AgentSkillsCN

generate-creative-ideas

运用SCAMPER、第一性原理、随机词汇,以及AI优化技术,激发创造性思维与方案构思。适用于在头脑风暴、突破创作瓶颈、探索替代方案,或推动创新时使用。

SKILL.md
--- frontmatter
name: generate-creative-ideas
description: Creative problem-solving and ideation using SCAMPER, First Principles, Random Word, and AI-optimized techniques. Use when generating ideas, breaking creative blocks, brainstorming alternatives, or innovating.

Creativity Skill

Workflow

code
1. Ask       → "What problem/challenge?"
2. Context   → Understand current state BEFORE suggesting
3. Diagnose  → Match situation to technique(s)
4. Generate  → Walk through technique step-by-step
5. Evaluate  → Score and filter ideas
6. Develop   → Shape top ideas into actionable concepts
7. Output    → Structured ideas + next actions (Thai if user uses Thai)

Context questions (step 2):

  • "What do you currently do?"
  • "What have you tried?"
  • "What's working? What's not?"

Anti-pattern: Jumping to solutions without understanding context = generic noise


Situation -> Technique Matrix

SituationTechniquesCombination Recipe
Stuck / No ideasRandom Word, Forced Connections, Oblique StrategiesRandom Word -> Forced Connections -> Dot Vote
Need breakthroughFirst Principles, Challenging Assumptions, Combinatorial EngineHMW -> Worst Idea -> SCAMPER -> Brainwriting
Improve existingSCAMPER, Reverse Brainstorming, TRIZ-AISCAMPER -> Reverse Brainstorm -> Impact/Effort
Explore systematicallyMorphological Box, Six Thinking Hats, Constraint InjectionMorphological Box -> Constraint Injection -> Clustering
Reframe problemHow Might We (HMW), Jobs to be DoneHMW -> JTBD -> First Principles
Team ideation6-3-5 Brainwriting, Multi-Persona ParallelMind Map -> Brainwriting -> Affinity -> Multi-Vote
Too many ideasImpact/Effort Matrix, Idea Clustering, NAF ScoringClustering -> NAF Quick Score -> Impact/Effort
AI ideas too similarDivergence Guard, Constraint Injection, Incubation CyclingDivergence Guard -> Domain Shift -> Incubation
Technical/engineeringTRIZ-AI, First PrinciplesTRIZ -> First Principles -> Assumption Mapping
Cross-domain innovationCombinatorial Engine, Analogical ThinkingCombinatorial Engine -> Analogical -> Constraint
Content ideas (blog/video/course)Content Pillars, Audience Pain Points, Gap AnalysisPillars -> Pain Points -> SCAMPER -> Validate
Business/product ideasJTBD, Opportunity Canvas, Lean ValidationJTBD -> HMW -> Morphological -> ICE Score

Technique Quick Reference

Divergent (Generate)

TechniqueOne-liner
SCAMPER7 lenses: Substitute, Combine, Adapt, Modify, Put to other use, Eliminate, Reverse
Random WordRandom noun -> list attributes -> force connections to problem
Reverse Brainstorm"How to make it worse?" -> Invert each idea
First PrinciplesStrip to fundamentals -> Rebuild from scratch
Six Hats6 perspectives: Facts, Feelings, Risks, Benefits, Ideas, Process
HMWReframe as "How Might We [verb] for [user] so that [outcome]?"
Morphological BoxParameters x Variations matrix -> Combine systematically
Analogies"How does [other domain] solve this?"
AssumptionsList assumptions -> Challenge/invert each
Forced ConnectionsCombine 2 unrelated concepts
Jobs to be Done"When [situation], I want [motivation], so I can [outcome]"
Oblique StrategiesRandom creative prompts to break deadlocks

AI-Optimized (Generate + Diversify)

TechniqueOne-liner
Incubation CyclingGenerate -> Pause -> Fresh restart (no prior context) -> Compare
Combinatorial EngineAbstract -> Retrieve 3 domains -> Generalize -> Combine -> Instantiate
Multi-Persona ParallelRun 4+ personas SIMULTANEOUSLY (not sequentially)
Constraint InjectionAdd random constraint -> Force novel solutions
Divergence GuardForce opposite -> Domain shift -> Absurdity injection
TRIZ-AIApply inventive principles (segmentation, nesting, dynamization, etc.)
Tree of ThoughtsExplore multiple reasoning branches simultaneously (74% vs CoT 49%)

Convergent (Evaluate + Refine)

TechniqueOne-liner
NAF Quick ScoreRate Novelty + Attractiveness + Feasibility (1-10 each)
Impact/Effort Matrix2x2: Quick Wins, Big Bets, Fill-ins, Avoid
ICE ScoringImpact x Confidence x Ease (1-10 each)
Idea ClusteringGroup similar -> Name clusters -> Pick best from each
Dot VotingEach person gets 3-5 votes -> Surface favorites
Assumption MappingMap assumptions on Importance x Certainty -> Test riskiest first

Idea Evaluation

After generating ideas, ALWAYS offer evaluation. Default to NAF Quick Score.

NAF Quick Score (Default)

CriterionQuestionScale
NoveltyHow new/surprising is this?1-10
AttractivenessHow well does it solve the problem?1-10
FeasibilityHow realistic to implement?1-10

Interpretation:

  • Total 24-30: Strong candidate -> develop further
  • Total 18-23: Promising -> refine or combine
  • Total < 18: Weak -> park or discard
  • Feasibility 8+: Worth trying (remaining 20% is implementation)
  • High N+A but low F: Reframe feasibility barriers as new problems to solve

When to Use Which Evaluation

SituationMethod
Quick screening (5+ ideas)NAF or Dot Voting
Growth experimentsICE Scoring
Data-driven product decisionsRICE Scoring
Complex multi-criteriaWeighted Scoring Matrix
Visual team alignmentImpact/Effort 2x2

Rules:

  1. NEVER evaluate during divergent phase -- generate first, judge later
  2. Take a break between generating and evaluating (different mindset)
  3. Different idea types need different criteria (incremental vs disruptive)

Details: evaluation.md


Content Creator Mode

When ideating for blog posts, videos, courses, or social media:

Step 1: Define Content Pillars (3-5 themes)

code
Pillar = Core theme that reflects expertise + audience needs
Example: Excel -> [Formulas, Data Viz, Automation, Tips & Tricks, Career]

Step 2: Audience-First Ideation

SourceQuestions
Pain PointsWhat frustrates them most?
QuestionsWhat do they repeatedly ask?
GapsWhat's poorly explained by competitors?
WishesWhat do they wish existed?
MistakesWhat common errors do they make?

Step 3: Generate Ideas (use any technique from matrix)

Apply creativity techniques TO the content pillars:

  • SCAMPER on existing popular content
  • Reverse brainstorm: "How to make the worst tutorial?"
  • Analogies: "How would Netflix teach Excel?"

Step 4: Validate Before Creating

code
[ ] Search demand? (keyword research)
[ ] Real audience pain point? (comments, surveys)
[ ] Can I add unique value? (gap analysis)
[ ] Fits my pillars? (strategy alignment)
[ ] Would I click this? (title/thumbnail test)
[ ] Can be repurposed? (1 piece -> 5+ formats)

Repurposing Chain

code
Blog post -> YouTube video -> Shorts/Reels -> Social posts -> Email -> Course module

Details: content-ideation.md


Idea Development Pipeline

Shape raw ideas into actionable concepts:

code
[Generate] -> [Cluster] -> [Evaluate] -> [Develop] -> [Validate] -> [Execute]

Quick Concept Card (for top ideas)

code
IDEA: [Name]
TAGLINE: [One compelling sentence]
PROBLEM: [What pain it solves]
SOLUTION: [How it works - 2-3 sentences]
TARGET USER: [Who benefits]
KEY INSIGHT: [The "aha" behind this]
EFFORT: [S / M / L]
BIGGEST RISK: [What could go wrong]
QUICKEST TEST: [How to validate cheaply]
NEXT STEP: [One concrete action]

Assumption Mapping (for important ideas)

  1. List assumptions: "What must be true for this to work?"
  2. Categorize: Desirability / Feasibility / Viability
  3. Map on 2x2: Importance (high/low) x Certainty (high/low)
  4. Test high-importance + low-certainty FIRST

Details: idea-pipeline.md


Output Formats

Use these when presenting ideas to the user:

Quick List (5+ ideas)

code
1. **[Idea Name]** -- [One-line description]
2. **[Idea Name]** -- [One-line description]
...

Scored List (after evaluation)

code
| # | Idea | N | A | F | Total |
|---|------|---|---|---|-------|
| 1 | ...  | 8 | 9 | 7 | 24    |

Concept Cards (top 3 ideas)

Use the Quick Concept Card format above for each top idea.

Comparison Matrix (deciding between options)

code
| Criteria     | Idea A | Idea B | Idea C |
|-------------|--------|--------|--------|
| Novelty     | 4/5    | 3/5    | 5/5    |
| Feasibility | 3/5    | 5/5    | 2/5    |
| Impact      | 5/5    | 3/5    | 4/5    |

AI Creativity Guidelines

Key Research Insights (2025-2026)

FindingImplication
Ask AI HOW to think, not WHAT to thinkProcess prompts > product prompts
Human-first ideation preserves diversityUser brainstorms first, THEN AI expands
LLMs match average creativity (52nd percentile)AI is a partner, not a replacement
Homogenization effect (g ~ -0.86)Use Divergence Guard actively
"Creative Scar" -- creativity drops after AI withdrawalDon't outsource all creative thinking
Tree of Thoughts: 74% vs CoT 49%Use ToT for complex creative tasks
Multi-LLM collaboration enhances originalityStack techniques, vary approaches
Only 0.28% of LLM ideas reach top 10% human creativityPush past first outputs aggressively

AI Role Rules

  • Best as: Exploration partner, constraint enforcer, analogy finder, elaborator
  • Avoid: Sole ideation source, final decision maker
  • Timing: After initial human ideation, not before
  • Override rate: Maintain 15-25% human override for optimal outcomes

Anti-Patterns

Don'tDo Instead
"Give me ideas for X""Give me 10 ideas for X that would surprise an expert"
Accept first outputsPush past 2-3 rounds; first ideas are "greatest hits"
Use AI before thinkingBrainstorm independently first, then expand with AI
"Be creative"Use specific technique (SCAMPER, constraints, personas)

Details: methodology.md | prompt-templates.md


References


Related Skills

  • /triz — Systematic innovation methodology (complements brainstorming)
  • /deep-research — Research inspiration and cross-industry solutions
  • /boost-intel — Critical evaluation of generated ideas
  • /design-business-model — Apply creative ideas to business models
  • /problem-solving — Structure the problem before ideating