Agent Config Audit

See what your agent instructions cost before they become permanent context.

Paste a file. It scores itself as you type — no signup, and nothing leaves your browser.

Step 1: Add your instructions

Paste the file your coding agent reads on every turn — usually CLAUDE.md or AGENTS.md at the root of your repo. Nothing is uploaded; the analysis runs in this browser tab.

Public repos only. Your browser fetches the file from GitHub directly — its contents never reach tokencheat.

Optional MCP config

Paste .mcp.json or compatible Claude config JSON to include detected MCP servers in the audit.

Step 2: Read the verdict

Two different questions. The score says how much was found. The label says whether anything was found at all — so a high score can still read NEEDS WORK.

Add a file above and the verdict appears here.

Step 3: Decide where each finding goes

Not every problem is fixed the same way. Some are already corrected below, some belong in a skill that loads on demand, some are policy that belongs in a Cheat Code, and some need a human.

Each finding gets a destination once there is something to route.

Step 4: Take the rewritten file

Only provable changes are applied — duplicate lines, filler phrasing, whitespace. Anything the engine could not verify was left alone and flagged instead.

The rewritten file appears once the audit has run.

Need more than one file scored?

Same engine, same privacy: both run in your browser and upload nothing.

Beyond one file

This scores the file you have. A Cheat Code decides what should be in it.

The audit grades one instruction file as written. A Cheat Code works the other way round — you state your durable preferences, technology policy, and action boundaries once, and it compiles the files. Free, and it runs in your browser too.

Build your Cheat Code

Founding waitlist

Keep this analysis. The workspace saves it.

The paid workspace adds saved reports and score history, API and CLI access, and team governance — opening to the founding waitlist first, at founding pricing locked in before public launch. This tool stays free either way.

Join the waitlist