0%
fewer tokens per run
About 690,000 tokens back on a single mid sized task.
Session limit reached
Four sessions, no finished work.
rewriting the prompts
Scroll
The difference
0%
fewer tokens per run
About 690,000 tokens back on a single mid sized task.
24→24
agent turns
Eleven round trips that never happen, so the session finishes.
$0
what Prompt me! costs
No subscription, no markup on tokens, no card.
The mechanism
Every turn resends the whole conversation. A run that takes twenty extra turns does not cost twenty extra messages, it re-pays for all the context it has already read, over and over.
The rewrite
>fix the login bug
Fix the 401 returned on valid credentials in src/auth/session.ts. Implement the change, do not only describe it.
Scope: token refresh only. Do not touch the signup flow or refactor unrelated auth helpers.
Done when: npm test passes and a valid login returns 200. Report the changed files and the test output.
How it goes
Type the request the way it is in your head. No structure required.
A short interview asks only the things that change the result, one at a time, with a recommended answer.
You get a prompt with paths, scope, and a verification step, plus what it will cost to run.
Write the prompt once, properly, and let the session finish inside its budget.
Start a promptYour first prompt runs on us. After that you connect your own API key and pay the model provider directly, at their price. Seven providers have a free tier, so that can be nothing at all.
Figures are estimates from the same cost model the app shows you on every result: predicted agent turns multiplied by the context resent each turn, priced at Claude Opus 5 list rates with cached reads. They are not measured benchmarks, and your run will differ.