Why is there a limit of 3 cycles?

A 3-cycle limit acts as a circuit breaker to prevent infinite loops and ensure the agent moves to execution once 'sufficiently good' context is gathered, balancing precision with speed.

What is the 'context problem' in multi-agent workflows?

The context problem occurs when a sub-agent is spawned to perform a task but lacks the specific knowledge of which files or patterns are relevant to that task, leading to failures or hallucinations.

Can this skill reduce my LLM token costs?

Yes. By iteratively narrowing down to the most relevant files rather than sending the entire codebase or guessing, it ensures only high-value information consumes your token window.

How does the refinement phase work?

The refinement phase analyzes the results of previous searches to extract new keywords and directory patterns while explicitly excluding paths that were scored low in relevance.

Iterative Context Retrieval

Name: Iterative Context Retrieval
Author: affaan-m

byaffaan-m

•

112,917

•

开发者工具

Optimizes AI context retrieval through a multi-stage feedback loop to solve the 'missing context' problem in complex multi-agent coding workflows.

Iterative Context Retrieval is a sophisticated pattern designed for Claude Code and agentic workflows to overcome the limitations of fixed-context retrieval. By implementing a 4-stage cycle—Dispatch, Evaluate, Refine, and Loop—this skill enables agents to progressively discover the most relevant files and code patterns within a large codebase. It prevents common failures like exceeding context limits or guessing incorrectly by scoring file relevance and refining search criteria based on discovered terminology. This ensures sub-agents have exactly the information they need to perform their tasks accurately while significantly optimizing token usage.

主要功能

01Token usage optimization by excluding irrelevant paths and prioritizing top candidates

02Four-stage feedback loop (Dispatch, Evaluate, Refine, Loop) for precise context discovery

03Relevance scoring system (0.0 to 1.0) to filter high-impact files

04Dynamic search term refinement based on discovered codebase terminology

05Automatic gap identification to target missing architectural context

06112,917 GitHub stars

使用场景

01Implementing new features that must align with existing internal patterns and middlewares

02Reducing token costs in RAG-based AI coding workflows by narrowing search focus

03Locating relevant files for bug fixes in unfamiliar or massive codebases

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add affaan-m/everything-claude-code iterative-retrieval

For use in Claude.ai and ChatGPT

主要功能

01Token usage optimization by excluding irrelevant paths and prioritizing top candidates

02Four-stage feedback loop (Dispatch, Evaluate, Refine, Loop) for precise context discovery

03Relevance scoring system (0.0 to 1.0) to filter high-impact files

04Dynamic search term refinement based on discovered codebase terminology

05Automatic gap identification to target missing architectural context

06112,917 GitHub stars

使用场景

01Implementing new features that must align with existing internal patterns and middlewares

02Reducing token costs in RAG-based AI coding workflows by narrowing search focus

03Locating relevant files for bug fixes in unfamiliar or massive codebases