fanout-analysis

Runs parallel analysis branches and synthesizes their findings into one evidence-based answer.

1.8k|303|Updated May 14, 2025
One-click install
npx skills add https://github.com/trpc-group/trpc-agent-go --skill fanout-analysis
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: fanout-analysis
Source: https://github.com/trpc-group/trpc-agent-go/tree/main/examples/dynamicworkflow/skills/skills/fanout-analysis
Command: npx skills add https://github.com/trpc-group/trpc-agent-go --skill fanout-analysis

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Complex questions often need multiple independent perspectives, but ad-hoc prompting mixes viewpoints together and hides disagreement or missing evidence. This Skill turns a request into a temporary fan-out/fan-in workflow so each angle is analyzed independently and then merged by a dedicated synthesizer.

Core Features & Use Cases

  • Parallel Branch Analysis: Runs two to four independent analysis branches concurrently, each with its own workflow-local Agent and only the context it needs.
  • Evidence-Based Synthesis: A separate synthesizer role distinguishes agreement, disagreement, and missing evidence, and can return a structured object with recommendation, reasons, and uncertainties.
  • Bounded Failure Handling: Failed branches return explicit null results that are surfaced as missing evidence instead of being silently treated as support or retried indefinitely.
  • Use Case: When evaluating whether to migrate a service to a new database, fan out branches for cost, performance, and operational risk, then synthesize a single recommendation with explicit uncertainties.

Quick Start

Use the fanout-analysis skill to analyze this decision from several independent angles and synthesize a recommendation with reasons and uncertainties.

Frequently Asked Questions about fanout-analysis

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run multiple analysis agents in parallel in a workflow?

Express the independent branches with a parallel construct, giving each branch its own workflow-local Agent instance and only the context it needs. Preserve the input order when passing the branch results to the synthesizer.

How many parallel branches should a fan-out analysis use?

Use two to four independent angles that cover different evidence or reasoning needs. Do not create extra roles just to make the workflow look larger, and never let one branch depend on another branch's unfinished answer.

What happens when a parallel branch fails in a fan-out workflow?

A failed independent branch returns a null result, which must be handled explicitly as missing evidence. The workflow or its caller decides whether a single bounded rerun is appropriate; unbounded retries and silent treatment as support are not allowed.

Can the synthesizer return structured output instead of text?

Yes. When a decision is required, the synthesizer returns a small structured object with recommendation, reasons, and uncertainties. Read the Agent result's explicit structured field rather than parsing JSON-looking text.

When should I not use a fan-out/fan-in analysis workflow?

Avoid it for simple questions that a single reasoning pass can answer, or when branches would depend on each other's results. It fits questions that genuinely benefit from several independent perspectives followed by one synthesis.