finops-observability
Measures what the brain and its agents cost and routes the question to the right tool: spend per arm, routing KPIs, marginal cost of each capability. Triggers on 'cuanto gastamos', 'cuanto cuesta', 'que se esta comiendo los tokens'.
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 18
- Forks
- 6
01 前言 Front Matter
- name
- finops-observability
- description
- Measures what the brain and its agents cost and routes the question to the right tool: spend per arm, routing KPIs, marginal cost of each capability. Triggers on 'cuanto gastamos', 'cuanto cuesta', 'que se esta comiendo los tokens'.
02 资源文件 1
- 参考文档skill.json351 B
03 SKILL.md 正文
FinOps Observability
The brain already measures cost across several surfaces. This skill is the map so you pick the right one instead of rebuilding. All use list-price ESTIMATES from local session JSONL (attributed by repo path) — honest scope: an estimate with a small unattributed remainder, not a per-request invoice. On a subscription, "estimated $" is API-equivalent value, not what's billed.
Which tool for which question
| Question | Tool |
|---|---|
| Per-arm $ + routing KPI ($ saved vs all-Opus) + est-vs-billed | python3 scripts/finops-digest.py [--days N] [--json] |
| Did a NEW capability make me cheaper/more expensive? | python3 scripts/cost-vs-change.py [--days N] [--window W] — correlates daily cost curve with the git log of skills/scripts merged |
| Daily roll-up (skills/agents activity + the FinOps routing section) | python3 scripts/brain-digest.py --stdout |
| Usage by day/week/month/model/project | python3 skills/claude-usage-report/scripts/usage_report.py |
| Per-skill token ROI (which skill earns its tokens) | python3 scripts/skill-cost-profiler.py --days N --json |
| Reconcile estimate vs ACTUAL billed (needs Admin API key) | python3 scripts/anthropic-analytics-pull.py then re-run finops-digest |
| Enforce a per-arm cap (not just measure) | [[finops-budget-policy]] — set budgets.yaml (template: budgets.yaml.example) |
usage_report.py owns the canonical PRICING table — every other tool reuses its price(); never fork the rates.
The two levers (measure → act)
- Caps ([[finops-budget-policy]]) = the ceiling: a
PreToolUsehook refuses the expensive tool when an arm burns through its cap. - Routing ([[model-routing-by-complexity]]) = the floor: frontier brain, tiered arms (Haiku mechanical, Sonnet well-specified bulk, Opus build default, Fable all judgment); delegate each task to the cheapest sufficient tier. The
finops-digestrouting KPI ($ saved vs all-frontier) is what proves routing is actually happening; if spend is ~100% frontier-tier, routing is unused.
Honest reading
- Daily cost tracks how much you worked, not only new capabilities —
cost-vs-changesurfaces signals to investigate, not proof (correlation ≠ causation). - The blog/news pipeline runs on Groq + GH Actions + CF, NOT Claude — those costs are in OTHER meters this suite doesn't see yet. Cost-per-output for content needs Groq ingestion (not built).
- Make the digest recurring (it's folded into
brain-digest.py); a metric nobody reads doesn't grow.
04 同仓库更多 Skills
yeet
Use only when the user explicitly asks to stage, commit, push, and open a GitHub pull request in one flow using the GitHub CLI (`gh`).
wrangler
Cloudflare Workers CLI for deploying, developing, and managing Workers, KV, R2, D1, Vectorize, Hyperdrive, Workers AI, Containers, Queues, Workflows, Pipelines, and Secrets Store. Load before running wrangler commands to ensure correct syntax and best practices. Biases towards retrieval from Cloudflare docs over pre-trained knowledge.
workspace-skill-discovery
Discover and include ALL skills across the workspace — not just global ones. Use at the start of every session or when the user asks about available skills. Ensures project-level skills (under .claude/skills/ or skills/ in any repo) are never missed.
workers-best-practices
Reviews and authors Cloudflare Workers code against production best practices. Load when writing new Workers, reviewing Worker code, configuring wrangler.jsonc, or checking for common Workers anti-patterns (streaming, floating promises, global state, secrets, bindings, observability). Biases towards retrieval from Cloudflare docs over pre-trained knowledge.
winui-app
Create, build and design WinUI 3 desktop apps with C# and the Windows App SDK following Microsoft's official guidance and CommunityToolkit. Covers XAML, navigation, theming, accessibility and deployment.
windows-brain-script-portability
Scripts in ~/.claude/scripts/ that run on a Windows brain: use the python3.cmd shim and not bare python3, and reconfigure stdout to UTF-8 if you print symbols. Use it when writing a brain script or triaging a Windows-only crash.