菜单
Multi-turn brainstorming session with RAG, domain profiles (code/tech/btp/cyber/finance), Claude Opus or local Qwen3.5 backend.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | brainstorm |
| description | Multi-turn brainstorming session with RAG, domain profiles (code/tech/btp/cyber/finance), Claude Opus or local Qwen3.5 backend. |
| trigger_patterns | ["brainstorm","session de réflexion","explorer idées","options pour","stratégie"] |
| tools_required | ["exec","chromadb_rag","search_router","llm_qwen35"] |
| examples | ["start","start tech --local","--code Hexagonal vs Clean Architecture pour un LLM gateway","--profile code Comment structurer un pipeline RAG avec ChromaDB ?","Quelle stratégie pour le Projet Chimere ?"] |
| model | qwen3.5-35b-a3b |
| execution | {"command":"/home/remondiere/.openclaw/bin/brainstorm.sh '$args'","timeout_ms":300000,"arg_mode":"raw"} |
Session de brainstorming avec mémoire persistante, RAG code/web, et profils de domaine.
Default engine: Claude Opus (via claude -p). Add --local for Qwen3.5-35B-A3B.
start [profile] — démarre session (profils: default, btp, cyber, finance, tech, code)stop — termine session (export automatique en markdown)status — état de la session couranteexport — exporte en markdown<message> — tour de brainstorming (RAG web + ChromaDB + LLM)Le profil code active :
code (1704 chunks : llama.cpp, OpenClaw, Python, Rust, Docker)enable_thinking=true, temp=0.8, top_p=0.95, top_k=20)--code — raccourci pour --profile code--profile NAME — profil explicite (default, btp, cyber, finance, tech, code)--local — utilise Qwen3.5-35B-A3B local au lieu de Claude Opus| Profile | Domain | Search | RAG | Thinking |
|---|---|---|---|---|
default | Brainstorming général | auto | auto | true |
tech | Architecture logicielle | auto | auto | true |
code | Code, patterns, libs, docs | deep | code | true, temp=0.8 |
btp | BTP / construction | auto | auto | true |
cyber | Cybersécurité | auto | auto | true |
finance | Finance / stratégie | auto | auto | true |
Code architecture design with web-informed debate between Architect and Reviewer agents — produces ASCII diagrams, justified stack, patterns (SOLID/DDD/CQRS), phased implementation plan.
CSV analysis via 2-pass SOTA orchestrator (score 54/55 on 7 CSVs). Statistics, data profiling, visualizations, insights.
Multi-perspective reasoning (advocate/critic/synthesizer). Three modes - general, --code for architecture decisions, --medical for EBM clinical analysis.
Deep reflection with web search — Brave/Perplexica pipeline fed into Qwen3.5-35B thinking mode for grounded, well-reasoned answers.
Threat intelligence analysis of IoCs (IPs, domains, hashes, URLs) via CyberBro (33 engines including Shodan/AbuseIPDB/VirusTotal) + Brave Search enrichment.
MCP CLI — list, configure, auth, and call MCP servers/tools directly (HTTP or stdio), including ad-hoc servers, config edits, and CLI/type generation.