Dream Swarm Search Skill

Launch parallel multi-agent search workflows across multiple domains.

Architecture: Parallel Domain Agents

              ┌────────────────────────┐
              │     SEARCH QUERY       │
              │  "AI safety research"  │
              └───────────┬────────────┘
                          │
        ┌─────────────────┼─────────────────┐
        │                 │                 │
   ┌────▼────┐       ┌────▼────┐       ┌────▼────┐
   │  arXiv  │       │ GitHub  │       │  News   │
   │  Agent  │       │  Agent  │       │  Agent  │
   └────┬────┘       └────┬────┘       └────┬────┘
        │                 │                 │
        └─────────────────┼─────────────────┘
                          │
              ┌───────────▼───────────┐
              │    AGGREGATOR         │
              │  Dedupe & Synthesize  │
              └───────────────────────┘

When to Use

• Multi-source search across different APIs
• Comparative analysis from diverse sources
• Rapid information gathering with parallelization
• Fact-checking against multiple sources
• Research aggregation from academic + news + code

Available Domains (17 sources)

Domain	Description	Example Query
arxiv	Academic papers	"quantum computing"
github	Code repositories	"react components"
news	Recent articles	"AI regulation 2024"
wikipedia	Encyclopedia	"machine learning"
pubmed	Medical literature	"CRISPR therapy"
semantic_scholar	Academic index	"transformer architectures"
census	US demographics	"population density"
nasa	Space/Earth data	"mars rover"
youtube	Video content	"tutorial python"
weather	Forecasts	"Seattle forecast"
openlibrary	Book metadata	"science fiction"
fec	Campaign finance	"2024 donations"
judiciary	Court records	"antitrust case"
archive	Wayback Machine	"historical websites"
finance	Market data	"AAPL stock"
mal	Anime database	"studio ghibli"
wolfram	Computation	"integral sin(x)"

Scripts

Start Search Workflow

scripts/swarm-search.py "AI safety research" --domains arxiv github news
scripts/swarm-search.py "climate change" --agents 10 --parallel 5
scripts/swarm-search.py "quantum computing" --all-domains

Check Search Status

scripts/swarm-status.py workflow-xyz789
scripts/swarm-status.py workflow-xyz789 --results

Configuration Options

Parameter	Default	Description
--domains	arxiv,news	Comma-separated domain list
--agents	5	Number of parallel agents (1-20)
--parallel	3	Max concurrent domain searches (1-10)
--provider	xai	LLM provider for synthesis
--max-results	10	Results per domain
--stream	false	Enable SSE streaming
--output	stdout	Output format (stdout, json, markdown)

Example Workflows

Academic + Code Search

scripts/swarm-search.py "WebAssembly optimization" \
  --domains arxiv,github,semantic_scholar \
  --agents 6 --parallel 3

News + Social Monitoring

scripts/swarm-search.py "AI regulation policy" \
  --domains news,wikipedia,youtube \
  --agents 5 --output markdown

Comprehensive Research

scripts/swarm-search.py "CRISPR gene therapy" \
  --domains arxiv,pubmed,news,wikipedia \
  --agents 8 --parallel 4 --stream

Output Structure

{
  "task_id": "swarm-xyz789",
  "status": "completed",
  "results": {
    "arxiv": [
      {"title": "...", "url": "...", "abstract": "..."}
    ],
    "github": [
      {"name": "...", "url": "...", "description": "..."}
    ],
    "news": [
      {"title": "...", "url": "...", "source": "..."}
    ]
  },
  "synthesis": {
    "summary": "...",
    "key_themes": [...],
    "cross_domain_insights": [...]
  },
  "metadata": {
    "domains_searched": 3,
    "total_results": 24,
    "execution_time": 12.5
  }
}

Advantages over Dream Cascade

Aspect	Dream Swarm	Dream Cascade
Speed	Faster (parallel)	Slower (hierarchical)
Depth	Broad coverage	Deep analysis
Use case	Data gathering	Research synthesis
Agent count	Many simple agents	Fewer complex agents
Cost	Lower	Higher

Prerequisites

• MCP server running on port 5060
• Valid API keys for data sources
• Python 3.10+ with requests library

Troubleshooting

"Domain not available": Check API key configured for that domain

"Rate limited": Reduce --parallel count

"No results": Try broader query or different domains

Related Skills

• dream-cascade - Hierarchical deep research
• data-fetch - Single domain direct access
• code-quality - Code-specific searches