/nw-discuss

name: nw-discuss
description: PO-led product conversation, visible feedback increments and human-readable briefs.
user-invocable: true
argument-hint: '<product question or outcome>'

NW-DISCUSS

Read `nw-human-collaboration` for interactive refinement. Honor an already selected final-review or full-delegation mode; the host manages dialogue and progression. Apply `flow:findings-go-to-the-owner-directly` from `nw-cross-cutting-invariants` when exchanging findings.

Product ownership

Read `nw-role-invocation` before delegating. The exact installed `nw-product-owner` owns semantic product facts and decomposition, using its collaboration, JTBD, BDD and value-slicing knowledge. The host relays human questions and executes requested tools within authorization. Offer proportionate visual exploration for changed experiences; select UX help or a bounded spike only when useful. Capturing product facts does not complete the conversation.

Supply promises, included/excluded scope, decisions, journeys and independently observable slices. Keep corrections with the Product Owner and unresolved prerequisites with their owner; do not silently substitute product meaning.

Explore relevant scenarios

Before settling product authority, read `nw-po-scenario-exploration` and apply its per-value relevance rule with the Product Owner. Do any values need bounded Quint exploration, or is an unexamined deferral hiding the uncertainty?

1. Obtain the PO's product questions and actor observations for relevant values. 2. Probe `quint --version`, including a configured local binary. If unavailable, follow that skill's one-time installation offer and authorization rules. 3. The host obtains a model-capable helper and bounded real `quint run` evidence: literal command, explicit seed/bounds, tool version, model identity and raw trace. Mechanically render events through deterministic domain templates; an LLM paraphrase is not deterministic rendering. The PO owns product meaning, not model implementation. 4. Return rendered scenarios and assumptions to the PO. In automatic mode keep them unreviewed/open, never human-approved; in interactive mode obtain the human's judgment through the exploration skill's collaboration sequence. 5. Have the PO record real scenarios as Scenario promises and assumptions or unresolved judgments as decisions. Tool unavailability or human refusal stays explicit. A missing model, broken renderer or other host-owned prerequisite returns to that owner, not to a silent skip or repeated unchanged PO invocation.

The LLM owns semantic values, their supplied order and dependencies. A simulation is not implementation verification. No solver is required for ordinary scenario discussion when the relevance rule finds no applicable value. Frontend/backend contracts and implementation planning remain design and delivery responsibilities.

Shared human view

Read `nw-doc-as-artifact` and use current briefs or HTML during meaningful refinement when the view helps the human. Inspect the result; if rendering is unavailable, expose the gap and keep the host conversation readable. Rendering is not publication, approval or a progression gate.

Keep intent, actor/value outcomes, confirmed decisions, proposals, assumptions and open questions distinct. Start from known facts; deepen only what affects a decision or the human asks to explore. Human emotion belongs to the human: record it only from the human's account, or explicitly mark it not supplied. Explain unfamiliar terms and preserve original scope. Document language follows `nw-human-collaboration`.

Forward human corrections to the Product Owner for revision. Regenerate the same local view after a meaningful authority change; never edit generated Markdown or HTML. No wave-entry setting, density gate, automatic expansion offer or separate completion ledger is needed.