Summary
Add a structured library of brainstorming techniques and elicitation methods as data files that /specify's exploration phase can use for systematic idea generation, replacing ad-hoc exploration with procedurally specific facilitation.
Motivation (from Kiln Analysis)
Kiln's brainstorm phase (Step 2) uses two data files:
62 Brainstorming Techniques Across 10 Categories
| Category |
Energy Level |
Example Techniques |
| collaborative |
High |
"Yes And Building" — take operator's last idea, say "yes, and..." then extend |
| creative |
High |
Metaphor generation, cross-domain analogy |
| deep |
Low |
Root cause analysis, constraint mapping |
| structured |
Medium |
SCAMPER, morphological analysis |
| theatrical |
High |
Role-play scenarios, persona adoption |
| wild |
High |
Extreme "what if" scenarios, impossible constraints |
| biomimetic |
Medium |
Nature-inspired solutions |
| quantum |
Medium |
Superposition thinking (explore contradictory options simultaneously) |
| cultural |
Medium |
Cross-cultural perspective shifts |
Technique data structure:
{
"name": "Yes And Building",
"energy": "collaborative",
"prompt": "Take operator's last idea. Say 'yes, and...' then extend it..."
}
50 Elicitation Methods with Output Patterns
Methods include thesis → antithesis → synthesis patterns for debate-style elicitation, structured interview techniques, and domain-specific probes.
How Da Vinci (Facilitator Agent) Uses Them
- Da Vinci's agent definition does NOT list the 62 techniques — it says "facilitate brainstorming using structured techniques, vary energy, track convergence"
- Da Vinci selects techniques heuristically based on what's stuck, varying energy levels
- Techniques are applied procedurally — they drive the conversation, not just display
- Critical rule: Da Vinci never generates ideas — every idea comes FROM the operator
- Fire-and-forget
VISION_UPDATE messages to Visionary agent for each approved section
Brainstorm Phase Communication Pattern
Da Vinci → Visionary (VISION_UPDATE: fire-and-forget, per approved section)
Da Vinci → Visionary (SERIALIZE_AND_SHUTDOWN: triggers file write)
Visionary → Da Vinci (SERIALIZATION_COMPLETE: confirms written)
Da Vinci → team-lead (BRAINSTORM_COMPLETE)
Current State in DevFlow
/specify has an Explore phase that investigates the codebase and gathers context
- Exploration is ad-hoc — agents explore based on their general capabilities
- No structured facilitation techniques for idea generation
- No data-driven approach to varying exploration strategies
Technical Approach
1. Create Technique Data Files
shared/data/
├── elicitation-techniques.json # 30-50 curated techniques
└── elicitation-methods.json # 20-30 structured methods
Focus on techniques that are genuinely useful for software specification (prune metaphorical/gimmicky ones like "Quantum Superposition"):
Curated categories for software:
| Category |
Purpose |
Example |
| decomposition |
Break down complex features |
"Identify the 3 independent sub-problems" |
| constraint |
Surface hidden requirements |
"What must NOT change? What's the blast radius?" |
| stakeholder |
Multiple user perspectives |
"How does the admin vs end-user experience differ?" |
| adversarial |
Edge cases and failures |
"How would a malicious user abuse this?" |
| simplification |
YAGNI enforcement |
"What's the simplest version that delivers value?" |
| analogy |
Pattern recognition |
"What existing feature is this most similar to?" |
| timeline |
Phased delivery |
"What ships in v1 vs v2 vs never?" |
2. Integration with /specify Explore Phase
The Explore agents in /specify select techniques based on feature type:
- Performance issue → decomposition + constraint techniques
- New feature → stakeholder + adversarial + simplification techniques
- Existing system change → constraint + analogy + timeline techniques
- API design → stakeholder + adversarial + analogy techniques
3. Skill Reference
Add technique awareness to the specification skill:
# In the specify command or a new elicitation skill
data-files:
- shared/data/elicitation-techniques.json
- shared/data/elicitation-methods.json
4. No Architectural Change Needed
This is a data file + skill reference addition. The /specify command structure stays the same — agents just have a richer toolkit for structured exploration.
Acceptance Criteria
References
- Kiln repo:
brainstorming-techniques.json (62 techniques, 10 categories), elicitation-methods.json (50 methods)
- Kiln's Da Vinci agent: facilitation-only, never generates ideas, varies energy heuristically
- DevFlow
/specify: plugins/devflow-specify/commands/
- Priority: Medium impact, Low effort — "Quick win"
Summary
Add a structured library of brainstorming techniques and elicitation methods as data files that
/specify's exploration phase can use for systematic idea generation, replacing ad-hoc exploration with procedurally specific facilitation.Motivation (from Kiln Analysis)
Kiln's brainstorm phase (Step 2) uses two data files:
62 Brainstorming Techniques Across 10 Categories
Technique data structure:
{ "name": "Yes And Building", "energy": "collaborative", "prompt": "Take operator's last idea. Say 'yes, and...' then extend it..." }50 Elicitation Methods with Output Patterns
Methods include thesis → antithesis → synthesis patterns for debate-style elicitation, structured interview techniques, and domain-specific probes.
How Da Vinci (Facilitator Agent) Uses Them
VISION_UPDATEmessages to Visionary agent for each approved sectionBrainstorm Phase Communication Pattern
Current State in DevFlow
/specifyhas an Explore phase that investigates the codebase and gathers contextTechnical Approach
1. Create Technique Data Files
Focus on techniques that are genuinely useful for software specification (prune metaphorical/gimmicky ones like "Quantum Superposition"):
Curated categories for software:
2. Integration with
/specifyExplore PhaseThe Explore agents in
/specifyselect techniques based on feature type:3. Skill Reference
Add technique awareness to the specification skill:
4. No Architectural Change Needed
This is a data file + skill reference addition. The
/specifycommand structure stays the same — agents just have a richer toolkit for structured exploration.Acceptance Criteria
shared/data/elicitation-techniques.jsoncreated with curated techniquesshared/data/elicitation-methods.jsoncreated with structured methods/specifyExplore phase can select techniques based on feature typeReferences
brainstorming-techniques.json(62 techniques, 10 categories),elicitation-methods.json(50 methods)/specify:plugins/devflow-specify/commands/