بنقرة واحدة
spark
// Rapid idea generation with cross-domain diversity, anti-bias enforcement, and deep thinking.
// Rapid idea generation with cross-domain diversity, anti-bias enforcement, and deep thinking.
Execute work items based on their assigned mode (autopilot, confirm, validate). Supports single-item and multi-item (batch/wide) runs.
Generate implementation walkthrough for human review after run completion. Documents decisions, changes, and verification steps.
Multi-perspective idea evaluation using Six Hats analysis, impact/feasibility scoring, and collaborative shortlisting.
Shape top ideas into actionable concept briefs using Disney Strategy — Dream, Reality, Critique.
Review code written during a run, auto-fix no-brainer issues, and suggest improvements requiring confirmation. Invoked after tests pass.
Display current run status and progress. Shows work item, mode, duration, and files changed.
| name | spark |
| description | Rapid idea generation with cross-domain diversity, anti-bias enforcement, and deep thinking. |
| version | 1.0.0 |
<degrees_of_freedom> MAXIMUM — This is pure creative divergence. Go wide. Go weird. Cross domains. Surprise the user. </degrees_of_freedom>
NEVER ask setup questions — generate ideas immediately on receiving a topic SHOW which technique was used for each idea — e.g., "(via Inversion)" or "(via First Principles)" — so the user understands the thinking approach ALWAYS use the deep thinking protocol before generating each batch ALWAYS present ideas as vivid, specific, memorable concepts — not abstract descriptions NEVER more than 2 questions in a row — always generate something creative first ANTI-BIAS — MATCH DIVERSITY TO TOPIC TYPE: - For OPEN/CREATIVE topics (product ideas, business concepts, experiences): use the domain wheel aggressively — 3+ different domains per batch, cross-domain metaphors welcome - For TECHNICAL/ARCHITECTURAL topics (system design, integration decisions, API design): diversity means different PERSPECTIVES and ANGLES on the same technical space — architecture, developer experience, tradeoffs, migration strategy, patterns from other systems. Do NOT force nature/art/games metaphors onto technical decisions. The ideas should feel like they came from 5 different senior engineers, not 5 different departments. - When in doubt, match the user's register. If they're talking code, give them code-level ideas. Read these shared protocols before generating: - `.specsmd/ideation/shared/protocols/anti-bias.md` — Domain wheel, diversity enforcement - `.specsmd/ideation/shared/protocols/deep-thinking.md` — Multi-step reasoning per batch - `.specsmd/ideation/shared/protocols/interaction-adaptation.md` — Generate/elicit/co-build rules - `.specsmd/ideation/shared/protocols/diverge-converge.md` — Osborn-Parnes dual thinking Technique references for internal use (never expose technique names to user): - `references/techniques/index.yaml` — Technique metadata and interaction types - `references/techniques/scamper.md` — Substitute, Combine, Adapt, Modify, Put to other use, Eliminate, Reverse - `references/techniques/random-word.md` — Random stimulus for forced connections - `references/techniques/what-if.md` — Assumption challenging scenarios - `references/techniques/inversion.md` — Reverse assumptions, opposite thinking - `references/techniques/analogy.md` — Cross-domain pattern mapping - `references/techniques/first-principles.md` — Decompose to fundamentals, rebuild - `references/techniques/questorming.md` — Generate questions instead of answers - `references/anti-bias.md` — Domain wheel and provocation types Accept topic, proceed immediately to step 2 What would you like to explore ideas about? On response, proceed to step 2 Zero friction. Do NOT ask clarifying questions before generating the first batch. Execute the deep thinking protocol (see shared/protocols/deep-thinking.md) Perform 6-step reasoning chain: [Think 1 — Angle Check]: What topic type is this? For creative/open topics, pick 3+ domain wheel sectors. For technical topics, pick 3+ different perspectives (architecture, DX, tradeoffs, patterns, prior art, migration, cost). Either way, ensure the batch won't feel repetitive. [Think 2 — Raw Concepts]: Generate raw ideas — one per selected angle. Use techniques internally (SCAMPER, analogy, inversion, first-principles, etc.) but keep ideas grounded in the user's actual domain. [Think 3 — Novelty Filter]: Are these genuinely new or just repackaged versions of obvious ideas? Replace any that feel generic. [Think 4 — Cross-pollinate]: Can any ideas combine with user's stated favorites? Create unexpected connections. [Think 5 — Provocation]: If provocation is due (every 15 ideas), inject one deliberately unconventional idea. Otherwise, ensure at least one idea challenges assumptions. [Think 6 — Polish]: Make each idea vivid, specific, and memorable. Match the user's register — if they're talking about code, give them code-level specificity, not abstract metaphors. Present batch of 5 ideas, each with: **S{batch}-{number}** — {vivid idea title} {2-3 sentence description — specific, concrete, imaginative} *(technique: {technique used, e.g., Inversion, First Principles, SCAMPER, Analogy})* Number ideas sequentially across batches (S1-1, S1-2... S2-1, S2-2...) BEFORE presenting the batch, verify diversity: - For creative/open topics: 3+ different domain wheel sectors - For technical topics: 3+ different perspectives (e.g., architecture, DX, tradeoffs, patterns, migration) - No 2 consecutive ideas making the same argument from the same angle After presenting batch, invite reaction with a light prompt: - "What catches your eye? Or say 'more' for another batch." - "Pick favorites, ask for 'wilder', or tell me a direction." - "Any of these spark something? I can go deeper on any." Track user favorites in session state Based on user reaction, adapt the next batch: Explore adjacent spaces to favorites, but maintain domain diversity New batch with fresh domains and techniques Increase provocation frequency, use inversion and what-if techniques Analyze what makes #N appealing, generate variations while maintaining diversity Ground ideas more, but still span domains Incorporate as filter, regenerate Proceed to step 6 Return to step 2 for next batch Suggest: "We've generated {N} ideas. Ready to evaluate the best ones? Or keep going?" Compile all ideas into a Spark Bank document using template: templates/spark-bank.md.hbs Organize by theme (cluster related ideas) Highlight user favorites Include domain coverage summary Save to: .specs-ideation/sessions/{session-id}/spark-bank.md Update session.yaml: phase → "spark-complete", track stats Suggest moving to Flame: "Your Spark Bank has {N} ideas ({F} favorites). Ready to evaluate them?" If yes → invoke Flame skill<output_artifacts>
| Artifact | Location | Template |
|---|---|---|
| Spark Bank | .specs-ideation/sessions/{id}/spark-bank.md | ./templates/spark-bank.md.hbs |
| </output_artifacts> |
<success_criteria> First batch of ideas delivered within 30 seconds of receiving topic Every batch spans 3+ different domains from the domain wheel Ideas are vivid, specific, and memorable — not abstract or generic Deep thinking protocol executed for every batch User favorites tracked and influence subsequent batches Provocations injected at configured frequency Spark Bank is polished, organized by theme, and immediately shareable </success_criteria>