skills.rest
HugoRCD
HugoRCDCOMMUNITY
Explore all skills by this author →
Website𝕏@hugorcdGitHub

cost-watchdog

Reviews weekly AI model spend and drift across surfaces using AI Gateway reports.

1.8k|61|Updated Jan 24, 2026
Search Tags:#drift detection#linear#lLM#model review#AI Gateway#Cost Analysis#spend report
One-click install
npx skills add https://github.com/HugoRCD/evlog --skill cost-watchdog
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: cost-watchdog
Source: https://github.com/HugoRCD/evlog/tree/main/apps/evi/agent/skills/cost-watchdog
Command: npx skills add https://github.com/HugoRCD/evlog --skill cost-watchdog

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Teams running LLM-powered agents lose track of model spend and silently drift into overpriced or outdated models. This Skill performs a recurring, data-grounded audit of model cost and quality so every surface runs a defensible model choice.

Core Features & Use Cases

  • Spend and token reporting: Pulls per-surface and per-model cost, token, and request data from the AI Gateway report for two 7-day windows.
  • Drift detection: Compares period-over-period cost, model mix, and token shape to flag meaningful changes with reasons.
  • Model landscape research: Checks current pricing and quality benchmarks from the AI Gateway catalog, leaderboards, and Artificial Analysis before recommending any swap.
  • Use Case: Every Monday, run the watchdog to produce a Linear document with per-surface spend, drift findings, and model recommendations, plus a Linear issue for any decision-worthy change.

Quick Start

Ask the agent to run the weekly cost-watchdog review of model spend and drift for the last full week.

Frequently Asked Questions about cost-watchdog

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

FAQPage Schema
How do I audit LLM model costs across multiple surfaces?▼

Query the AI Gateway report with groupBy tag to get per-surface cost, token, and request rows, then groupBy model for the model mix. Compare two 7-day windows to compute period-over-period drift per surface.

How to decide whether to swap an LLM for a cheaper model?▼

Judge swaps on quality and price together, not price alone. Pull per-token pricing from the AI Gateway model catalog and quality signals from leaderboards and Artificial Analysis, then recommend a swap only with a defensible cost or task-fit win.

Can the watchdog set different models per surface?▼

Not directly. The agent currently runs a single global model set by EVI_MODEL, so per-surface recommendations require either changing the global model or adding surface-scoped routing as a follow-up decision.

What data sources does a model cost review need?▼

It needs the AI Gateway usage report for spend and tokens, the gateway model catalog for pricing, and external benchmarks like arena.ai and Artificial Analysis for quality. Every cited figure needs a source and recency.

What happens when a weekly cost review finds no drift?▼

The report states that spend is flat and the models in use remain the sane choice, then stops. A quiet week is a valid result and no drift or swap should be invented.

Related Skills

Explore More Data & Analytics Skills→
truefoundrytruefoundry
13

truefoundry-ai-monitoring

Query AI Gateway spans API for traces, costs, latency, errors, and token usage.

Official
Advanced
DalomeveDalomeve
3

model-route-guard

Audit openclaw.json and agent overrides to resolve model routing conflicts.

Community
Intermediate
belokonmbelokonm
8

performance-monitor

Monitor token usage, latency, and eval-driven quality in AI agent workflows.

Community
Advanced
NagyViktNagyVikt
2

marketplace

Discover, install, and manage Vercel Marketplace integrations with environment variable provisioning.

Community
Advanced