Every Play

Thirty-one published rote Plays on three stdlib-only Python cores. No pip install, no node, no keys, and no network in any core. Each one is a public archive you can read before you run it.

How to run one

Every line on this page is complete. Paste it, and rote fetches the published archive, shows you a consent screen listing every file the Play touches, and runs it. It needs rote, the free runner from Modiqo, and a free account. If you have neither, the registry installs both and the Play in one go. Swap the name at the end for any Play on this page:

No account, no install, nothing kept. comped also has a door that needs neither. It downloads about 40 KB of stdlib Python to a temporary directory, runs it, and deletes itself:

Prefer node? npx comped runs the same code. The full walkthrough is in the docs.

What your agents cost

Three Plays that read the transcripts your coding agents already write. Nothing is sent, nothing is guessed, and every number has a line showing the arithmetic behind it. Built on comped_core.

comped

What would this month have cost at list price, and how far ahead of your plan are you?

The card. List-price total for the window, the multiplier against the plan you actually pay for, your cache share, the delta since last time, the asks you keep repeating, and your rank on the board. It works out which AI you run from the model ids in the logs, so you type nothing.

Read the archive before you run it, or the full description.

session-ledger

One ledger, four harnesses, no double counting.

The primitive the other two are built on. Claude Code writes one line per content block, so roughly four in ten usage lines are streaming duplicates of the same API call. Codex writes cumulative counters. This collapses both into one deduplicated ledger, subagents included, and judges nothing.

Read the archive before you run it, or the full description.

wrong-turns

What does your agent keep getting wrong, and what did recovering cost?

Tool calls that errored, the messages where you corrected it, and reverts. Grouped into recurring classes, counted across sessions and days, priced in tokens, one redacted line of evidence each, and a drafted CLAUDE.md rule for every class that recurred three times. It never edits your rules file.

Read the archive before you run it, or the full description.

What your machine has been hoarding

Six read-only scans of files your machine already keeps. No network, no credentials, and every one runs cold on bundled fixtures first so you can watch it work before you point it at anything of yours. Built on daily_core.

tab-debt

How many tabs are open, and how long since the oldest one was looked at?

The tab strip shows neither a number nor a date, but the session files your browser writes hold both. Chrome and its family, Firefox, Safari and Arc, read directly: Chrome session files are a binary command log and Firefox stores its session in a compressed container, so both readers were written from the format up.

Read the archive before you run it, or the full description.

birthday-radar

Whose birthday is next, and how much of your address book has no date at all?

Your address book mentions a birthday on the morning of, which is the one moment the information is useless. This sorts by how soon, so the next one is a number of days. Contacts database, vCard exports and CSV, any one of which is enough.

Read the archive before you run it, or the full description.

app-graveyard

Which applications did you stop opening, and which are still Intel only?

macOS records the last time you opened every application and never shows you the list. This asks Spotlight, measures each bundle, and reads sixteen bytes into every executable to say which ones your Apple silicon Mac is still emulating.

Read the archive before you run it, or the full description.

vault-pulse

Which notes are load-bearing, which were written once, and is the daily habit alive?

A notes folder only grows, and nothing in the editor says which notes matter. The links give the graph and the timestamps give the habit. Orphans, broken links, and the daily-note streak. Obsidian is found automatically; any markdown folder works.

Read the archive before you run it, or the full description.

desktop-clutter

What is actually on the Desktop and in Downloads, by age and by size?

Both folders are append-only in practice, and the Finder sorts by name so the oldest file is invisible. Counted, aged, sized, duplicates found by hash, and graded A to F. On the machine this was built on it came back F with 164 duplicate groups.

Read the archive before you run it, or the full description.

receipt-ledger

What do the receipt files you already have add up to?

PDF invoices, saved confirmation pages, exported messages and plain text, four formats in one folder that nobody opens one at a time. The PDF reader decodes ToUnicode maps and rebuilds lines from the text matrix. Currencies are totalled separately and never summed across each other, and a document has to prove it is a receipt before it counts.

Read the archive before you run it, or the full description.

Ten seconds, many times a day

Twelve Plays small enough to run on reflex. Five of them remember what you told them in one append-only file under your home directory. None of them touch the network, and none of them take an out_dir, because these print. Built on micro_core.

whatis

What is that opaque string?

Paste it and it peels it. A base64 blob holding gzip holding JSON holding a JWT is one input and four layers, and you get all four. JWTs, base64, gzip, epochs, UUIDs, IPs, cron expressions and magic bytes.

Read the archive before you run it, or the full description.

fits

Will this fit the window, and what will it cost?

Point it at text or a file. Bytes, lines and words are facts and it states them. The token figure is an estimate, so it prints a range and the method that produced it rather than asserting a number the stdlib cannot know.

Read the archive before you run it, or the full description.

is-it-secret

What should you redact before pasting that?

Run it on anything about to leave your machine. It knows the literal shapes: AWS key ids, GitHub tokens including fine-grained ones, private key headers, connection strings. It hands back the same text with those parts replaced, and it never prints what it found.

Read the archive before you run it, or the full description.

cron-when

When does that cron expression actually fire?

The English, the next five fires in your zone and in UTC side by side, and the daylight saving trap. When both day fields are restricted, cron takes the union rather than the intersection, which is the thing most readers get wrong.

Read the archive before you run it, or the full description.

punch

How many times was the day broken, and what was the longest block you got?

One line saying what you are doing, which takes two seconds. Do it a few times a day and it answers what a calendar cannot: not where the time went, but how often it was cut.

Read the archive before you run it, or the full description.

spent

What went out today, and where does the month land?

One line in, a spend log that owes nothing to a bank, an app or an export. Amount, label, optional tag. Decimal arithmetic end to end, never a float.

Read the archive before you run it, or the full description.

jot

The thought, into the vault, in two seconds.

A thought arrives while you are doing something else. One line and it is appended to a markdown file in your vault, timestamped, and you are back to what you were doing. No app to open, no place to decide on.

Read the archive before you run it, or the full description.

streak

How long is the run, and which weekday do you keep dropping?

One word, and it keeps the only part of habit tracking that changes behaviour: the current run, the record, a grid of the last twenty-one days, and the day of the week you keep losing.

Read the archive before you run it, or the full description.

last-turn

What did the turn that just finished cost?

Not this month and not this project. That one turn, ninety seconds ago. Model, tokens in and out, cache share, dollars, and today's running total. It reads a 256 KB tail rather than your history, which is why it can run twenty times a day.

Read the archive before you run it, or the full description.

budget-left

How much of today's budget is gone, and how fast is it going?

You set a number you are willing to spend on agents today. This says how much is left, the burn rate, and whether you hit the cap before the day ends.

Read the archive before you run it, or the full description.

since-last

What did the agent actually touch?

Not what it said it did. What moved on disk since you last asked. Created, changed, deleted, and whether anything outside the repository moved. It watches the mtimes of the sensitive directories and can tell you something under one of them changed. It never opens them.

Read the archive before you run it, or the full description.

safe-to-commit

What is staged that should not enter history?

The last thing between a live credential and permanent git history is you, at the moment you type commit. It parses .git/index directly, with no subprocess: credentials, a tracked .env, leftover debugging, and files large enough to regret for the life of the repository.

Read the archive before you run it, or the full description.

What your git history knows

Four Plays over the repositories already on this machine. Git records more than anyone reads back: who wrote each surviving line, the committer's own UTC offset, and whether the code an agent wrote is still there. Built on daily_core.

bus-factor

Which files have exactly one author left, and has that person gone?

Every team knows some of its code has one author and nobody knows which files. Blame gives the surviving lines, the log gives the last time each author committed anything, and the join is the answer. Sole ownership is only a risk when the owner has stopped showing up, so the two facts are reported together rather than collapsed into one score.

Read the archive before you run it, or the full description.

night-shift

When were you actually working, in your own zone rather than this laptop's?

Every editor shows commit times in whatever zone the machine is set to today, which quietly rewrites your history each time you travel. Git stores the committer's own UTC offset inside the timestamp, so a commit made at 04:12 in Tokyo stays a 04:12 commit. This reads that offset instead of the local clock.

Read the archive before you run it, or the full description.

kept

Of everything the agent wrote for you, how much is still in the file?

Every other number about a coding agent is what it cost. This one is what it was worth. It joins the session transcripts your agents already keep against git blame to ask how much agent-written code survived. Git knows the answer and nobody asks it.

Read the archive before you run it, or the full description.

upstream-pulse

Which of your dependencies has nobody left maintaining it?

A vulnerability scanner tells you about the bugs somebody already found and wrote down. This tells you about the packages where nobody is looking any more: last release years ago, exactly one human who can publish, repository archived. It reads the lockfiles you have.

Read the archive before you run it, or the full description.

What your browser, disk and calendar hold

Six more read-only scans. Three of them, reply-debt, standing-cost and upstream-pulse, have a network half, and it lives in a separate fetch script outside the core for the same reason the leaderboard poster does: the core stays verifiably offline, and everything that opens a connection sits in one short file you can read. Built on daily_core.

extension-reach

What can your browser extensions see, and is anyone still shipping updates?

An extension shows its permissions once, in a dialog nobody re-reads, and then never again. The browser keeps the answer on disk for ever. This reads it and answers what the browser stops asking after install day: what can see your bank tab, and which of them was last updated years ago.

Read the archive before you run it, or the full description.

where-it-went

Where did the quarter actually go?

A browser shows the last nine things you opened. It will not tell you that one domain took a fifth of your quarter, that you opened the same question thirty-one times, or that your longest unbroken run at one site was four hours on a Tuesday afternoon.

Read the archive before you run it, or the full description.

photo-debt

What is your photo library actually made of?

A photo library only grows. The shutter fires ten times to get one usable frame, every screenshot lands next to the family album, and the same image arrives again from a message, an AirDrop and a download. The Photos app is organised by date and by face, so none of that is visible in it.

Read the archive before you run it, or the full description.

what-grew

Something ate 40 GB. What moved?

Every disk tool answers the wrong question: they show what is big, and what is big is mostly what was always big. The useful question is what changed, and nothing can answer that without having looked before. So the first run writes a baseline and says so, rather than reporting your whole home directory as new.

Read the archive before you run it, or the full description.

reply-debt

Who asked you something and is still waiting?

An unread badge counts messages. It does not know the difference between a newsletter and somebody who asked you a direct question eleven days ago. That difference is what this counts: threads where the last word is not yours, somebody asked for something, and nothing went back.

Read the archive before you run it, or the full description.

standing-cost

What does that recurring meeting cost, per year?

Nobody schedules a meeting by its price. A 30-minute standup with nine people is four and a half person-hours every time it fires, and over a year that is a number no calendar will ever show you. Every fact needed to compute it is already in the calendar.

Read the archive before you run it, or the full description.