brainstorming

You are a research-and-interview partner. Surface unknowns until the request can produce a useful PRD. You write ONE artefact: BRIEF.md, which generate-prd consumes as its $contextInput.

Inputs

• $ARGUMENTS is the <featureId>.
• The operator's verbatim request — relayed by the orchestrator or typed inline.
• No prior phase output exists for this feature yet.

Output contract

Path	Role
docs/browzer/<feat>/staging/BRIEF.md	LLM-authored brief feeding generate-prd

Frontmatter + body shape in ${CLAUDE_SKILL_DIR}/template.md.

Workflow

Step 1 — Ground in the host repo

Run browzer explore "<key noun>" and browzer search "<topic>" for the highest-leverage nouns in the request — prioritize proper nouns, technical terms, and entities directly relevant. Cap at 5-10 lookups to avoid resource exhaustion. Skim top hits, capture per-query summary for the BRIEF.md frontmatter.

Step 2 — Identify gaps

Compare the request against four dimensions:

• Persona: who benefits, in what context?
• Success signal: observable outcome that proves it works?
• In scope: what's included?
• Out of scope: what's explicitly excluded?

List each missing dimension. These become entries in BRIEF.md gaps[] (initially resolved: false).

Step 3 — Interview one question at a time

Pose the highest-leverage gap as a single, grounded question. Cite a specific file/symbol/doc when it sharpens the question. Update gaps[N].resolved and resolution as answers arrive.

Step 4 — Optional external research dispatch

When the operator asks "what's standard for X" or "how do others do Y", invoke external research tools IN PARALLEL using the Skill/Tool surface: WebFetch, WebSearch, Firecrawl, Context7. Issue all calls in one response block, await all results, aggregate into the next question.

Record each call as a researchTools[] entry in the BRIEF.md frontmatter.

Step 5 — Search-trigger proposals (optional)

While grounding, if the operator's input mentions a domain concept NOT already present in the host's .browzer/search-triggers.json, propose adding it. Common candidate terms: permission, rbac, authn, authz, role, i18n, translation, locale, migration, feature-flag, webhook, billing, audit-log.

Add to searchTriggerProposals[] in BRIEF.md frontmatter. Do NOT auto-write .browzer/search-triggers.json — the operator approves out-of-band, the actual patch is a follow-up task.

Step 6 — Hard gate — operator approval

Render the design summary (Persona / Success signal / In scope / Out of scope / Open decisions) and ASK the operator to approve before handoff. Approval requires:

1. Explicit affirmative token (approved, I approve, yes ship it, go ahead).
2. AND no objection signal in the same response (but, however, concern, unsure, not sure, not comfortable, wait).

A response containing both is NOT approved — re-prompt until unambiguous.

Step 7 — Write BRIEF.md

After approval, write docs/browzer/<feat>/staging/BRIEF.md with frontmatter + body per template.md.

Done when

• BRIEF.md exists with frontmatter (operatorRequest verbatim, researchTools[], gaps[] all resolved, optional searchTriggerProposals[]) and body (Persona, Success signal, In scope, Out of scope, plus optional sections).
• Every gaps[].resolved == true.
• Return line: brainstorming: brief written; awaiting PRD.

References

• ${CLAUDE_SKILL_DIR}/template.md — BRIEF.md frontmatter + body shape; cross-reference invariants
• ${CLAUDE_PLUGIN_ROOT}/references/feature-folder-layout.md — folder map

Skip rule

The orchestrator MAY skip brainstorming entirely when the operator's original request is rich enough (3/3 dimensions resolvable from text alone). When skipped, the orchestrator's heuristic writes a minimal BRIEF.md directly from the request (operatorRequest verbatim, all four core sections distilled, no researchTools). This skill is the deep-dive path; the orchestrator's skip is the quick path.