caveman-discover Skill for Claude: Find and Label Every LLM Workflow in Your Repo

caveman-discover makes your assistant find every LLM job in your repo and label it so Caveman Cloud can split spend by workflow.

Installs: 111.2K; on skills.sh
Stars: 116.7K; on GitHub
Updated: Nov 2026; last repo commit

caveman-discover walks your repo from its entry points: handlers, scheduled jobs, CLI scripts, eval harnesses and agent definitions. It groups them into workflows, one per job a person would name, and proposes slugs such as support-reply. It shows you a table first and changes code only after you agree.

Then it adds a workflow label with the lightest method for each callsite, such as an x-cave-workflow header or a setting in the Caveman SDK, and checks one labeled path still works. Labels only count on traffic that already goes through the Caveman gateway. It bundles no scripts.

Good for

Developers who ship LLM features and want spend split by job
Teams already sending traffic through the Caveman gateway
Anyone who needs an inventory of LLM entry points in a repo

Install caveman-discover

Claude Code

Copy this prompt.

Install the caveman-discover skill in Claude Code. Run this command:

npx skills add https://github.com/juliusbrussee/caveman --skill caveman-discover

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

From its listing on skills.sh. Source.

What's in the skill

When it loads

Find and label every LLM workflow in the repository so Caveman Cloud groups spend by workflow instead of one bucket. Use for "discover workflows" or breaking LLM spend down by workflow.

Sections in SKILL.md

Step 1 — Inventory the workflows
Step 2 — Name them
Step 3 — Propose, then apply
Step 4 — Verify
Step 5 — Report
Workflows labeled

Key facts

Author: juliusbrussee (GitHub)
Repository: juliusbrussee/caveman, folder skills/caveman-discover
Installs on skills.sh: 111,210 (as of 2026-11-02)
Repo stars: 116,702 (whole repository, as of 2026-11-02)
License: Apache-2.0
Last repo commit: 2026-11-01
Files in the skill folder: 1 file: SKILL.md
skills.sh audits: Gen Agent Trust Hub: pass, Socket: warn, Snyk: pass

What it includes

Find handlers, jobs, scripts, evals and agents that call an LLM
Walks the repo from its entry points and counts one workflow per job a person would name.
Read · Step 1: Inventory the workflows
Slugs like support-reply; lowercase letters, digits, - and _
Names the job, not the tech, in 1 to 96 characters, and marks guessed names for review.
Step 2: Name them
Show a table, wait for your OK, then add labels
Adds a label per callsite: an SDK option, an x-cave-workflow header, a --workflow flag or CAVE_WORKFLOW variable for wrapped agents, or a header on raw HTTP.
Write · Step 3: Propose, then apply
Run one existing labeled path and check it still works
Runs a test, dev script or one request, and reports what it saw. The gateway rejects a bad label with an error.
Runs code · Step 4: Verify
Table of workflows, what was verified, what is not wired
Lists labeled workflows, verification, calls not routed through the gateway, and names marked for review.
Step 5: Report

From the official docs

Permissions

Medium risk

It edits code at each LLM callsite and runs one test request that may call a paid model. Approve the table first and review the diff.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

It edits code across your repo
It adds labels at every LLM callsite it finds, and runs one request to check. Review the table before you agree and review the diff after.
Its test request uses your code path
Verification runs a path your repo already has, such as a test or one request. That path may call a paid model.

Technical details

Talks to	Caveman gateway, when it runs a labeled request

Questions

What does caveman-discover do?

It finds the jobs in your repo that call an LLM, names each one, and adds a workflow label so Caveman Cloud groups spend by workflow instead of one unlabeled bucket.

How do I install caveman-discover?

Use the install box at the top of the page. The skill file is plain instructions, but the labels only help if your traffic goes through the Caveman gateway, which caveman-setup wires up.

Will it change my code without asking?

No. The SKILL.md says to propose the table first and apply after you agree. Running it again on an already labeled repo should change nothing.

Does it need an API key?

The SKILL.md asks for no key itself. The labels travel with requests that already carry your Caveman gateway key, and calls not routed through the gateway are listed as not wired instead of labeled.

Is caveman-discover safe?

It edits code, so use a branch and review the diff. It shows the table of changes before it applies them, and verification runs a path your repo already has.

How is it different from caveman-setup?

caveman-setup routes your LLM calls through the Caveman gateway so they are measured. caveman-discover then labels those calls by workflow. The SKILL.md says wiring is the setup skill's job.

Information

updated Nov 2, 2026

Repository: juliusbrussee/caveman
Folder: skills/caveman-discover

juliusbrussee/caveman skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 3

caveman-discover SKILL.md · github.com
caveman README · github.com
caveman-setup SKILL.md · github.com