cavecrew

Route code investigation, small edits, and diff review to compressed subagents.

Updated Aug 27, 2026
One-click install
npx skills add https://github.com/ebarroso12/personal-skills --skill cavecrew-ebarroso12
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: cavecrew
Source: https://github.com/ebarroso12/personal-skills/tree/main/cavecrew
Command: npx skills add https://github.com/ebarroso12/personal-skills --skill cavecrew-ebarroso12

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It reduces main-context token usage by delegating code discovery, small edits, and diff review to caveman-style subagents that return compressed, tool-injected outputs.

Core Features & Use Cases

  • Decision guide for delegation: Chooses between cavecrew-investigator, cavecrew-builder, and cavecrew-reviewer versus inline work or vanilla alternatives based on task size and desired output style.
  • Scope-appropriate subagent routing: Uses investigator for “where is X / what calls Y / uses of Z”, builder for ≤2 file surgical edits, and reviewer for findings-only diff audits.
  • Output contracts for predictable reuse: Provides strict formats the main thread can rely on (e.g., path:line listings, edit verification, severity-sorted review).

Quick Start

Ask the AI to delegate code location, edit, and verification work to cavecrew by using the appropriate cavecrew subagent for your task scope.

Frequently Asked Questions about cavecrew

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

FAQPage Schema
How do I reduce main-context token consumption when navigating a codebase?

Reduce main-context token consumption by routing codebase navigation and surgical edits to compressed subagents that return path-anchored, tool-injected outputs. This prevents large file contents from bloating your main context window during investigation.

What is the best way to delegate small multi-file edits without losing scope?

Delegate small multi-file edits to a builder subagent designed for changes within a strict scope of two files or fewer. It performs surgical edits and verifies modifications while returning compressed, structured outputs to the main thread.

How do I perform findings-based diff review without loading the entire codebase?

Perform findings-based diff review by routing the audit task to a reviewer subagent. It returns severity-sorted findings without loading full file contents into the main context, ensuring structured, grep-safe outputs.

When should I use subagent delegation instead of inline code investigation?

Use subagent delegation for locating symbols, discovering what calls a function, or auditing diffs to save context tokens. Rely on inline work or vanilla alternatives for tasks outside these specific investigation, build, or review scopes.

Can I use compressed subagents for large codebase changes exceeding two files?

Compressed subagents are intended for investigations, findings-based reviews, and surgical edits of two files or fewer. Tasks requiring broader multi-file modifications should be handled inline or with alternative delegation methods to avoid strict scope limitations.

What output format should I expect from a code investigation subagent?

Code investigation subagents return path and line-anchored listings that are grep-safe and structured for predictable reuse. These compressed outputs ensure the main thread receives exact symbol locations without consuming unnecessary context tokens.