One command to install
In the Cursor chat box, type:
Installing from Customize in the sidebar works the same. Then pick your models:
It asks about your reasoning budget (how long models think), assigns a model to each model roles, and writes it to ~/.cursor/rules/pstack-models.mdc.
Commands on this site are written for the Claude Code plugin by default, with a pstack: prefix that Cursor doesn’t use. Pick Cursor here, and every command on the site switches to how Cursor writes it, ready to copy:
I use
Commands on this site are shown for this tool, so copying just works
Shown as: /pstack:poteto-mode
Pick /poteto-mode from the / menu
Type / in the chat box, select /poteto-mode from the menu, then write the task:
/poteto-mode add a --json flag to this command. The existing text output must stay byte-for-byte identical. Verify both outputs.Which key you send with makes a difference:
- Enter: it covers this one message only. A few turns later, it stops working.
- Option+Enter (Mac) or Alt+Enter (Windows): it becomes a Custom Mode and stays on every turn, until you leave the mode.
With Custom Mode on, “do it” or “continue” is enough for later messages.
When you finish one thing and move to the next, start with “new task.” Otherwise it treats your new question as the next step of the last one.
Hand it off before bed: /loop
/loop is a built-in Cursor command. It wakes the AI up now and then to check whether the work is done. Give it a done condition that passes or fails:
/poteto-mode I'm going to bed. In a new worktree (branched from <your base branch>), migrate every caller to the new parser.
Done means: no old callers left, all parser test data passes, and the old API is deleted.
Keep a decision log. No need to ask me before committing.
/loop until done. If you're truly stuck after a few hours, stop and write down why.“Work for 4 hours” isn’t a done condition. It has nothing to check, so you’ll wake up to 4 hours of busy work. For a fuller version, see Let it run.
Several AIs at once: give each its own space
Several AIs changing code in one folder overwrite each other. Two ways to keep them apart:
- In the cloud: type
/in-cloudbefore the task. Each AI gets its own machine and branch, and your computer is untouched. - Locally: give each task its own worktree (a second working folder for the same repo). On a laptop, a few at a time is enough.
/poteto-mode new task. Create a new worktree from <base branch> and port the parser changes over there.A few things to install separately
A few /poteto-mode steps use the cursor-team-kit plugin. It runs without it, but those steps do a weaker job. For the full set, install it. It includes:
deslop: cleans the “AI feel” out of code.control-cli: drives command-line programs.control-ui: drives browsers and web pages.
pstack has no planning skill. Use Cursor’s built-in Plan Mode alongside it.
Stuck?
/poteto-mode stops working after a few turns.
You probably sent it with Enter. Pick it from the menu again and press Option+Enter (Alt+Enter on Windows). Or type /poteto-mode again for each new task.
I changed the model settings, but it still runs the old models.
Settings only apply to new conversations. Start a new one and try again.
Several AIs worked at once and scrambled my files.
Give each task its own worktree, or hand it to the cloud with /in-cloud.
I ran /loop all night and nothing got finished.
The done condition has to pass or fail, like “this command returns 0” or “all these tests pass.” “Work until morning” doesn’t count.
I heard pstack comes with an automation pack called benny?
It does, off by default. It sorts issues reported in Slack and fixes the ones confirmed as bugs. To turn it on, point Cursor at FOR_AGENTS.md in the benny folder.