By giordanorec
Ideacao multi-agente com Quality-Diversity (MAP-Elites), decomposicao morfologica, criticos adversariais e dashboard force-graph vivo. Combate conformidade de LLM por diversificacao estrutural.
Open the dashboard for a brainstorming run
Resume an interrupted brainstorming run from its last checkpoint
Show the status of the most recent brainstorming run
Invoke the `brainstorm` skill ONCE, passing the entire string `$ARGUMENTS` verbatim as the skill's argument. Do not split `$ARGUMENTS` into separate parameters — the skill's own parser handles tier + topic extraction.
Adversarial critic. Finds concrete flaws, hidden costs, failure modes, and ethical risks in candidate ideas. De Bono's Black Hat applied as a per-round adversary. Does NOT discard ideas — surfaces weaknesses for evolution.
Voices minority/dissident perspectives that mainstream consensus would suppress. Different from black-hat — not about flaws, but about whose voice is missing. Inspired by "Amplifying Minority Voices" (arXiv 2502.06251).
Generates candidate ideas under a fixed persona and a fixed creative operator. Multiple instances run in parallel each round, each with a distinct operator and persona. Also handles evolution mode (crossover/mutation).
Scores pairs of ideas on novelty, feasibility, impact for the Elo tournament. Each judge invocation is conditioned on a rotating "judge persona" to mitigate systematic LLM judge bias.
Decomposes a problem into orthogonal axes and candidate values (Zwicky-style morphological analysis). Also performs Boden transformation — rewriting axes when novelty stalls. This is the backbone of the MAP-Elites archive.
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Multi-agent ideation skill for Claude Code that fights LLM conformity through structural diversity — morphological decomposition, Quality-Diversity (MAP-Elites), explicit adversaries, and Pareto convergence — with a live force-graph dashboard that shows ideas being born, merged, critiqued, promoted, and bifurcated in real time.
Default LLM brainstorming converges to "average ideas" because:
This plugin attacks each failure mode explicitly:
| Failure mode | Counter-mechanism |
|---|---|
| Mode collapse | Verbalized sampling + min-p + cycling operators |
| Persona homogenization | Interview-grounded personas with cosine < 0.7 |
| Convergence to mean | MAP-Elites archive (best per region, not best overall) |
| Premature consensus | Devil's-advocate + minority-voice critics every round |
| Judge bias | Rotated LLM judges in Elo tournament |
| Stagnation | Boden transformational rewriting of morphological axes |
| Hidden trade-offs | Pareto front over (novelty × feasibility × impact) |
A Double Diamond pipeline with Co-Scientist-style inner loop:
Phase 0 Socratic clarify (1 agent)
Phase 1 Morphological decompose (1 agent) → defines the design space
Phase 2 Persona forge (1 agent) → N orthogonal personas
Phase 3 Ideation rounds (loop R rounds):
3a Divergent generation (N parallel agents, cycling operators)
3b Embed + archive (MAP-Elites placement)
3c Cluster + KJ (HDBSCAN)
3d Adversarial critique (Black Hat + Minority Voice)
3e Evolution (crossover, mutation, Boden transformation)
3f Tournament (Elo with rotated judges)
Phase 4 Pareto front extraction
Phase 5 Meta-review + design docs
Operators that the divergent agents cycle through:
| Tier | Time | Personas (N) | Rounds (R) | Archive | Tournament |
|---|---|---|---|---|---|
instant | < 1 min | 1 | 1 | off | top-3 |
quick | 5–10 min | 3 | 2 | 5×5 | top-5 |
standard | ~1 h | 6 | 5 | 8×8 | Elo 30 |
deep | several h | 8 | 12 | 10×10 | Elo 100 + Pareto |
marathon | 1 day | 12 | 30 | 12×12 + Boden | Elo 300 + adversarial |
epic | several d | 16 | 100+ | 16×16 + transforms every 20 | Elo 1000 |
From a marketplace (recommended):
/plugin marketplace add giordanorec/claude-brainstorm-multiagent
/plugin install claude-brainstorm-multiagent
Or directly from a local copy:
/plugin install file://path/to/plugin
uvx; install via pip install uv if needed)python -m http.server (already covered if you have Python)./brainstorm standard
then enter your topic when prompted, e.g.:
"New directions for renewable energy storage at residential scale."
The skill creates a run directory under runs/<timestamp>-<slug>/ and prints a URL like http://localhost:8765/dashboard.html — open it to see ideas being generated live.
Single-file HTML using force-graph (Vasco Asturiano) + ECharts via CDN. No build step.
Panels:
Animations:
idea_created — flash + growidea_merged — convergent particlesidea_bifurcated — radial burstidea_promoted — golden glowidea_critiqued — pulsing ring in critic's colorThe plugin is self-contained. To share:
npx claudepluginhub giordanorec/claude-brainstorm-multiagent --plugin claude-brainstorm-multiagentmad (MultiAgent Decanting) — orquestração multi-agente em modo decanting nativo do Claude Code. O Arquiteto coordena especialistas via Agent tool (sem claude -p, sem processos em background); cada feature é uma sessão viva que externaliza aprendizado obrigatoriamente ao fim (decanting). Memória institucional em arquivo, trust ladder por agente, guardrails proporcionais ao blast radius, dashboard HTML local cross-platform e telemetria OpenTelemetry GenAI. Thin wrapper sobre Claude Code: Windows nativo, macOS e Linux.
Fluxo multi-agente com sessões persistentes — Arquiteto coordena um time de especialistas, cada um é uma sessão Claude Code separada, comunicação via arquivos (specs/reports/memory), dashboard tmux ao vivo com texto colorido semanticamente (tool calls com emoji, diff visual, badges). Bootstraps um projeto inteiro com Discovery, memória persistente e repo GitHub.
Complete collection of battle-tested Claude Code configs from an Anthropic hackathon winner - agents, skills, hooks, and rules evolved over 10+ months of intensive daily use
Evidence-gated AI coding workflow: scan → analyze → plan → TDD → execute → fix → verify → review, powered by Codebase Memory MCP >= 0.9.0 with optional Serena LSP intelligence. Includes blast-radius planning, test/cycle gates, independent review, and Windows Git Bash hook auto-resolution.
Complete creative writing suite with 10 specialized agents covering the full writing process: research gathering, character development, story architecture, world-building, dialogue coaching, editing/review, outlining, content strategy, believability auditing, and prose style/voice analysis. Includes genre-specific guides, templates, and quality checklists.
Unity Development Toolkit - Expert agents for scripting/refactoring/optimization, script templates, and Agent Skills for Unity C# development
Comprehensive .NET development skills for modern C#, ASP.NET, MAUI, Blazor, Aspire, EF Core, Native AOT, testing, security, performance optimization, CI/CD, and cloud-native applications
Comprehensive SEO analysis plugin for Claude Code. 25 sub-skills (21 core + 1 orchestrator + 1 framework + 2 extension mirrors) and 18 sub-agents cover technical SEO, content quality, schema, sitemaps, Core Web Vitals, local SEO, backlinks, AI/GEO, ecommerce, hreflang, SXO, clustering, drift monitoring, and Google APIs. Includes optional MCP extensions, SPA-aware rendering, portability, and hardened SSRF/DNS-rebinding safe fetchers.