RizwanZafaris/product-manager-OS / skills/conductor
conductor
VerifiedInterview-driven operation of the six-stage product loop. Use when a user says "start" on a new product, "resume" or "where are we" on an existing one, or asks to be walked through a stage rather than handed a template. Asks one question at a time from the stage's question bank, cross-examines weak answers with a capped challenge grammar, lands every accepted answer in STATE.md and the target template, renders the stage gate on evidence, and never advances past a gate silently.
| File | Kind | Size |
|---|---|---|
| SKILL.graph.yml | config | 211 B |
| SKILL.md | skill md | 10.7 KB |
| questions/README.md | markdown | 4.0 KB |
| questions/build.md | markdown | 5.6 KB |
| questions/define.md | markdown | 12.7 KB |
| questions/deliver.md | markdown | 10.4 KB |
| questions/design.md | markdown | 7.6 KB |
| questions/discover.md | markdown | 8.4 KB |
| questions/operate.md | markdown | 9.0 KB |
questions/define.md
View on GitHub---
layer: skills
stage: DEFINE
gate: 2
feeds: []
method: ""
aliases: ["DEFINE bank"]
---
# DEFINE bank
Stage: DEFINE, feeds Gate 2 (requirements signed off) in [../../../os/STAGE-GATES.md](../../../os/STAGE-GATES.md).
Working handoffs: drafting via [../../../agents/drafting-agent.md](../../../agents/drafting-agent.md); [../../ai-prd/SKILL.md](../../ai-prd/SKILL.md) when a model is inside; [../../../agents/validation-agent.md](../../../agents/validation-agent.md) before the gate.
Questions 1 to 3 run the tree in [../../../os/WHICH-DOCUMENT.md](../../../os/WHICH-DOCUMENT.md) and pick the artifact weight before any template opens. The honest outcome may be "decide and log", in which case the interview ends with one decision-log entry and no document.
Questions 10 to 12 ask for the vision, strategy and roadmap, which the work produces before the definition set; they come last here only because IDs append rather than insert.
Format and ladder: [README.md](README.md).
### DEFINE-1: stakes
Ask: What does being wrong here cost?
Wrong costs: Weight follows stakes; misjudge this and either a quarter of work gets a ticket or a two-day change gets a twelve-section spec.
Evidence class: 5 is acceptable, this is a judgment the user owns; the judgment must still pick one option.
Options: a) an afternoon, which implies decide and log or a ticket. b) a sprint, which implies a ticket with acceptance criteria. c) a quarter, which implies a one-pager or a full PRD. d) a license or a contract, which implies the full stack and a regulatory look.
Cross-examine when: the answer hedges across two options. Move: banned openers, then force one.
Accept when: exactly one option, with one line of reasoning.
Lands in: `execution/decision-log.md` weight entry and STATE.md accepted answers, at every weight.
### DEFINE-2: audience
Ask: Who must read this document and act on it?
Wrong costs: A document written for nobody in particular is read by nobody at all.
Evidence class: 5, a judgment the user owns: named roles, or named people where they exist.
Cross-examine when: the answer is "the team" or "stakeholders". Move: category to name.
Accept when: each reader named with the action their reading triggers.
Lands in: `execution/decision-log.md` weight entry always; at one-pager weight, also `definition/one-pager.md` "Reviewers who must not be surprised" once DEFINE-3 lands the weight (full-PRD and BRD-stack weight route this answer to section 3 "Users and stories" of the PRD and section 4 "Stakeholders" of the BRD instead, filled during those templates' own drafting), and STATE.md accepted answers.
### DEFINE-3: reversibility
Ask: Can this be undone in a day with a flag, or does it set contracts, data models, or public commitments that outlive the team?
Wrong costs: Treating a one-way door as reversible is how a weekend decision becomes a two-year migration.
Evidence class: 2 where the claim is a flag or a rollback path, an artifact should exist; otherwise a reasoned judgment.
Cross-examine when: "we can always roll it back" arrives with no named mechanism. Move: interest to behavior, has that rollback ever been performed?
Accept when: reversible with the mechanism named, or irreversible with what it locks in named. The three answers so far pick the weight; the Conductor states the pick and why before continuing.
Lands in: `execution/decision-log.md` weight entry, plus `definition/one-pager.md` section 3 "Scope" at one-pager weight (full-PRD and BRD-stack weight route this answer into section 4 "Functional scope" of the PRD instead, filled during that template's own drafting), and STATE.md accepted answers.
### DEFINE-4: objective traceability
Ask: Which Gate 1 problem statement does each objective trace to?
Wrong costs: An objective that traces to nothing is scope smuggled past discovery.
Evidence class: 2, the trace itself, objective to problem statement, line by line.
Cross-examine when: an objective traces to a goal, a competitor, or a wish instead of the Gate 1 problem. Move: banned openers, then: cut it, or name the gap out loud in the document.
Accept when: every objective traces, or the untraced ones are cut or explicitly flagged as new scope needing its own discovery.
Lands in: `definition/prd.md` section 2 at full-PRD or BRD-stack weight; `definition/one-pager.md` section 4 at one-pager weight, and STATE.md accepted answers.
### DEFINE-5: how requirements fail
Ask: For each requirement, how does it fail?
Wrong costs: Prose that cannot fail cannot be tested, and its gaps ship silently.
Evidence class: 2, a condition, an expected result, and a measurable threshold per requirement.
Cross-examine when: a criterion contains "works well", "fast", "intuitive", or any word a demo could not falsify. Move: naked numbers, threshold or it is not a criterion.
Accept when: every requirement has a failing condition a tester could stage. Prose that cannot fail is returned, not accepted.
Lands in: `definition/acceptance-criteria.md`, plus `definition/frd.md` at BRD-stack weight, and STATE.md accepted answers.
### DEFINE-6: out of scope
Ask: What is out of scope, and has the sponsor read the list?
Wrong costs: The unwritten exclusion is the launch-week surprise, delivered by the sponsor.
Evidence class: 3, named commitment, the sponsor's written acknowledgment of the list.
Cross-examine when: the list exists but the sponsor has "seen it around". Move: interest to behavior, where is the written yes?
Accept when: the list is written and the sponsor's acknowledgment is dated.
Lands in: `definition/prd.md` section 7 at full-PRD or BRD-stack weight; `definition/one-pager.md` section 5 at one-pager weight, and STATE.md accepted answers.
### DEFINE-7: the assumptions
Ask: Which assumptions is this built on?
Wrong costs: The unsigned assumption is the one that resurfaces in the launch review.
Evidence class: 5, filed properly: each assumption gets a confidence, a validation method, and a validate-by date.
Cross-examine when: the register is empty or every confidence is high. Move: banned openers, an empty register on a new product is itself the strongest claim in the room.
Accept when: each assumption carries all three fields and an owner.
Lands in: `definition/assumptions-register.md` section 1, and STATE.md accepted answers.
### DEFINE-8: overlays
Ask: Does a model produce any user-facing output, and does a financial or data regulator govern any target market?
Wrong costs: A deferred overlay resurfaces later with an eval gap or a regulator's reference number attached.
Evidence class: 5, a stated yes or no per half, each with one line of grounds.
Cross-examine when: either half is "probably not". Move: banned openers, then: which markets, which data, which model, checked against what?
Accept when: both halves answered with grounds. Yes to the model half attaches the AI overlay, eval rows replacing prose criteria via [../../ai-prd/SKILL.md](../../ai-prd/SKILL.md). Yes to the regulator half routes through [../../reg-gap-check/SKILL.md](../../reg-gap-check/SKILL.md) before Gate 2, because the regulated overlay's preconditions freeze here. Yes to both halves activates the regulated overlay itself, which is the rule in [../../../os/STAGE-GATES.md](../../../os/STAGE-GATES.md); a yes on the regulator half alone does not, and STATE.md records what the regulatory owner carries instead.
Lands in: STATE.md position block (overlays active) and `execution/decision-log.md`, and STATE.md accepted answers.
### DEFINE-9: user-facing surface
Ask: Does this change have a user-facing surface? If so, does the UI state inventory list every screen's states with an owner?
Wrong costs: A screen whose states are unlisted ships with gaps in behavior, accessibility, and ownership.
Evidence class: 2, the inventory, or a stated reason the product has no UI.
Cross-examine when: the answer is "we have a UI" but the inventory is missing or incomplete. Move: banned openers, then: name every screen, name its states, name its owner.
Accept when: a filled inventory, or a stated reason the product has no UI.
Lands in: `definition/ui-state-inventory.md` and STATE.md accepted answers.
### DEFINE-10: product vision
Ask: Which customer, which Gate 1 problem and which change does the product vision name, and has the business sponsor approved it?
Wrong costs: With no agreed destination, the strategy and the roadmap each supply their own, and the definition set is built toward whichever of them was written last.
Evidence class: 2, the vision itself, or at one-pager weight the one-pager lines that carry it, with the business sponsor's approval dated on it.
Cross-examine when: the vision names a market or a mood where a customer belongs, or answers a problem Gate 1 never accepted. Move: category to name, then: which Gate 1 sentence does this answer?
Accept when: the vision names who it serves, which Gate 1 problem it answers and the change it aims for, and the business sponsor's approval is dated. At one-pager weight, a few lines of the one-pager that carry all three pass in place of a separate file, under the same dated approval; a vision the one-pager does not carry is missing, not implied.
Lands in: `planning/vision.md`, or at one-pager weight the lines of `definition/one-pager.md` that carry it, and STATE.md accepted answers.
### DEFINE-11: product strategy
Ask: Where does the product strategy play, how does it win, what does it refuse, and have the product owner and the business sponsor approved it?
Wrong costs: A strategy that refuses nothing cannot tell the roadmap what to drop, so every request that arrives becomes a roadmap row.
Evidence class: 2, the strategy itself, or at one-pager weight the one-pager lines that carry it, with both approvals dated on it.
Cross-examine when: a bet refuses nothing, or the strategy never says which vision it serves. Move: banned openers, then: name the option this bet gives up.
Accept when: the strategy names where to play, how to win and what the product will not do, traces to the vision, and the product owner's and the business sponsor's approvals are both dated. At one-pager weight, a few lines of the one-pager that carry all of that pass in place of a separate file, under the same two dated approvals.
Lands in: `planning/product-strategy.md`, or at one-pager weight the lines of `definition/one-pager.md` that carry it, and STATE.md accepted answers.
### DEFINE-12: roadmap
Ask: Which phases does the roadmap order, each with what outcome, success measure and dependencies, and have the product owner and the engineering lead approved it?
Wrong costs: A roadmap of dates with no outcomes becomes a delivery contract nobody signed, and the PRD's objectives trace to deadlines instead of results.
Evidence class: 2, the roadmap itself, or at one-pager weight the one-pager lines that carry it, with both approvals dated on it.
Cross-examine when: a phase carries a date and no outcome, or names a dependency that appears nowhere else. Move: naked numbers, then: what outcome does this phase buy, and who else must deliver for it?
Accept when: the roadmap orders phases, each with an outcome, a success measure and its dependencies, traces to the strategy, and the product owner's and the engineering lead's approvals are both dated. At one-pager weight, a few lines of the one-pager that carry all of that pass in place of a separate file, under the same two dated approvals.
Lands in: `planning/roadmap.md`, or at one-pager weight the lines of `definition/one-pager.md` that carry it, and STATE.md accepted answers.
## Forced pair
On "advance anyway": DEFINE-5, then DEFINE-8. Untestable requirements and an unexamined overlay are the two skips that multiply in cost with every stage they survive.
## Gate 2 rendering
| Gate 2 checklist line | Evidenced by |
|---|---|
| Vision names who it serves, the Gate 1 problem and the change; sponsor approved | DEFINE-10 |
| Strategy names where to play, how to win and what it will not do, traced to the vision; owner and sponsor approved | DEFINE-11 |
| Roadmap phases each carry an outcome, a success measure and dependencies, traced to the strategy; owner and engineering lead approved | DEFINE-12 |
| Every objective traces to Gate 1; every requirement traces to a PRD item | DEFINE-4, DEFINE-5 |
| Every acceptance criterion can fail | DEFINE-5 |
| Every NFR target is a number or names a dated owner | DEFINE-5, checked by the validation agent against `definition/nfr.md` |
| Assumptions register complete per assumption | DEFINE-7 |
| Out of scope written and read by the sponsor | DEFINE-6 |
| Sponsor signed the definition artifact at the chosen weight | Human signature on the one-pager, PRD, or BRD sign-off block; the Conductor reports presence or absence, never supplies it |
| AI overlay: eval sets with thresholds, not prose | DEFINE-8, via the ai-prd handoff |
| Regulated overlay: section 0 answered per market, lint green | DEFINE-8, via the reg-gap-check handoff |
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |