The carpool nobody could solve

What the agent made

Ran it for real with Claude Code 2.1.295 and OMP 18.8.6 on 2026-10-09, in an empty session with no saved memory. Your result will differ.

Open what Claude Code made, full screen

The same prompt in OMP made its own version: open what OMP made. Two runs, two different results: that is normal.

Make your own

1. Get the files

   Download the practice folder (ZIP)

   Unzip it: you get a folder called cli-practice with 1 fictional file. Download the fictional families.txt: six kids, four cars, two booster seats, and who can drive when.

   families.txt

2. Open your agent in that folder

   Open a terminal in cli-practice and type claude or omp --approval-mode always-ask. First time? Set up in about five minutes, with steps for Mac, Windows, and Linux.

3. Paste this prompt

   Read families.txt without changing it. It is a fictional school carpool. Find out whether any two-week schedule follows every rule.
   Write only these files: check.py, solve.py, why-impossible.md, fix-1.json, fix-2.json, and carpool.html. Use Python 3 standard library only.
   A schedule is JSON: {"change": "<one sentence or none>", "runs": [{"week": 1, "day": "Mon", "run": "morning", "cars": [{"family": "Abara", "kids": ["Ada Abara"]}]}]} with every run of both weeks listed.
   1. check.py <schedule.json> [--fix 1|2] checks every rule on its own, independently of solve.py, prints each broken rule with its week, day, and run, and exits nonzero if any rule is broken. With no --fix it checks the rules exactly as written; --fix N applies only the one-fact change of fix N.
   2. solve.py searches every possible schedule. If none follows the rules, prove it in why-impossible.md: name the exact rules that clash and show why, in words and counts a parent could check by hand.
   3. Find the two smallest changes that each make a schedule possible. A change is smallest when it changes one fact: one seat, one driver's slot, or one fairness number. Save a passing schedule for each as fix-1.json and fix-2.json, and run check.py on each with its --fix.
   4. Build one self-contained carpool.html: why the rules as written fail, and a printable two-week calendar for each fix with every car, driver, and kid, plus each family's trip count.
   Report every check.py command and its exit status in your final reply. Work only inside this folder; never install packages, use the network, call APIs, or load external assets. Use no em dashes or en dashes.

   The prompt asks the agent to stay in this folder; it cannot enforce that. Read each file change and command before you approve it.

4. Check it yourself

   Open cli-practice/carpool.html in your web browser (double-click it), then:

   Human check: Read why-impossible.md and check its reason by hand against families.txt. Run python3 check.py fix-1.json --fix 1 and python3 check.py fix-2.json --fix 2 and see exit 0. Open carpool.html and count one family's trips on the calendar.

   When we checked the recording above: why-impossible.md: every run needs Abara or Delgado for the two boosters, 20 runs vs fairness cap 8+8=16, checked by hand against families.txt. check.py fix-1.json --fix 1 exit 0, fix-2.json --fix 2 exit 0, original rules exit 1. Independent checker: fix 1 (cap 10) trips 10/8/8/10 PASS; fix 2 (Chen booster) 8/8/8/8 PASS. carpool.html calendar counted, Brooks 8.

Do it for real

Write your real constraints as plain rules in a text file: who, which seats, which days, what is fair. Ask the agent for a checker first and a schedule second, and only trust a schedule the checker passes. Leave out addresses and kids' full names.

Why an agent, not a chat window

A chat reply will hand you a schedule that looks right. An agent writes a checker for every rule, searches every schedule, and shows you the checker's verdict on every schedule, so the answer is proven, not just plausible.