What is the 'context problem' in AI coding?

It refers to the challenge of providing an AI agent with enough relevant information to solve a task without exceeding its context window or polluting its reasoning with irrelevant data.

Can this skill identify project-specific terminology?

Yes, during the 'Refine' phase, it extracts keywords and naming patterns from initial results to better target subsequent searches for related code.

How many cycles does the retrieval loop perform?

The pattern is designed to perform a maximum of three cycles, which is typically sufficient to find the 'high-relevance' files needed for most development tasks.

Does this help reduce token consumption?

Yes, by iteratively narrowing down the focus to files with a relevance score of 0.7 or higher, it avoids sending massive, unnecessary code blocks to the LLM.

How does it decide if a file is relevant?

It uses a scoring system where files are rated from 0 to 1 based on whether they implement target functionality, contain related types, or are tangentially related.

Iterative Context Retrieval

Name: Iterative Context Retrieval
Author: Eotel

byEotel

0•

Herramientas para Desarrolladores

Implements a multi-phase loop to progressively refine code context retrieval for more accurate AI agent performance.

The Iterative Retrieval skill solves the common 'context problem' in AI-driven development by preventing agents from being overwhelmed by irrelevant data or hampered by insufficient information. By utilizing a standardized four-phase cycle—Dispatch, Evaluate, Refine, and Loop—this skill allows Claude to explore a codebase intelligently, learning local terminology and hidden file relationships on the fly. This ensures that subagents receive the most relevant code snippets for a given task while staying within context window limits, significantly increasing the success rate of complex refactoring and bug-fixing operations.

Características Principales

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

02Automated relevance scoring (0.0 to 1.0) for every retrieved file

03Contextual gap identification to find missing dependencies or related patterns

040 GitHub stars

05Optimized token management by filtering out low-relevance candidate files

06Dynamic keyword discovery to adapt to project-specific naming conventions

Casos de Uso

01Locating relevant logic for complex bug fixes in large, unfamiliar repositories

02Mapping dependencies and middleware patterns before implementing new features

03Improving subagent accuracy in multi-agent workflows by providing precise, filtered context

Características Principales

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

02Automated relevance scoring (0.0 to 1.0) for every retrieved file

03Contextual gap identification to find missing dependencies or related patterns

040 GitHub stars

05Optimized token management by filtering out low-relevance candidate files

06Dynamic keyword discovery to adapt to project-specific naming conventions

Casos de Uso

01Locating relevant logic for complex bug fixes in large, unfamiliar repositories

02Mapping dependencies and middleware patterns before implementing new features

03Improving subagent accuracy in multi-agent workflows by providing precise, filtered context