Is this skill useful for multi-agent systems?

Yes, it is specifically designed to help sub-agents gain the precise context they need when they are spawned with limited information to perform a sub-task.

How does it solve context window issues?

Instead of sending the entire codebase to the AI, it selectively retrieves and evaluates files, only keeping those with high relevance scores (typically 0.7 or higher).

What is the Iterative Retrieval pattern?

It is a workflow where an AI agent performs an initial broad search, evaluates the results, and refines its search criteria multiple times to find the most relevant code for a specific task.

How many cycles are recommended for context retrieval?

The skill suggests a maximum of 3 cycles, which is usually sufficient to identify necessary context without incurring excessive latency or token costs.

Why use this over standard keyword search?

Standard search often misses files if the exact terminology isn't known; this pattern learns the codebase's specific jargon (e.g., using 'throttle' instead of 'rate limit') during the first retrieval cycle.

Iterative Context Retrieval

Name: Iterative Context Retrieval
Author: Nixdorfer

byNixdorfer

•

개발자 도구

Optimizes code context for sub-agents through a multi-stage search and refinement loop to resolve complex repository queries.

Iterative Context Retrieval is a specialized pattern designed to solve the 'context problem' in multi-agent workflows where sub-agents lack sufficient information to start tasks. By implementing a four-stage cycle—Dispatch, Evaluate, Refine, and Loop—this skill enables Claude to progressively discover relevant files, learn project-specific terminology, and identify architectural patterns without exceeding context window limits. It effectively bridges the gap between massive codebases and specific coding tasks by filtering for high-relevance files and identifying critical information gaps in every cycle.

주요 기능

01Dynamic terminology discovery to learn codebase-specific naming conventions

02Intelligent context filtering to prevent context window overflow

03Gap analysis to drive targeted subsequent searches

04Automated relevance scoring (0-1 scale) for retrieved files

051 GitHub stars

06Four-stage retrieval cycle: Dispatch, Evaluate, Refine, and Loop

사용 사례

01Discovering existing implementation patterns before adding new features

02Locating relevant files for complex bug fixes in unfamiliar codebases

03Optimizing sub-agent performance by providing precise, high-relevance context

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