fanout

name	fanout
description	Fan-out/fan-in research. Spawns N researchers (Sonnet) in parallel to explore a question from different angles, then synthesizes findings with Opus.
user-invocable	true
allowed-tools	["Read","Agent","Grep","Glob","Bash","WebSearch","WebFetch"]
effort	high

/fanout — Fan-Out/Fan-In Research

Spawn N researcher agents (default 5) in parallel using Sonnet. Each explores the research question independently. An Opus synthesizer then merges all findings into a coherent answer.

$ARGUMENTS — the research question or topic.

Optional:

• n=X — number of researchers (default 5)
• model=haiku — use Haiku instead of Sonnet for researchers

Why this works

Parallel research catches angles a single agent misses. Each researcher has its own context window to dig deep. The Opus synthesizer sees all findings and produces higher-quality synthesis than any individual researcher.

How it works

1. Parse the research question
2. Spawn N researcher agents (Sonnet) in parallel
3. Each researcher independently explores, searches, reads
4. Collect all research outputs
5. Opus synthesizer merges findings into final answer

Researcher prompt

Research this question thoroughly:
[QUESTION]

You are one of [N] parallel researchers. Explore your own angle — don't assume others will cover what you skip.

Use available tools: WebSearch, WebFetch, Read, Grep, Glob.

Report findings in under 500 words. Lead with key insights. Cite sources.

Synthesizer prompt

Here are findings from [N] independent researchers on:
[QUESTION]

[ALL FINDINGS]

Synthesize into a comprehensive answer:
- What the researchers agree on
- Conflicting findings (if any)
- Key insights and evidence
- Your final synthesis

Be thorough but direct.

Execution

// Parse n from args, default 5
n = parse_arg("n") || 5
researcher_model = parse_arg("model") == "haiku" ? "haiku" : "sonnet"

// Fan out: spawn N researchers in parallel
researchers = []
for i in 1..n:
  researchers.push(Agent({
    subagent_type: "researcher",
    model: researcher_model,
    prompt: RESEARCHER_PROMPT.replace("[QUESTION]", question).replace("[N]", n),
    description: "Researcher " + i
  }))

// Collect all findings
findings = await_all(researchers)

// Fan in: Opus synthesizes
synthesis = Agent({
  subagent_type: "researcher",
  model: "opus",
  prompt: SYNTHESIZER_PROMPT.replace("[QUESTION]", question).replace("[N]", n).replace("[ALL FINDINGS]", format_findings(findings)),
  description: "Opus Synthesizer"
})

Output

Report directly:

• Researchers: N (model used)
• Individual researcher summaries (1-2 lines each)
• Opus synthesis