What happens if a subagent fails during the process?

The skill includes failure recovery logic that can retry tasks or split a failing agent's workload into smaller pieces to bypass context limit errors.

How does it handle files that are too large for one agent?

The skill automatically calculates the corpus size and segments the data into chunks that fit within agent context limits, using overlap to ensure no context is lost between segments.

What is a two-stage fan-out pattern?

It is an orchestration strategy where data is first distributed to parallel 'Worker' agents for analysis, then verified by 'Critic' agents for consistency, before being synthesized into a final report.

Can I customize the depth of the analysis?

Yes, you can choose from three effort levels: 'Some effort' (default), 'A lot of effort' (thorough), or 'Herculean effort' (maximum redundancy) depending on your needs.

Is this suitable for source code analysis?

Absolutely. It includes specific logic for token estimation in source code and can navigate complex directory structures to gather analysis inputs.

Large Corpus Fan-Out Analyzer

Name: Large Corpus Fan-Out Analyzer
Author: ed3dai

byed3dai

•

数据科学与机器学习

Orchestrates a multi-stage subagent pipeline to analyze large datasets or codebases that exceed standard context windows.

This skill enables Claude to process massive volumes of text, code, or data by distributing the workload across a structured hierarchy of subagents. It implements a two-stage fan-out pattern where 'Worker' agents analyze segments, 'Critic' agents review findings for consistency and gaps, and a 'Summarizer' synthesizes the final report. With built-in task tracking, failure recovery, and configurable effort levels, it is an ideal solution for comprehensive security audits, deep documentation reviews, or large-scale codebase analysis where accuracy and thoroughness are paramount.

主要功能

01Redundant review cycles (Critics) to identify gaps and inconsistencies

02Dynamic task dependency tracking with failure recovery and auto-splitting

035 GitHub stars

04Parallel subagent orchestration for Workers, Critics, and Summarizers

05Automatic corpus segmentation and token-limit management

06Configurable effort levels to balance analysis speed and depth

使用场景

01Synthesizing insights from extensive legal, medical, or technical documentation sets

02Performing comprehensive architectural or security audits on massive codebases

03Identifying cross-module inconsistencies and patterns in large-scale data repositories

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add ed3dai/ed3d-plugins-testing doing-a-simple-two-stage-fanout

For use in Claude.ai and ChatGPT

Download Skill