skills.rest
leogurja
leogurjaCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

caveman-evidence-review

Reviews Caveman Cloud cost, trace, and savings evidence through read-only queries.

1|Updated May 3, 2021
Search Tags:#observability#savings#read-only#caveman#trace analysis#llm cost#cost review
One-click install
npx skills add https://github.com/leogurja/dotfiles --skill caveman-evidence-review-leogurja
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: caveman-evidence-review
Source: https://github.com/leogurja/dotfiles/tree/main/home/dot_agents/skills/caveman-evidence-review
Command: npx skills add https://github.com/leogurja/dotfiles --skill caveman-evidence-review-leogurja

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? It answers questions about where LLM spend goes and what Caveman Cloud found, using only measured evidence instead of guesses, while keeping cost, inferred headroom, and verified savings strictly separate. ## Core Features & Use Cases - Baseline reporting: Pulls overview, costs, Cave Score, workflows, and verified savings reports via MCP tools or the caveman CLI. - Trace investigation: Searches and inspects bounded trace cohorts by workflow, model, session, error code, or cost to test explanations against control groups. - Read-only safety: Never starts, approves, cancels, or rolls back experiments, and never fetches prompt or completion payloads unless explicitly requested. - Use Case: When asked why LLM costs spiked last week, the skill loads project context, compares cost reports across windows, isolates the driving workflow via trace search, and reports findings with cited trace ids. ## Quick Start Ask the assistant to review what Caveman found about LLM spend and errors for the current project over the past week.

Frequently Asked Questions about caveman-evidence-review

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

FAQPage Schema
How do I find where my LLM spend is going?▼

Load the Caveman project context, pull the costs and workflows reports, then use trace search grouped by workflow or model to identify the jobs driving spend. Compare the suspect cohort against a control window before drawing conclusions.

How do I investigate LLM errors and latency with traces?▼

Use caveman_trace_search with a bounded time window and closed filters such as error code, model, or latency bounds, then inspect a few high-signal trace ids with caveman_trace_get. Keep payload retrieval off and cite trace ids in findings.

Can this skill start or roll back Caveman experiments?▼

No. The skill is strictly read-only and never starts, approves, cancels, or rolls back experiments. It only proposes possible actions and directs lifecycle changes to a separate management workflow with a safety gate.

What is the difference between measured cost and verified savings?▼

Measured cost is the provider-complete list-price cost, while verified savings come from the savings ledger. Inferred daily headroom is a separate estimate, and the skill never adds or relabels these buckets.

What happens if Caveman login or project selection is missing?▼

The review stops and asks the user to run caveman login or select a project. It never guesses an organization id or proceeds without scoped project context.

Related Skills

Explore More Data & Analytics Skills→
ronnycodingronnycoding
20

connect-mcp-server

Connect MCP servers to Claude Code across HTTP, stdio, and SSE transports.

Community
Advanced
digital-stoic-orgdigital-stoic-org
19

openspec-develop

Automate OpenSpec change workflows by converting proposals into tasks and managing gates.

Official
Advanced
starwreckntxstarwreckntx

NetworkMonitorAgent (chatllm)

Detect anomalies in LLM chat traffic and analyze latency patterns.

Community
Basic
gke-labsgke-labs
42

gke-cost-analysis

Analyze GKE infrastructure costs from BigQuery billing export datasets.

Official
Intermediate