octocode-brainstorming

Octocode Brainstorming

Evidence-grounded idea exploration. Flow: FRAME → DIVERGE → RESEARCH → CROSS-POLLINATE → STRESS-TEST → SYNTHESIZE → DECIDE.

Modes and lobby rules

• Generate: create 6–10 angles, then validate the best 2–3. Validate: produce 2–4 reframings, then investigate deeply. Map: expand adjacent terms and landscape existing solutions.
• Capture framing before judging. When direction, audience, or research surface is unclear, ask the user to choose before deep research.
• Declare a Surface Plan: mark local, top resources/web, and repository/package/code evidence active or skipped with a reason.
• Treat snippets and summaries as leads; cite exact sources or mark claims weak. Track claim → source → confidence → next query.
• Cross-pollinate across each active surface and run Critical Architect, Visionary Entrepreneur, and Product lenses before verdict.
• Recall potentially useful context first and validate it; capture only durable lessons that survive rebuttal.

Hard gate

Stop, recommend one option, and wait when the idea spans three unrelated spaces, active evidence stays thin after synonym retries, evidence materially conflicts, or delegation would exceed five workers.

Smart routes — load only what the current step needs

• When framing the idea and diverging into angles, build the Surface Plan with references/tools.md; when code/repository/package evidence is active, load references/octocode.md — choose sources deliberately and delegate technical research correctly.
• When generic results cannot prove momentum, crowdedness, publication, or shipped prior art, load references/trend-sources.md — add time-sensitive evidence without domain lock-in.
• When cross-pollinating leads from one active surface into another, stay in references/tools.md and references/web-search-workers.md — carry each finding across surfaces instead of closing a surface early.
• When stress-testing, load references/debate.md — run the three lenses and cross-examination before converging.
• When research is substantial, multi-turn, or delegated, load references/hook-communication.md and run scripts/brainstorm-run.mjs — preserve a resumable claim/source/decision ledger.
• When synthesizing and deciding, load references/output.md and score every prior-art claim with references/confidence.md; if the user approves a durable artifact, load references/brief-template.md — match chat brevity or saved decision depth.
• When methods or source contracts are challenged, load references/grounding.md — make the process falsifiable.
• When improving this skill, prefer octocode-graph-eval; otherwise load references/improve-loop.md — require measurable acceptance.