| name | skillweave-council |
| description | Multi-Model LLM Council — Deliberation, Peer Review, and Chairman Synthesis with Web Search and Faigate Routing |
| argument-hint | topic="[topic]" time_range="[30d|quarter|6mo|1yr|any]" mode="[quick|standard|full]" profile="[default|quick|deep|expert]" output="[markdown|json]" phase="[discovery|design|code_review|post_release]" |
skillweave-council
Collective AI Intelligence — Instead of asking one LLM, convene a council of AI models that deliberate, peer-review, and synthesize the best answer.
Overview
skillweave-council is a standalone SkillWeave skill that implements the LLM Council pattern:
- Stage 1 — Deliberation: Multiple models answer independently in parallel (via Faigate routing)
- Stage 2 — Peer Review: Models anonymously rank each other's responses
- Stage 3 — Chairman Synthesis: A chairman model synthesizes the final answer
Optional Stage 0: Web Search (DuckDuckGo, Serper, Tavily, Brave) grounds responses in real-time data with configurable time ranges.
Usage
/skillweave-council topic="Wettbewerbsanalyse Q2 2026" time_range="quarter" mode="full"
/skillweave-council topic="Evaluate this architecture decision..." mode="full" output="json" phase="code_review"
/skillweave-council command="search" topic="latest AI trends" time_range="30d"
/skillweave-council command="profiles"
/skillweave-council command="compare" topic="..." profile="quick"
Commands
| Command | Description | Stages |
|---|
deliberate | Full council deliberation (default) | 0→1→2→3 |
compare | Side-by-side model comparison | 0→1 |
search | Web search only (no models) | 0 |
profiles | List available council profiles | — |
Parameters
| Parameter | Values | Default | Description |
|---|
topic | string | required | Question or topic for the council |
time_range | 30d, quarter, 6mo, 1yr, any | any | Web search time window |
mode | quick, standard, full | full | Execution depth |
profile | default, quick, deep, expert | default | Pre-configured model set |
output | markdown, json | markdown | Output format |
phase | discovery, design, code_review, post_release | — | SkillWeave lifecycle phase context |
Council Profiles
| Profile | Models | Mode | Use Case |
|---|
| default | 4 models (balanced) | standard | General-purpose deliberation |
| quick | 2 models (fastest) | quick | Rapid comparison, low latency |
| deep | 6 models (diverse) | full | Comprehensive analysis |
| expert | 4 models (top-tier) | full | High-stakes decisions, research |
Execution Modes
| Mode | Stages | Description |
|---|
| quick | Stage 1 only | Models answer independently, no review |
| standard | Stages 1 + 2 | Models answer + peer review with rankings |
| full | All 3 stages | Deliberation + peer review + chairman synthesis |
Web Search Providers
| Provider | Type | Key Required |
|---|
| DuckDuckGo | Free web+news | No |
| Serper | Google results | Yes (SERPER_API_KEY) |
| Tavily | LLM-optimized | Yes (TAVILY_API_KEY) |
| Brave | Privacy-focused | Yes (BRAVE_API_KEY) |
Phase Integration
skillweave-council integrates with the SkillWeave 7-phase lifecycle:
| Phase | Council Focus | Typical Use |
|---|
| discovery | Market research, competitive analysis | Replaces last30days |
| design | Design evaluation, UX critique | Optional design review |
| code_review | Architecture review, code quality | Build-phase quality gate |
| post_release | Feedback analysis, pattern detection | Release retrospective |
Output Format
Markdown (default)
A structured markdown document with the chairman's synthesis, council rankings, and individual model responses.
JSON
Structured output with schema validation:
{
"title": "Concise answer title",
"summary": "Executive summary",
"key_insights": ["Insight 1", "Insight 2"],
"consensus_score": 0.85,
"dissent": "Minor disagreement on timeline",
"sources": ["Source A", "Source B"]
}
Faigate Integration
skillweave-council uses Faigate for model routing:
- Automatic availability check before queries
- Credit balance verification
- Failed models gracefully excluded (no council abort)
- Multi-provider routing through a single API
No API key management needed — Faigate handles authentication internally.
Sandbox Preflight
The council validates:
- Faigate endpoint reachable
- At least 2 models available (for peer review)
- Chairman model available (for full mode)
- Web search provider configured (if time_range specified)
- JSON schema valid (if output=json)
Testing
| Test | Description |
|---|
| Stage 1 parallel execution | All models respond within timeout |
| Stage 1 graceful failure | 1 model fails, others succeed |
| Stage 2 anonymization | Responses anonymized as A/B/C |
| Stage 2 ranking parse | FINAL RANKING: format parsed correctly |
| Stage 3 markdown synthesis | Chairman produces valid markdown |
| Stage 3 JSON output | Valid JSON with schema compliance |
| Faigate availability mock | Check returns model statuses |
| Web search DuckDuckGo | Real search returns results |
| Time-range filter | 30d/quarter/6mo/1yr filtering |
| Council profile selection | All 4 profiles load correctly |
| Phase context injection | Discovery/design prompts modify output |
| Standalone mode | No phase context = generic prompt |
