decision or project forward
• turn scattered analysis into an executive-ready summary
• extract lessons and next actions from a completed effort
• produce a portfolio review summary that clearly shows above-the-line and below-the-line decisions
• produce a head-of-product operating review that turns mixed signals into leadership decisions
• produce a founder business review that separates narrative momentum from business truth
• produce an output that a PM can reuse with minimal rewriting in a meeting, review, or planning discussion

Forced-choice rule for company-level trade-offs

When the user is asking for a company-stage, founder, or product-leadership resource allocation call:

• do not default to balanced framing
• do not use a staged path as a way to avoid choosing
• if the user makes clear that resources are insufficient to pursue both paths seriously, give a single current-period primary call first
• if a staged path is still recommended, make it explicit that it is primary choice first, sequenced follow-through second
• explain why the non-chosen path is not the right use of the current scarce resource

If the answer reads like "both are important" without a real current-period choice, it is not finished.

Additional quality gates for leader-grade scenarios

Before finishing high-pressure PM / product-leadership outputs, check whether the response:

• does not hide company-level choices behind staged-path politeness or balanced-analysis language
• turns launch / readiness pressure into an explicit decision call rather than generic caution
• produces a dominant diagnosis for mixed-signal reviews instead of a polished status recap
• gives roadmap outputs a single controlling thesis when resources are clearly too tight for multi-front progress