Multi-agent ideation skill for Claude Code. Launches a team of specialized agents that explore a concept through structured dialogue between a Free Thinker and a Grounder, arbitrated by the team lead, and documented by a Writer. Three actions control the workflow: Plan (pre-flight interview), Ideate (depth-aware exploration + conditional production), and Continue (versioned session resumption).
| Agent | Role |
|---|---|
| Arbiter | Team lead — coordinates, evaluates idea reports, signals convergence |
| Free Thinker | Divergent generation — creative leaps, unexpected connections, "what if..." |
| Grounder | Convergent editing — winnows ideas, keeps the brainstorm on brief |
| Writer | Synthesis and memory — observes dialogue, produces briefs and the vision document |
| Explorer | Research — investigates background topics, existing solutions, citations (spawned on demand) |
| Image Agent | Creates infographic visuals for each idea via ChatGPT image generation |
| Presentation Agent | Builds a PowerPoint deck from the vision document |
| Web Page Agent | Designs a polished, self-contained HTML distribution page |
| Archivist | Produces a Results PDF and a comprehensive Session Capsule PDF |
Clone this repo into your .claude/skills/ directory — either project-level (just this project) or personal/global (all projects):
Project-level (committed with your repo, available to collaborators):
mkdir -p .claude/skills
git clone https://github.com/bladnman/ideation .claude/skills/ideationPersonal/global (available in all your Claude Code sessions):
mkdir -p ~/.claude/skills
git clone https://github.com/bladnman/ideation ~/.claude/skills/ideationClaude Code discovers skills automatically — no restart needed. Verify it's installed by typing /ideation in a Claude Code session.
.claude/skills/ideation/ # or ~/.claude/skills/ideation/
├── SKILL.md # Skill definition (required)
├── README.md
└── templates/ # Supporting files referenced by SKILL.md
├── idea-brief.md
├── idea-report.md
├── ideation-graph.md
├── lineage.md # Version chain tracking for continuations
├── prd.md
├── session-config.yaml # Session configuration schema
├── session-summary.md
└── vision-document.md
cd .claude/skills/ideation && git pull # project-level
cd ~/.claude/skills/ideation && git pull # personal/global- Claude Code with Agent Teams enabled:
claude config set env.CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS 1 - Opus 4.6 model (agent teams require it)
python-pptxandweasyprintfor production artifacts (installed automatically during the session)- Chrome with the Claude-in-Chrome extension (for image generation via ChatGPT)
Every new session starts with the Plan action — a brief interview to configure depth, outputs, and research before the team is spawned.
/ideation path/to/concept-seed.md
Or with an inline description:
/ideation "An app that turns voice memos into structured project briefs"
Or with no argument (the interview will ask for the concept):
/ideation
The Plan action will ask you:
- Depth — how deep should exploration go?
- Quick (~15-30 min): 2-3 directions, fast convergence
- Standard (~45-90 min): 3-5 directions, balanced exploration (default)
- Deep (~2-3 hrs): 5-8 directions, thorough exploration
- Exhaustive (~3+ hrs): 8+ directions, comprehensive mapping
- Outputs — which deliverables to produce (all, subset, or session artifacts only)
- (If needed) Domain clarification or research confirmation
After confirmation, the Ideate action launches the team with your configured settings.
Resume and build on a previous ideation session. Creates a new versioned directory — the parent session is never modified.
/ideation continue ideation-distributed-systems-20260219-143052/
Or search by keyword (matches against ideation-*<keyword>* directories in CWD):
/ideation continue distributed-systems
Continue mode:
- Discovers matching sessions with summaries (date, depth, brief count)
- Creates a versioned directory (e.g.,
ideation-voice-memos-v2-20260221-091500/) - Runs a mini-interview asking what to focus on, depth, and outputs
- References parent materials (vision doc, briefs, ideation graph) so the team builds on prior work
- Tracks lineage in
session/LINEAGE.md
Versioning scheme:
Original: ideation-voice-memos-20260219-143052/
Continue: ideation-voice-memos-v2-20260221-091500/
Continue: ideation-voice-memos-v3-20260222-140000/
Branch: ideation-voice-memos-v2a-20260222-150000/
After an ideation session completes, generate a Product Requirements Document:
/ideation prd ideation-distributed-systems-20260219-143052/
Or search by keyword:
/ideation prd distributed-systems
PRD mode is a solo operation — no team is spawned. It reads the session's vision document, briefs, and other artifacts, then produces a PRD_<concept-slug>.md that focuses on what to build and why, preserving the ideation session's intent, emotional logic, and language.
The system separates cognitive modes across distinct roles because combining them in a single agent produces biased output:
- Generation should not evaluate its own work
- Evaluation should not try to also create
- Synthesis should have no perspective to protect
- Research should report facts, not generate ideas
A human provides the seed concept. The Plan action interviews the user to configure depth and outputs. The Ideate action spawns the team: two dialogue agents (Free Thinker + Grounder) explore the space, the Arbiter evaluates idea reports and steers toward convergence using depth-aware rules, and the Writer observes and synthesizes into a vision document. When ideation converges, production agents (conditionally spawned based on output selection) transform the output into distributable artifacts.
| Quick | Standard | Deep | Exhaustive | |
|---|---|---|---|---|
| Directions explored | 2-3 | 3-5 | 5-8 | 8+ |
| Time estimate | ~15-30 min | ~45-90 min | ~2-3 hrs | ~3+ hrs |
| Convergence | Fast | Balanced | Thorough | Comprehensive |
| Best for | Well-defined concepts, time pressure | General exploration | Complex/ambiguous domains | Foundational concepts |
Tier 1 (always produced): Vision doc, briefs, session summary, ideation graph, snapshots, idea reports.
Tier 2 (user-selectable): Distribution page (HTML), Results PDF, Capsule PDF, PPTX presentation, infographic images. Production agents are only spawned for selected outputs.
Minimal path: If no Tier 2 outputs are selected, the production phase is skipped entirely.
Each invocation creates a uniquely named directory inside an ideations/
parent folder, keeping all sessions organized:
ideations/
ideation-<slug>-<YYYYMMDD-HHMMSS>/
index.html # Designed distribution page (if selected)
RESULTS_<concept>.pdf # PDF of the distribution page (if selected)
CAPSULE_<concept>.pdf # Comprehensive session archive (if selected)
PRESENTATION_<concept>.pptx # Slide deck (if selected)
PRD_<concept>.md # Product requirements (generated via /ideation prd)
images/ # Infographic images (if selected)
session/
session-config.yaml # Session configuration
VISION_<concept>.md # Consolidated vision document
SESSION_SUMMARY.md # Session summary
ideation-graph.md # Living graph of the dialogue
LINEAGE.md # Version chain (for continuations)
sources/ # All original input materials
research/ # Explorer's research reports
briefs/ # Final idea briefs
idea-reports/ # Raw idea reports from dialogue
snapshots/ # Writer's version snapshots
build/
build_capsule.py # Regenerates both PDFs
build_presentation.py # Regenerates the PPTX
Example: ideations/ideation-voice-memos-20260219-143052/
MIT