What is the 'context problem' this skill solves?

It solves the issue where AI sub-agents lack the necessary information to perform tasks because they don't know which files or patterns are relevant to a codebase before they begin working.

How does the skill handle irrelevant search results?

The Evaluate stage scores files on a scale; files with low relevance (below 0.2) are explicitly excluded from the Refine stage to ensure future search cycles are more accurate.

How many cycles does the iterative loop typically run?

The pattern is designed to run for a maximum of 3 cycles, which is usually sufficient to narrow down the most relevant files while maintaining efficiency.

Can this skill adapt to different coding styles?

Yes, it learns the project's specific terminology and patterns during the first broad search cycle and uses those discoveries to refine its subsequent queries.

Iterative Context Retrieval

Name: Iterative Context Retrieval
Author: unju-ai

byunju-ai

0•

开发者工具

Optimizes AI context retrieval through a multi-stage evaluation loop to provide sub-agents with precisely the information they need.

The Iterative Retrieval skill addresses the 'context problem' in multi-agent workflows where sub-agents struggle to identify relevant files in large codebases. By implementing a four-stage process—Dispatch, Evaluate, Refine, and Loop—this pattern enables agents to start with broad queries and progressively narrow down the most relevant code and patterns. This prevents context window overflow and eliminates guesswork, ensuring that the AI has the specific terminology, architecture, and files required to complete tasks accurately without manual intervention.

主要功能

010 GitHub stars

02Precise relevance scoring (0.0 to 1.0) for granular file selection

03Identification of context gaps to drive targeted subsequent searches

04Four-stage retrieval loop: Dispatch, Evaluate, Refine, and Loop

05Automatic discovery of project-specific terminology and naming conventions

06Configurable cycle limits to optimize performance and prevent token waste

使用场景

01Identifying architectural patterns and existing middleware when implementing new API features

02Locating specific logic across large, unfamiliar codebases for complex bug fixes

03Gathering comprehensive context for multi-agent sub-tasks without exceeding token limits

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add unju-ai/ecc iterative-retrieval

For use in Claude.ai and ChatGPT