caveman-evidence-review Skill for Claude: Read-Only Review of LLM Spend and Traces

caveman-evidence-review makes your assistant read your Caveman Cloud data and explain where LLM spend goes, without changing anything.

Installs: 111.3K; on skills.sh
Stars: 116.7K; on GitHub
Updated: Nov 2026; last repo commit

caveman-evidence-review loads the project you have selected in Caveman, reads cost, score, workflow and savings reports, and then checks the leading explanation against traces. It keeps four numbers apart: measured cost, inferred daily headroom, verified savings and evidence cost. It cites trace ids and time windows and does not claim a cause from an aggregate alone.

It is read-only. It never starts, approves, cancels or rolls back an experiment, and it does not fetch prompt, completion or tool payloads unless you ask. It needs a Caveman login and project, through the MCP tools or the CLI fallback. It bundles no scripts.

Good for

Teams running LLM agents who want to know where the money goes
Developers debugging cost spikes, errors or slow workflows
Reviewers who want findings tied to trace ids

Install caveman-evidence-review

Claude Code

Copy this prompt.

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

npx @caveman-ai/cli@latest

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 22.13+.

This skill drives the caveman command line tool, so the skill file alone is not enough. The README lists it under the commands you use in your agent after the full install. Run caveman login before first use. Pick one route, not both.

From the project README. Source.

What's in the skill

When it loads

Read-only review of Caveman Cloud evidence: cost, Cave Score, workflows, traces, latency, errors, routing, savings. Use when asked what Caveman found or where LLM spend goes.

Sections in SKILL.md

Hard rules
Step 1 — Load context
Step 2 — Establish baseline
Step 3 — Test the leading explanation with traces
Step 4 — Inspect representative traces
Step 5 — Report
Caveman evidence review

Key facts

Author: juliusbrussee (GitHub)
Repository: juliusbrussee/caveman, folder skills/caveman-evidence-review
Installs on skills.sh: 111,260 (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: pass, Snyk: pass

What it includes

Read your Caveman project; stop if not logged in
Uses caveman_context, or the CLI commands to check login and projects. Stops and asks you to log in or pick a project.
Network · Step 1: Load context
Read overview, costs, score, workflows, verified savings
Reads those reports and the ranked daily headroom, and states the report window and basis first.
Network · Step 2: Establish baseline
Search traces in a bounded window with closed filters
Searches traces by workflow, model, error, cache status and similar filters, and compares a suspect group with a control.
Network · Step 3: Test with traces
Look at a few high-signal traces, metadata only
Reads span metadata, latency, status, token counts and the applied optimizers, with payload retrieval off.
Network · Step 4: Inspect traces
Fixed format with findings, unproven items, next check
Reports scope, measured cost, verified savings, inferred headroom, findings with trace ids, and one next read-only check.
Step 5: Report

From the official docs

Permissions

Low risk

It reads cost and trace data from your Caveman project and changes nothing. Make sure the right project is selected before you start.

What it can do

Read your data
Create and change things
Delete things
Read-only modeHow to turn it on

Watch out for

The install route changes your agent setup
The README says the full Caveman install finds your agents and logins, shows what it will change and waits for you to press Enter. The Claude Code plugin route auto-starts caveman in every interactive session. The README's Privacy section says the CLI sends usage stats by default (commands run, token counts, a random install ID, OS and IP, not prompts, code or file paths); `caveman telemetry off` or DO_NOT_TRACK=1 turns it off.
It reads your Caveman project data
It uses your Caveman login to read cost, trace and workflow data. Check that you have selected the right project.
Payloads are off by default
It does not fetch prompt, completion, tool or artifact payloads unless you explicitly ask for payload review.

Technical details

Turning on read-only	The SKILL.md says to act as a read-only operator and never start, approve, cancel or roll back an experiment.
Talks to	Caveman Cloud, through MCP tools or the caveman CLI

Questions

What does caveman-evidence-review do?

It reads your Caveman Cloud data and explains where LLM spend, errors or latency come from. It tests the leading explanation against traces and reports what is proven and what is not.

How do I install caveman-evidence-review?

Use the install box at the top of the page. The skill needs the caveman command line tool or its MCP tools and a login, so the box lists the full Caveman setup first. Then run caveman login.

Does it need an account?

Yes. It needs a Caveman login and a selected project. If either is missing, the SKILL.md says to stop and ask you to run caveman login or pick a project instead of guessing.

Can it change my experiments or settings?

No. The SKILL.md says it must never start, approve, cancel or roll back an experiment. It is a read-only review.

Is caveman-evidence-review safe?

It is built to be read-only and to skip prompt and completion payloads by default. It does use your Caveman login to read project data, so select the right project first.

How is it different from caveman-manage?

caveman-evidence-review reads cost, trace and savings evidence. caveman-manage reads the experiment lifecycle and blocks unsafe actions. The evidence review points to caveman-manage for lifecycle questions.

Information

updated Nov 2, 2026

Repository: juliusbrussee/caveman
Folder: skills/caveman-evidence-review

juliusbrussee/caveman skills.sh

Sources

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

Show all 3

caveman README · github.com
caveman-evidence-review SKILL.md · github.com
caveman-manage SKILL.md · github.com