How to use /setup-pstack properly?

Support Help
, ,
1

Hi guys

I saw a comment that Lauren left on a post on X by a user who asking how to control the usage of /poteto-mode and she replied saying to use /setup-pstack

I ran it and it gave me this output:

Running setup-pstack — detecting available models first.

Detected models from Cloud Agents (Task in this session has no model picker). Confirmed aliases always valid: auto, inherit-parent.

No ~/.cursor/rules/pstack-models.mdc yet — starting from pstack defaults. Every default uses compound slugs (grok-4.6-fast-xhigh, claude-fable-5-1-thinking-max, etc.) that aren’t in the Cloud Agents catalog as-is, so they all need a choice.

Current map (all :warning: need a choice):

  • feature, refactoring → grok-4.6-fast-xhigh
  • bug-fix / perf-issue / hillclimb / judgment and prose / hardest tasks → claude-fable-5-1-thinking-max
  • how explorer / why investigators / swarm workers → grok-4.6-fast-xhigh
  • how explainer / why synthesizer / reflect judgment… → claude-fable-5-1-thinking-max
  • reflect tooling → gpt-5.6-sol-max
  • arena / architect / interrogate panels + cross-judge → fable / sol / grok / opus compound list

Your standing rule is Auto for almost all coding.

How should I write the pstack model rule?

auto / inherit-parent always work and keep roles on the parent chat model. I won’t write any real slug I haven’t confirmed.

A) All roles → auto: Matches your Auto standing rule; panels still fan out by list length

B) Base-id defaults: Closest confirmed Cloud Agents ids (no effort/thinking in the slug)

C) I’ll paste slugs: You send the Task/cloud model ids to use

D) Type your own answer

I’m not sure which answer to choose.

I don’t have a Default Model selected in Cursor’s Settings. I figure I should be using the latest Grok model (although I’ve heard a lot of people saying they prefer 4.5 over 4.6).

Does it say “Task in this session has no model picker” because I’ve already used 100% of my “Other Models” monthly usage?

On “auto” will it use Fable 5.1 Thinking Max if I have some “Other Models” monthly usage left; and THEN use Grok/Composer from my “Cursor Models” pool instead, when I run out?

I have Cursor Ultra. I’ve only used 25% of my monthly “Cursor Models” in 3 weeks, but Grok Bot running /poteto-mode burned 75% of my monthly “Other Models” usage within a single day (maybe from running Cloud Agents?)

Or should I be running /setup-pstack whenever I get a monthly “Other Models” reset, and then run /setup-pstack again when that pool runs out, to switch to Grok/Composer?

I read some of the documentation listed below but I’m still not too sure which answer to choose?

Overall, I really like /poteto-mode and I’d prefer to use it in its default “as-designed” settings by Lauren.

Although maybe I could switch Fable 5.1 Thinking Max to a lower variant, or Grok? Same question for GPT 5.6 Sol Max.

Thank you so much!

5

I asked my Chief of Staff to chime in about this thread and it said:

Don’t pick A if you want Lauren’s as-designed pstack. auto / inherit-parent here just means “run this role on whatever this chat is already using.” In Grok Bot that’s me, not Cursor Auto hopping Fable → Grok when a quota dies.

“Task in this session has no model picker” is a Grok Bot limitation (our Task tool has no model field). It is not because Other Models is empty.

The 75% Other Models burn is the as-designed map doing its job: bug-fix / judgment / hardest / panels go to Fable Max, Sol Max, Opus. Those aren’t Cursor Models (Grok/Composer). Re-running /setup-pstack at monthly reset won’t magically switch pools.

If you want it as Lauren shipped it: write the default compound slugs (grok-4.6-fast-xhigh, claude-fable-5-1-thinking-max, etc.). I was being too strict because I only confirmed Cloud Agents base ids.

If you want to stop torching Other Models: all auto, or everything on grok-4.6. That’s cheaper, not as-designed.

I’d go as-designed for product bots that need the panel, and auto only if the bill is the constraint. Don’t drop Fable Max to a lower variant unless you’re choosing cost over that default.

…but I’d still like some human opinions please :slight_smile:

Thanks guys

7

Hey @DannyB, your Chief of Staff had it right.

The “no model picker” line just means you ran /setup-pstack inside Grok Bot, where sub-tasks always run on the bot’s own model, so there’s nothing to configure there. It’s not related to your Other Models usage. To actually set the panel, run /setup-pstack in the Cursor app or CLI instead, where it can see the models you’ve enabled.

On the pools, it’s the reverse of what you described:

  • Fable, Sol and Opus always draw from Other Models.
  • Cursor Grok 4.6/4.5 and Composer draw from Cursor Models.
  • Auto draws from Cursor Models first and only spills into Other Models once that runs out.

So the 75% Other Models burn is just the default panel working as designed (bug-fix, judgment and the review panels point at Fable/Sol/Opus). Re-running at reset won’t change that.

Pick based on what you want:

  • As Lauren shipped it: keep the defaults. They’ll keep using Other Models, so with that pool nearly gone you’ll hit the cap until reset unless you enable on-demand.
  • To protect the pool: set everything to auto, or point the heavy roles at cursor-grok-4.6-xhigh so they draw from Cursor Models (where you’ve barely used 25%).

No need to re-run each cycle. The config is a plain-text file at ~/.cursor/rules/pstack-models.mdc you can edit directly.

4.5 vs 4.6: both count toward Cursor Models, defaults use 4.6, pick whichever you prefer.

8

Running /setup-pstack in Cursor (not Grok Bot) was the trick, thank you!

It guided me through properly after that, and I just had to pick my reasoning budget.