cavecrew

Delegate coding tasks to specialized subagents with compressed output contracts.

Updated Jul 6, 2026
One-click install
npx skills add https://github.com/shirulot/codex-skill --skill cavecrew-shirulot
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: cavecrew
Source: https://github.com/shirulot/codex-skill/tree/main/cavecrew
Command: npx skills add https://github.com/shirulot/codex-skill --skill cavecrew-shirulot

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill solves the issue of context exhaustion during long coding sessions by delegating tasks to specialized subagents that return highly compressed, token-efficient output.

Core Features & Use Cases

  • Context Compression: Reduces subagent output size by approximately 60% compared to standard agents, allowing for longer, more complex task chains.
  • Specialized Roles: Provides three distinct subagents: an investigator for code location, a builder for surgical edits, and a reviewer for diff analysis.
  • Use Case: When performing a cross-file refactor, use the investigator to locate symbols, the builder to apply changes, and the reviewer to audit the diff, all while keeping the main conversation context lean.

Quick Start

Delegate the task of locating the definition of the user authentication function to the cavecrew investigator.

Frequently Asked Questions about cavecrew

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I optimize context usage during long coding sessions?

To optimize context usage, delegate tasks to specialized subagents that return highly compressed output. This reduces subagent response size by approximately 60% compared to standard agents, extending the main thread context window longevity.

What's the best way to handle cross-file refactoring without exhausting context?

Handling cross-file refactoring without exhausting context involves using specialized subagent roles: an investigator to locate symbols, a builder to apply surgical edits, and a reviewer to audit the diff. This framework keeps the main conversation context lean.

Can I use subagent delegation for surgical file editing and diff review?

Yes, you can use subagent delegation for surgical file editing and diff review by utilizing a builder subagent for edits and a reviewer subagent for diff analysis. They enforce compressed output contracts to minimize verbose prose in responses.

Why does context exhaustion happen in complex task chains and how to prevent it?

Context exhaustion in complex task chains happens because standard agents generate verbose prose that fills the context window. Prevent it by enforcing compressed output contracts on subagents, reducing output size and allowing for longer task chains.

Does this token-efficiency approach work for code investigation tasks?

Yes, this token-efficiency approach works for code investigation by using a specialized investigator subagent. It locates code definitions and symbols while returning compressed output to maintain main thread context window longevity.