What is the 'context problem' in multi-agent workflows?

It refers to the challenge where subagents are spawned with limited information, leading them to either guess which files are relevant or fail because they lack critical project-specific knowledge.

Can this skill help with large-scale codebase migrations?

Yes, by iteratively refining its search, the agent can map out dependencies and patterns across a large codebase that would otherwise exceed the context window.

How many cycles does the Iterative Retrieval loop perform?

The pattern is designed to run for a maximum of 3 cycles, which is typically sufficient to find high-relevance files without wasting time or tokens.

How does the skill determine if a file is relevant?

It uses an Evaluate phase that scores files from 0 to 1 based on how directly they implement target functionality or contain related patterns and types.

Iterative Context Retrieval

Name: Iterative Context Retrieval
Author: Nixdorfer

byNixdorfer

•

開発者ツール

Optimizes AI agent context by progressively refining code search through a multi-phase evaluation and feedback loop.

The Iterative Retrieval skill implements a sophisticated pattern to solve the 'context problem' in multi-agent AI workflows, where subagents often struggle with insufficient or irrelevant project data. By utilizing a four-phase cycle—Dispatch, Evaluate, Refine, and Loop—this skill ensures that Claude intelligently explores a codebase, discovers domain-specific terminology, and gathers exactly the right files needed for complex tasks. This method prevents context window overflow while ensuring the agent has the deep technical background necessary for accurate bug fixes and feature implementations.

主な機能

01Four-phase retrieval loop: Dispatch, Evaluate, Refine, and Loop

02Automated relevance scoring (0-1 scale) for retrieved source files

03Context gap identification to drive subsequent search refinements

04Dynamic keyword discovery to learn project-specific terminology

05Configurable cycle limits to balance context depth with processing efficiency

061 GitHub stars

ユースケース

01Locating critical logic for bug fixes in massive, unfamiliar repositories

02Identifying relevant architectural patterns when implementing new features

03Streamlining subagent workflows by providing precise context without manual file selection

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add nixdorfer/claudecodetool iterative-retrieval

For use in Claude.ai and ChatGPT

Download Skill