claude-cooked
v0.1.0
Published
Get roasted for how you use Claude Code. Your Cooked score, what your tokens would cost at API prices, a roast from your own Claude, a shareable receipt and roast battles with friends. Runs locally; your prompts never leave your machine.
Maintainers
Readme
claude-cooked
Get roasted for how you use Claude Code.
npx claude-cookedIt reads your local Claude Code history, gives you a Cooked score out of 100, has your own Claude roast you with your real numbers, tells you how to cut your usage, and prints a receipt you can share.
claude-cooked 80/100 FULLY COOKED The Context Hoarder
"Pays $20 a month for $9,234 of compute and still types 'dude' at me 340 times."
THE ROAST
* Your first 4 messages to the most capable AI on Earth were "hey". It took you 4 tries to say hi.
* You've typed /quit 433 times. Claude has never once quit on you.
* 1,065 prompts after midnight, the latest at 4:58am. Your sleep schedule is "whenever the rate limit hits".
...What you get
- The roast. If Claude Code is installed, your own Claude writes it (about a cent of usage, no API key). If not, built-in roasts fill in your numbers. Every line quotes a real stat.
- The evidence. All-time tokens and what they would cost at API prices, month by month, model by model, your hours, streaks, night-owl share, favourite words, most-used commands, tools and lines written.
- The redemption arc. Fixes ranked by how much they'd save: when to start fresh, what loads before you type, which work can move to a cheaper model.
- A receipt. Opens in your browser with a Download PNG button.
- A battle code. Send it to a friend; they run
npx claude-cooked --vs <code>and Claude roasts you both.
Options
npx claude-cooked medium roast
npx claude-cooked mild|ramsay spice level
npx claude-cooked stats numbers and tips only, no roast
npx claude-cooked --vs <code> roast battle
--name <name> your name on the receipt and battle code
--no-claude built-in roasts only
--no-open don't open the browser
--json everything computed, as JSON (includes folder paths and your first prompt: keep it private)
--dir <path> Claude config folder (default $CLAUDE_CONFIG_DIR or ~/.claude)Privacy
Everything is computed on your machine. Your prompts and files are never uploaded.
When Claude Code is installed, the numbers on your card (counts, dollars, hours, a word from a fixed list like "dude", the country your timezone suggests, and the --name you pick, if any) are sent to your own Claude to write the roast. Your prompts, file names and project names never are. The receipt and battle code hold only those numbers, so they're safe to post.
Where the numbers come from
stats-cache.json: Claude Code's own all-time token totals and daily activityhistory.jsonl: every prompt you've typed, with timestampsprojects/**/*.jsonl: full session transcripts (Claude Code deletes these after 30 days by default)- your plan, from
.credentials.jsonor~/.claude.json
Most usage tools only read the transcripts, so they forget everything older than a month. This one reads all three.
"API price" is what your tokens would cost with an API key, using list prices (see PRICES in stats.js). On a subscription you don't pay that; it's a measure of how hard you're leaning on your plan.
The Cooked score
Five parts, 20 points each, computed from the card so battles are fair:
| Part | Measures | |---|---| | burn | API-price value per month | | night | share of prompts between midnight and 5am | | hoard | average context per request | | manners | please and thanks per 100 prompts (fewer = more cooked) | | chaos | /usage checks, interrupts, corrections, model-hopping |
Requires Node 18.17+. Zero dependencies. MIT.
