caveman-explore

Locate relevant code in repositories and return compact path:line citations.

Updated Aug 20, 2026
One-click install
npx skills add https://github.com/bittlinkm/claude-setup --skill caveman-explore-bittlinkm
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: caveman-explore
Source: https://github.com/bittlinkm/claude-setup/tree/main/skills/caveman-explore
Command: npx skills add https://github.com/bittlinkm/claude-setup --skill caveman-explore-bittlinkm

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Cold-start exploration of an unfamiliar codebase wastes tokens and time when an agent reads files one by one. This Skill performs fast, read-only repository searches and returns only compact file:line citations, keeping the main conversation context clean. ## Core Features & Use Cases - Parallel Broad Search: Issues multiple Glob, Grep, and Read calls in parallel to cover complementary hypotheses in a single turn. - Citation-Only Output: Returns strictly one path:START-END citation per line with a short relevance note, with no preamble or summary. - Honest Fallback: Reports "no relevant locations found" instead of guessing when nothing relevant exists. - Use Case: A solver agent needs to know where route selection logic lives in a large Go codebase. Delegate the question, and receive citations like src/router/pick.go:42-71 without any search noise entering the main conversation. ## Quick Start Ask the explorer to find where a specific feature or symbol is implemented in this repository and return only file:line citations.

Frequently Asked Questions about caveman-explore

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

FAQPage Schema
How do I find where a feature is implemented in an unfamiliar codebase?

Delegate a localization question to this read-only explorer, which fans out parallel Glob, Grep, and Read calls across likely paths and symbols. It returns a compact list of file:line citations identifying exactly where the relevant code lives.

How to search a repository without polluting the main conversation context?

Use a delegated read-only explorer whose reads and greps never enter the main conversation. Only the final citation block of path:line references is returned to the solver, minimizing token usage.

When should I use broad codebase exploration versus a direct search?

Use broad exploration for cold-start questions, cross-file localization, or after a direct search fails. Skip it when the issue already names the exact file or symbol, or a previous turn already returned usable file:line evidence.

Can this explorer edit files or run commands in the repository?

No. It is strictly read-only, using only Read, Glob, and Grep tools. It never edits files, executes commands, or proposes solutions; it only reports where relevant code lives.

What happens when the search finds nothing relevant in the repository?

The explorer replies with the single line "no relevant locations found" rather than guessing. It only cites line ranges it actually read and never invents or estimates ranges.