← ClaudeAtlas

cost-analysislisted

Use this skill when the user asks for a cost audit, margin analysis, unit economics, "how much does this cost to run", "what's our margin", "are we losing money on X plan", "compare top-up packs", pricing health-check, or any variant of "is this priced right". Inventories every paid API call, SDK invocation, scraper, render provider, and managed service (Vercel, Resend, Supabase, Stripe, Replicate, Fal, Apify, Upstash, Anthropic, OpenAI, Gemini, etc.), pulls per-unit rates from canonical sources, then compares against the project's pricing model (subscription tiers, top-up packs, usage-based, free). Computes margins for redlining / average / power users, surfaces loss-making paths, and diffs against prior snapshots stored in memory. Optionally writes a COSTS.md to the repo. Also invocable as `/cost-analysis`.
kevincui1034/cost-analysis · ★ 0 · AI & Automation · score 75
Install: claude install-skill kevincui1034/cost-analysis
# Cost & Margin Analysis Inventory every $-generating call in the project, attach a per-unit cost, model the pricing structure, and produce **per-plan / per-user / per-feature** margin math with redlines flagged. Optionally persist the result as a dated memory snapshot and a checked-in `COSTS.md`. This skill is **stack-agnostic**. It detects the project shape first, then runs only the analyses that apply. If the repo has prior cost snapshots in auto-memory, it diffs against them instead of starting cold. ## How to use this skill Run the **workflow** in order. The discovery phases (1–3) feed the math (4–6); don't skip them or numbers will be fabricated rather than derived. The **catalog of paid surfaces** at the bottom is the reference for what to look for. --- ## Workflow ### Step 1 — Memory recall (do this first, before any file reads) Look for prior cost / pricing snapshots in auto-memory. Read in this order: 1. The `MEMORY.md` index for entries matching: `pricing-snapshot-*`, `cost-snapshot-*`, `pricing-redlines`, `pricing-sources`, `pricing-roadmap`, or anything tagged `type: project` referencing $ values. 2. Any matched memory files in full. If snapshots exist, this run becomes a **diff against the most recent one**. State that upfront in the report so the user knows you're not starting from scratch. If no snapshots exist, this is a cold audit — say so, and after the report ask the user whether to persist a baseline snapshot to memory. ### Step 2 — Stack & su