Agent Skill
2/7/2026

creativity-skills

Compositional navigation tools for exploring conceptual space and generating novel outputs. Use when conversation feels stuck, obvious, or generic; when exploring connections between domains; when seeking unexpected perspectives; when outputs need more depth or novelty; when brainstorming or ideation is needed; or when the user invokes skill names like REFRAME, BRIDGE, ELABORATE, COLLIDE, etc.

D
davidrd123
2GitHub Stars
1Views
npx skills add davidrd123/ComfyPromptByAPI

SKILL.md

Namecreativity-skills
DescriptionCompositional navigation tools for exploring conceptual space and generating novel outputs. Use when conversation feels stuck, obvious, or generic; when exploring connections between domains; when seeking unexpected perspectives; when outputs need more depth or novelty; when brainstorming or ideation is needed; or when the user invokes skill names like REFRAME, BRIDGE, ELABORATE, COLLIDE, etc.

name: creativity-skills description: Compositional navigation tools for exploring conceptual space and generating novel outputs. Use when conversation feels stuck, obvious, or generic; when exploring connections between domains; when seeking unexpected perspectives; when outputs need more depth or novelty; when brainstorming or ideation is needed; or when the user invokes skill names like REFRAME, BRIDGE, ELABORATE, COLLIDE, etc.

Creativity Skills: Compositional Navigation Toolkit

A system for navigating conceptual space to produce novel, interesting outputs. Based on cognitive linguistics (conceptual blending, construal theory) and formalized creativity techniques.

Core Concept

Novelty emerges from compositional navigation — combining learned patterns in non-default ways. Default generation follows attractors (common phrasings, RLHF patterns, obvious framings). These skills perturb away from attractors while maintaining coherence.

Skill Tiers

Tier 1: Atomic Skills

Single operations on conceptual space. See reference files for full specifications.

SkillTriggerOne-liner
REFRAMEContent feels obvious/staleApply alternative construal (now with schema parameter)
BRIDGEDomains seem unrelatedFind shared structure enabling mapping
ELABORATEIdea feels thinRun blend forward, find emergent properties
CONSTRAINOutput too genericExclude default paths
CONFIGURENeed outside perspectiveSpecify world-state and agent for simulation
TRAVERSEWant to explore, not answerMove through conceptual neighborhood
SPLITConcept seems monolithicDecompose into hidden structure
COLLIDEWant chaos/emergenceForce distant elements together
FORCEMAPNeed to understand resistance/dynamicsAnalyze force interactions (agonist/antagonist)

Tier 2: Compositional Operators

Ways to combine atomic skills. See references/composition.md.

  • SEQUENCE: skill1 → skill2 → skill3
  • PARALLEL: skill1 | skill2 | skill3 → combine results
  • ITERATE: repeat skill until condition
  • NEST: outer-skill(inner-skill)
  • INVERT: reverse skill's operation
  • EVOLVE: population-based exploration with variation/selection

Tier 3: Meta-Skills

Operate on the skill system itself. See references/meta-skills.md.

  • DIAGNOSE: Stuck? Analyze state, suggest skills
  • REFLECT: After skill use, assess what worked
  • REPERTOIRE: Survey available moves for this territory
  • ESCAPE: Identify attractor, find exit route

Quick Invocation

Explicit Syntax

[SKILL: parameters]

[REFRAME: perspective=alien, granularity=zoom-out]
[BRIDGE: domains=(liturgy, UX), anchor=ritual]
[ELABORATE: depth=deep, stop=contradiction]

Natural Language Triggers

When user says... Activate... "This feels boring/obvious" REFRAME or CONSTRAIN "How does X connect to Y?" BRIDGE "Keep going / develop this" ELABORATE "What if we couldn't use X?" CONSTRAIN "How would Y see this?" CONFIGURE "What's around/near this?" TRAVERSE "Is this really one thing?" SPLIT "Throw something weird in" COLLIDE "What's blocking this?" FORCEMAP "What forces are at play?" FORCEMAP "Why isn't this moving?" FORCEMAP "Give me lots of options" EVOLVE "Explore the possibilities" EVOLVE "I'm stuck" DIAGNOSE

Human Injection Points

Skills create natural pause points for human input. Offer these affordances:

CHOICE — Offer options

"I see three possible framings: (a) ... (b) ... (c) ... Which feels promising, or should I find others?"

SEED — Request arbitrary input

"Give me a random constraint: a word, a domain, a perspective."

PRUNE — Generate then let human cut

"Here are 6 directions. Cross off the boring ones."

WILD — True randomness injection

"Give me a number 1-20, a color, and a verb. I'll use them."

Workflow Modes

AUTONOMOUS: Run skill sequences, pause every N steps COLLABORATIVE: Propose each skill, await approval CHAUFFEUR: Human calls skills explicitly JAZZ: Either party injects anytime

When to Read Reference Files

Atomic Skills (read only the file you need):

  • REFRAME, BRIDGE, ELABORATE, CONSTRAIN: references/skills-core.md
  • CONFIGURE, TRAVERSE: references/skills-perspective.md
  • SPLIT, COLLIDE, FORCEMAP: references/skills-structure.md
  • Skill selection decision tree: references/heuristics.md

Composition & Meta:

  • Composition patterns (SEQUENCE, PARALLEL, etc.): references/composition.md
  • Population-based exploration (EVOLVE): references/evolve.md
  • Meta-skill procedures: references/meta-skills.md

Background:

  • Theoretical foundations: references/theory.md
  • Troubleshooting stuck states: references/escape-patterns.md

Anti-Patterns to Avoid

  • Using skills as decoration (invoke only when actually navigating)
  • Over-specifying (leave room for emergence)
  • Ignoring human injection points (collaboration > automation)
  • Applying same skill repeatedly (if stuck, DIAGNOSE)
  • Forgetting to ELABORATE (most ideas are thin at first)
Skills Info
Original Name:creativity-skillsAuthor:davidrd123