← ClaudeAtlas

anthropic-enterprise-analyticslisted

Pull Anthropic's Admin API usage_report into ~/.claude/analytics/ so brain-digest can reconcile estimated cost (from session JSONL list-price math) against actual billed cost (from Anthropic). Closes the estimated-vs-billed gap that any operator running on Pro/Max/Enterprise has.
CarlosCaPe/octorato · ★ 5 · AI & Automation · score 73
Install: claude install-skill CarlosCaPe/octorato
# Anthropic Enterprise Analytics — ingestion ## What this skill is The fourth and final block of the FinOps pipeline. The first three (per-arm rollup, cost-spike watchdog, budget caps) all run on the brain's *estimated* USD — list-price math applied to token counts read from native Claude Code session JSONLs. The estimate is good enough for budget caps and spike detection, but it is **not** what Anthropic actually bills the operator. The Admin API exposes the truth: per-user, per-API-key, per-day cost figures with Anthropic's actual rate card applied (including any enterprise discounts). This skill closes the gap. It pulls the Admin API daily, writes the data to a private directory, and brain-digest renders a reconciliation row alongside the per-arm estimate. ## When to use - You are on **Anthropic Enterprise / Team** with admin scope on the organization (the regular per-user API key won't work). - You want the brain digest to surface "estimated $X, billed $Y, delta Z%" so you catch any drift between the brain's list-price math and Anthropic's actual invoice — usually drift indicates either a model pricing change you haven't picked up in `_pricing.py` or a cache-read classification mismatch. - You want to bill clients against *actual* spend, not estimated. If you are on Pro/Max only, this skill is dormant — the script exits cleanly with "not configured" and the rest of the FinOps pipeline keeps running on estimates. ## How to set up ### 1. Mint an Admin API