How many cycles does the retrieval pattern run?

The pattern is designed to run for a maximum of 3 cycles, stopping as soon as sufficient high-relevance context (typically 3+ files with a score above 0.7) is identified.

Does this skill help reduce LLM token costs?

Yes. By progressively refining searches and excluding irrelevant paths, it avoids 'sending everything' to the model, which keeps the context window lean and cost-effective.

Can I use this for non-TypeScript projects?

Absolutely. While the examples use TypeScript, the iterative logic and relevance scoring patterns are language-agnostic and work for any codebase structure.

What is the 'subagent context problem'?

It occurs when secondary agents are spawned without knowing which specific files they need to solve a task, often leading to missing information or context window overflow.

How does it handle project-specific terminology?

During the evaluation phase, the skill identifies terminology gaps (e.g., discovering the project uses 'throttle' instead of 'rate-limit') and adds these keywords to the next search cycle.

Iterative Context Retrieval

Name: Iterative Context Retrieval
Author: saar120

bysaar120

0•

生产力与工作流

Refines codebase context through an intelligent multi-cycle search pattern to provide subagents with the precise information needed for complex tasks.

The Iterative Retrieval skill addresses the critical 'context problem' in multi-agent workflows where subagents often lack the necessary codebase information to perform tasks effectively. By implementing a sophisticated 4-phase loop—Dispatch, Evaluate, Refine, and Loop—it allows agents to progressively discover relevant files, identify project-specific terminology, and filter out noise. This method prevents the failure modes of sending too much or too little information, ensuring that Claude Code subagents operate with high-relevance context while minimizing token consumption and maximizing accuracy in large or unfamiliar repositories.

主要功能

014-phase iterative loop for progressive context discovery

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

03Dynamic query refinement based on discovered project terminology

040 GitHub stars

05Intelligent gap identification to target missing context in subsequent cycles

06Token optimization by filtering out low-relevance content

使用场景

01Navigating large legacy codebases where logic is distributed across multiple undocumented layers

02Implementing new features that must strictly adhere to existing project-specific naming conventions and patterns

03Resolving complex bugs that require tracing data flow through specific project middlewares and utilities

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

主要功能

014-phase iterative loop for progressive context discovery

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

03Dynamic query refinement based on discovered project terminology

040 GitHub stars

05Intelligent gap identification to target missing context in subsequent cycles

06Token optimization by filtering out low-relevance content

使用场景

01Navigating large legacy codebases where logic is distributed across multiple undocumented layers

02Implementing new features that must strictly adhere to existing project-specific naming conventions and patterns

03Resolving complex bugs that require tracing data flow through specific project middlewares and utilities