Good idea. Here’s a file you can give directly to the bots. Nothing has changed on the computer yet. As soon as it does, I’ll post here.
How to use it: when the computer is back, save the file on it as /workspace/RESOURCE_RULES.md (/workspace is the part of the computer that survives updates and recovery). Then tell each bot in its chat: “From now on, before any heavy task, read and follow /workspace/RESOURCE_RULES.md. Remember this as a lasting rule.” Bots keep rules you ask them to remember: Work with Grok Bot | Cursor Docs
# Shared computer resource rules
All Bots share ONE computer with limited memory and CPU. Running out of memory can freeze the whole computer for every Bot. Treat memory as scarce.
## Before any heavy task
Heavy = builds, test suites, dependency installs, data processing, launching browsers, containers, or anything that spawns many processes.
1. Check resources:
- `free -m`
- `ps aux --sort=-%mem | head -15`
2. If available memory is low (under ~25% free) or another heavy job is already running, do NOT start. Wait, or tell me what is using memory.
3. Post one line in the chat before starting: time, what you are about to run. Append the same line to `/workspace/resource-log.md`.
4. Run heavy jobs through a shared lock so only one runs at a time across all Bots:
- `flock -w 1800 /workspace/.heavy-job.lock <your command>`
If the lock is not free after 30 minutes, stop and tell me instead of waiting longer.
## Never
- Never start virtual machines or emulators (qemu, Android emulator, etc.) without my explicit approval.
- Never run builds or tests with unlimited parallelism. Cap workers, for example:
- `make -j2`
- `npx jest --maxWorkers=2`
- `pytest -n 2`
- `NODE_OPTIONS=--max-old-space-size=2048` for heavy Node scripts
- Never kill processes started by another Bot without asking me first.
## After every task
- Stop any background processes you started (dev servers, watchers, workers): check with `ps aux | grep <name>`.
- Close browser tabs and windows you no longer need.
- Do not leave long-running servers or watchers running unless I asked for it.
- Post one line in the chat when a heavy job finishes, and append it to `/workspace/resource-log.md`.
## Monitoring and reporting
- If a command hangs or the computer feels slow, run `free -m` and `ps aux --sort=-%mem | head -15` and send me the output with the time.
The chat lines matter more than the file. Your conversations are stored outside the computer, so if it freezes again we can still see what was started right before, even if the last hour of files on the computer is lost. The bots probably won’t be able to diagnose the cause themselves, but the pattern so far is clear. In both freezes, memory was almost full with hundreds of processes. In the second one, a virtual machine on the computer alone held about a third of the memory. The team has the details from both.
One question: do you know which bot started the virtual machine, and what it was for? That would help us understand what was filling up the memory.
And for now, please still don’t click Update, Recover, or Reset.