What is the subagent context problem?

It occurs when a specialized AI agent lacks the necessary codebase knowledge (files, patterns, or terminology) to begin its task effectively, leading to 'guessing' or context window limits.

How many cycles does the iterative retrieval process take?

The pattern is designed to run for a maximum of 3 cycles, which is typically sufficient to find high-relevance context without excessive overhead.

How does this skill handle different naming conventions?

During the 'Refine' phase, the skill identifies local terms used in the codebase (e.g., finding 'throttle' when searching for 'rate limit') and updates subsequent queries to match the project's vocabulary.

Can I use this for any programming language?

Yes, the iterative retrieval pattern is language-agnostic as it relies on directory structures, keyword matching, and AI-driven relevance evaluation.

Iterative Context Retrieval

Name: Iterative Context Retrieval
Author: Nixdorfer

byNixdorfer

•

生産性とワークフロー

Optimizes AI agent performance by progressively refining codebase context through a multi-stage evaluation and search loop.

This skill implements the Iterative Retrieval pattern to solve the 'context gap' common in multi-agent workflows. Instead of overwhelming Claude with irrelevant files or guessing requirements, it uses a structured 4-phase cycle (Dispatch, Evaluate, Refine, Loop) to identify high-relevance code patterns and terminology. By scoring file relevance and identifying information gaps, it ensures that agents receive precise, necessary context—typically within three refinement cycles—making it ideal for complex bug fixes and feature implementations in large, unfamiliar repositories.

主な機能

01Context Optimization: Filters and merges context to ensure only high-relevance files are passed to sub-agents.

024-Phase Retrieval Lifecycle: Implements Dispatch, Evaluate, Refine, and Loop stages for systematic context gathering.

03Adaptive Search Refinement: Learns codebase-specific terminology and naming conventions during the retrieval process.

04Relevance Scoring: Ranks files on a 0-1 scale to prioritize critical implementation details and patterns.

05Gap Identification: Automatically detects missing terminology or dependencies to guide subsequent search queries.

061 GitHub stars

ユースケース

01Implementing new features that require understanding specific architectural patterns across multiple directories.

02Onboarding AI sub-agents to large codebases where manual context selection is inefficient or prone to error.

03Resolving complex bugs where the specific failure point or relevant utility files are unknown.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add nixdorfer/claudecodetool iterative-retrieval

For use in Claude.ai and ChatGPT

Download Skill