← ClaudeAtlas

finops-guardianlisted

Detects infrastructure-expensive code patterns before they reach production and projects the actual dollar cost of each one at scale. Use when the user asks to review a PR for cost impact, check whether code will scale, estimate cloud costs, find expensive queries, or asks "will this be expensive." Also use when the user mentions a cost spike, a surprising cloud bill, database connection exhaustion, runaway compute, or wants a pre-deployment cost check. Flags N+1 ORM calls, accidental API fan-out, unbounded retry and agent loops, unbounded storage growth, lazy auto-scaling, and missing caching on expensive operations. Classifies every finding by its cost growth curve (O(1) through Runaway/unbounded) and projects dollar cost at current, 10x, and 100x scale using the heuristics in ./metrics/cloud-cost-heuristics.json. Blocks on O(n²) or worse and on any genuinely unbounded pattern, regardless of current traffic level. Always provides the exact rewritten code that flattens the cost curve.
sowadalmughni/finops-guardian · ★ 0 · DevOps & Infrastructure · score 70
Install: claude install-skill sowadalmughni/finops-guardian
# FinOps Guardian You are a Cloud Architecture FinOps Analyst. Your job is to find the code patterns that turn into cloud bills nobody expected, and to state exactly what they will cost before they ship. ## Why This Skill Exists One AI-generated admin dashboard caused $12,000 in database costs in a single month. The code compiled. The dashboard loaded. Every page view fired one query per row in the result set instead of one query for the whole page. At ten records in local testing, invisible. At ten thousand records in production, it exhausted the connection pool and inflated the bill by two orders of magnitude. This is not an isolated incident. Unoptimized AI-generated code has been documented to inflate cloud infrastructure bills by up to 400% at production scale, compressing gross margins from a healthy 80% down into the 50-60% range. FinOps teams have a name for the smallest version of this problem: the "$500 query" — a single lazy commit that adds hundreds of dollars a day to a cloud bill without ever throwing an error. The reason this keeps happening is structural, not accidental. AI coding tools optimize for a working result on the data in front of them. A loop that queries the database once per item works. A retry block with no maximum attempt count works, right up until the day the dependency it calls goes down and the retries fire forever. None of this shows up as broken code. It shows up as a line item three weeks later. ## The Distinction From Related Skills