A ranked list of 10 to 15 fixes, each with what to change, where, and what it should save per year at your current usage and list prices. Quick-win scripts for the top three fixes. A one-hour walkthrough call. One week of async questions after delivery. I spend 4 to 6 hours on it. You get the report within five business days after I receive your Token Triage output.

It is not Agent Stack Setup , and it is not that page's $250 Harness Audit. Agent Stack Setup installs Claude Code, Cursor, or Codex on your repo. Harness Audit reviews your repo and rules before an install. Standard includes a small burn view. This audit is the full read of what a team actually spends. It is not implementation. You or your team make the changes. If you want 60 days of async help after the audit, plus a routing and caching review, ask about the Premium audit at $4,500. It is not a SaaS or a dashboard. It is a report, written by me, from a tool you run.

Token Triage burn audit

A 4-6 hour audit of what your coding agents actually spend. $1,500. If the report does not identify at least $15,000 a year in savings, you pay nothing.

I am Ryan Doubrava. I audit Claude Code token spend for teams on API keys. Book a 20-minute discovery call so I can check the fit before you pay.

You run Token Triage, the free Apache-2.0 CLI, on your own machine. It reads the usage blocks your harness already logs and ranks the burn by project, session, and model. The tool sends nothing anywhere: no network code, no telemetry. You send me its report, not your logs. I read it and tell you what to change.

Receipts: the Token Triage release and 147 PRs across 11 repos in 39 days. Questions: ryan@cursedagentic.com.

Who is this for?

Teams that pay for coding agents by the token. Five to fifty engineers on API keys, spending $2,000 a month or more across Claude, OpenAI, or a mix. The usual trigger is a surprise bill, a first agentic engineer, or a CFO asking what all the AI spend is for.

Token Triage reads Claude Code logs today; other harnesses are out of scope until the tool parses them.

If you are one person on a flat Max or Pro plan, you are not the buyer. Run the free tool; it shows you your own numbers.

What do you get?

A ranked list of 10 to 15 fixes, each with what to change, where, and what it should save per year at your current usage and list prices.
Quick-win scripts for the top three fixes.
A one-hour walkthrough call.
One week of async questions after delivery.

I spend 4 to 6 hours on it. You get the report within five business days after I receive your Token Triage output.

What does it cost?

$1,500, fixed, paid up front. If the report does not identify at least $15,000 a year in savings, you pay nothing. If the report misses that threshold, I refund your payment in full.

Identified means the fixes in the report add up to that number at your current usage and list prices. I do not bill on what you later realize, and I never need ongoing access to check. That keeps it a one-time invoice.

What is it not?

It is not Agent Stack Setup, and it is not that page's $250 Harness Audit. Agent Stack Setup installs Claude Code, Cursor, or Codex on your repo. Harness Audit reviews your repo and rules before an install. Standard includes a small burn view. This audit is the full read of what a team actually spends.
It is not implementation. You or your team make the changes. If you want 60 days of async help after the audit, plus a routing and caching review, ask about the Premium audit at $4,500.
It is not a SaaS or a dashboard. It is a report, written by me, from a tool you run.

What leaves your machine?

Token Triage has no network code and no telemetry, and a CI check keeps it that way. What you send me is its markdown report: aggregate numbers (tokens, models, cache reads, costs) per project and per session. No prompts, no code, no conversation text. Mutual NDA on request.

How do I start?

Book a 20-minute discovery call. I check that the audit fits your spend before you pay. Or pay now: Pay $1,500 for the burn audit. Checkout runs through Doubrava Consulting. After payment, complete a short intake (team size, tools, monthly spend), run Token Triage locally, and send me its markdown report.