Caveman Evidence Review
Lets an agent read-only audit your Caveman Cloud LLM cost, traces and savings data.
- Repository
- JuliusBrussee/caveman
- Stars
- 116,835
- Licence
- Apache-2.0
- Files
- 1 file
What it does
This skill teaches an AI agent to act as a read-only analyst for Caveman Cloud, a service that tracks LLM API spending and performance. When you ask what Caveman found or where your model spend goes, the agent pulls cost reports, scores, workflows, traces and savings figures, then reports findings with trace IDs and time windows. It keeps measured costs, estimated headroom and verified savings clearly separate, and it is told not to start, approve, cancel or roll back experiments or fetch raw prompt and completion content by default.
Great for
- Engineering teams using Caveman Cloud who want to understand LLM spend by workflow or model
- Operators investigating cost spikes, retry loops or latency issues using trace data
- People who want evidence-backed summaries rather than guesses about where money goes
Skip it if
- Teams that do not use Caveman Cloud, since the skill only works against that service's data
- Anyone needing to change experiments or routing, because the skill is deliberately read-only and defers actions to another skill (caveman-manage)
- Users who need prompt or completion content reviewed by default, since payload retrieval is off unless explicitly requested
Try asking
- “What did Caveman find about our LLM costs this week?”
- “Which workflows drove the most spend last month, and show me the top expensive traces.”
- “Compare model mix for the support-bot workflow against the previous two weeks.”
- “Are there retry loops in our agent sessions yesterday? Cite trace IDs.”
Install in 3 steps
- 1
Download the skill
The ZIP contains the 1 original files, the Apache-2.0 licence and an attribution note.
caveman-evidence-review.zip - 2
Add it to your agent
Claude Code (all projects)
mkdir -p ~/.claude/skills && unzip caveman-evidence-review.zip -d ~/.claude/skills/Claude Code / Agent SDK (this project only)
mkdir -p .claude/skills && unzip caveman-evidence-review.zip -d .claude/skills/Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.
- 3
Ask for the task
No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.
Get better results
- State the time window explicitly (for example, last 7 days) so the agent can scope its queries and cite exact ranges.
- Ask narrow questions; the skill reads only the reports needed to answer, which keeps queries bounded.
- Treat the output as a starting point: unproven explanations are flagged, and you should verify suspected causes with a control cohort or evaluation before acting.
What's inside
- SKILL.md
Redistributed unmodified from JuliusBrussee/caveman under the Apache-2.0 licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.