What is the subagent context problem?

It refers to the challenge where secondary AI agents are initialized with limited context and don't know which specific files or patterns are needed to complete their assigned task.

Does this skill help with token limit issues?

Yes, it specifically prevents 'sending everything' which often exceeds context limits, by identifying and only including high-relevance code snippets.

How does the evaluation scoring work?

Files are scored from 0.0 to 1.0. High relevance (0.8-1.0) files are directly related to the task, while low relevance files are filtered out to keep the context window clean.

Why is a 3-cycle limit recommended for the loop?

Three cycles typically provide enough opportunity to discover project-specific terminology and find the top 3-5 most relevant files without wasting processing time.

Iterative Retrieval Pattern

Name: Iterative Retrieval Pattern
Author: oabdelmaksoud

byoabdelmaksoud

0•

Developer Tools

Optimizes AI agent performance by progressively refining context retrieval through a four-stage evaluation loop.

The Iterative Retrieval skill solves the 'subagent context problem' where AI agents lack sufficient information about a codebase before starting a task. By implementing a systematic loop of Dispatch, Evaluate, Refine, and Loop, this skill allows Claude to start with broad queries and narrow down to the most relevant files based on scoring and gap identification. This pattern prevents token limit exhaustion from over-sending files while ensuring the agent doesn't work with missing or guessed information, making it essential for complex debugging and feature implementation in large-scale repositories.

Key Features

01Intelligent gap identification to drive subsequent search cycles

02Relevance-based scoring (0.0 to 1.0) for context quality control

030 GitHub stars

044-Stage Context Refinement Loop (Dispatch, Evaluate, Refine, Loop)

05Automated discovery of codebase-specific terminology and patterns

06Optimized token usage by filtering for high-relevance content

Use Cases

01Isolating cross-dependency bugs where relevant files aren't immediately obvious

02Standardizing context gathering for subagents in multi-step workflows

03Navigating unfamiliar file structures in massive monorepos

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add oabdelmaksoud/superclaw iterative-retrieval

For use in Claude.ai and ChatGPT

Key Features

01Intelligent gap identification to drive subsequent search cycles

02Relevance-based scoring (0.0 to 1.0) for context quality control

030 GitHub stars

044-Stage Context Refinement Loop (Dispatch, Evaluate, Refine, Loop)

05Automated discovery of codebase-specific terminology and patterns

06Optimized token usage by filtering for high-relevance content

Use Cases

01Isolating cross-dependency bugs where relevant files aren't immediately obvious

02Standardizing context gathering for subagents in multi-step workflows

03Navigating unfamiliar file structures in massive monorepos

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add oabdelmaksoud/superclaw iterative-retrieval

For use in Claude.ai and ChatGPT