Can it handle codebases with unique terminology?

Yes, the Refine stage identifies specific keywords and patterns found during the first cycle to improve the accuracy of subsequent retrieval attempts.

How does this skill prevent context window limits?

Instead of sending all files at once, it evaluates the relevance of retrieved files and only keeps those with high scores (0.7+), effectively filtering out noise.

What is the Iterative Retrieval pattern?

It is a 4-stage process—Dispatch, Evaluate, Refine, and Loop—that helps AI agents find the most relevant files in a codebase by starting with broad searches and narrowing down based on findings.

How many cycles should the retrieval process run?

The pattern recommends a maximum of 3 cycles, which is typically enough to find the necessary context without causing unnecessary delays or token costs.

Iterative Retrieval Pattern

Name: Iterative Retrieval Pattern
Author: saar120

bysaar120

0•

Developer Tools

Optimizes code context retrieval for multi-agent workflows through a systematic four-stage evaluation and refinement loop.

The Iterative Retrieval skill implements a sophisticated RAG-like pattern for Claude Code, designed to solve the common 'context problem' in complex multi-agent orchestrations. By moving through iterative phases of dispatch, evaluation, and refinement, it allows agents to discover and identify high-relevance files without overwhelming the context window or making incorrect guesses about codebase structure. This skill is essential for large-scale bug fixing and feature implementation tasks where the necessary files and terminology aren't immediately obvious, ensuring agents have exactly the information they need to succeed.

Key Features

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

02Gaps identification to drive targeted follow-up queries

030 GitHub stars

04Sophisticated relevance scoring (0-1 scale) to eliminate context noise

05Automated terminology discovery to adapt to codebase-specific naming conventions

06Token usage optimization by prioritizing high-impact files over broad searches

Use Cases

01Resolving 'context too large' or 'missing context' failures in autonomous agent tasks

02Optimizing multi-agent workflows where sub-agents require specific domain knowledge

03Navigating large, unfamiliar codebases where search terms are unknown

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add saar120/skillgate-registry iterative-retrieval

For use in Claude.ai and ChatGPT

Key Features

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

02Gaps identification to drive targeted follow-up queries

030 GitHub stars

04Sophisticated relevance scoring (0-1 scale) to eliminate context noise

05Automated terminology discovery to adapt to codebase-specific naming conventions

06Token usage optimization by prioritizing high-impact files over broad searches

Use Cases

01Resolving 'context too large' or 'missing context' failures in autonomous agent tasks

02Optimizing multi-agent workflows where sub-agents require specific domain knowledge

03Navigating large, unfamiliar codebases where search terms are unknown

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add saar120/skillgate-registry iterative-retrieval

For use in Claude.ai and ChatGPT