Skip to content
Skills Library
AI & agents · Planning

setup-models

/setup-models

Configure astack models and reasoning budgets by role. Use for /astack:setup-models, model selection, or budget changes.

Repository
astack ↗
Licence
MIT
Last repository change
05.11.2026

Assessment and security

100/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content35 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
1
Scripts
0
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 05.11.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What it does and when to use it

Configure astack models and reasoning budgets by role. Use for /astack:setup-models, model selection, or budget changes.

Original author description

Configure astack models and reasoning budgets by role. Use for /astack:setup-models, model selection, or budget changes.

How it works

  1. Accepted values
  2. 1. Detect available models
  3. 2. Read current choices
  4. 3. Choose the budget and roles
  5. 4. Validate choices
  6. 5. Write the rule
  7. 6. Confirm the result
  8. 7. Offer verification when useful
Original SKILL.md headings · 157 lines, 957 words
SKILL.md preview
---
name: setup-models
description: "Configure astack models and reasoning budgets by role. Use for /astack:setup-models, model selection, or budget changes."
---

# Set up models

Write the user rule `${CLAUDE_CONFIG_DIR:-~/.claude}/rules/astack-models.md`.
Claude Code loads it in every session.
Its role values override the astack skill defaults.

## Accepted values

- A Claude alias: `opus`, `sonnet`, `haiku`, or `fable`.
  The Agent call sets that `model`.
- `inherit-parent` or `auto`.
  The Agent call omits `model` and uses the parent chat model.
- An external seat: `codex`, `codex:<model>`, or `cursor-agent:<model>`.
  Follow `${CLAUDE_SKILL_DIR}/../arena/references/external-seats.md` for that CLI.
  Bare `codex` uses its configured default model.

External seats are valid only for `arena runners`, `architect runners`, `interrogate reviewers`, and `arena cross-judge pool`.
Every other role runs as a Claude subagent with the session's tools.
The `why` and `reflect` roles need MCP access.
Code roles need the astack agents.

## 1. Detect available models

Read the Agent tool's `model` enum for the Claude models available in this session.
Use that enum as the source.

Check each external CLI and record it only after a working run:

- Run `command -v codex`.
  Check login with `codex exec -s read-only --skip-git-repo-check "Reply OK"`.
  Read the default from the `model` line in `~/.codex/config.toml`.
- Run `command -v cursor-agent`.
  Check login with `cursor-agent status` and list models with `cursor-agent models`.

If detection cannot find a requested model, ask the user for its name.
Never write an external value without a successful run.
`inherit-parent` and `auto` are always valid.

## 2. Read current choices

Start from the default role map in step 5.
If the rule already exists, read its `# budget` line and role values as the current choices.
Drop a role absent from the step 5 map, such as retired `how critics`.
Mark a value outside the accepted forms as needing a choice.
A model slug from another tool is not a Claude alias.

## 3. Choose the budget and roles

### Ask for the budget

Prefer AskUserQuestion over free text.
Show the current budget if the rule records one.
Without a rule, explain that `large` matches the skill defaults.
Offer these exact labels:

- `unlimited — max reasoning`
- `large — xhigh reasoning`
- `medium — high reasoning`
- `small — medium reasoning`

### Apply the budget

Build the working table from skill defaults.
On a later run, keep role overrides, including values, lists, `inherit-parent`, and `auto`.
The budget sets target effort to `max`, `xhigh`, `high`, or `medium`.

Claude and `codex` values contain no effort token, so their values stay unchanged.
Pass target effort through the Agent call's `effort` for a Claude seat.
Pass `-c model_reasoning_effort=<level>` for a codex seat.

For a `cursor-agent:<model>` slug with an effort suffix, replace that token with the target.
The token is last, or immediately before a trailing `fast`.
The ladder is `max` > `xhigh` > `high` > `medium` > `low`.
If the result is not listed, use that family's highest listed effort at or below the target.
If none fits, mark the role as needing a choice.

If step 1 found no working codex, replace each `codex` default with `sonnet`.

### Show the role map

Show every role and its value.
Mark values outside the detected set as needing a choice.
List every role that step 2 dropped.

Ask whether to accept the map or change specific roles.
Offer detected values, `inherit-parent`, and `auto`.
Explain that both aliases use the parent chat model.
Prefer AskUserQuestion.
It accepts at most four options and adds Other for the rest.

Panel roles use lists.
Each entry starts one seat, including alias entries, so list length sets the count.
`arena cross-judge pool` is a list too.
Arena chooses one entry from a different family than the parent when possible.
For a Claude parent, choose an available external seat.
`swarm workers` sets worker defaults unless a race or comparison assigns a model to an arm.

## 4. Validate choices

Require each Claude value to appear in the Agent tool's `model` enum.
Require each external value to have passed step 1.
`inherit-parent` and `auto` always pass.
If a choice is unavailable, stop and ask again.

## 5. Write the rule

Write plain Markdown without frontmatter so the rule loads in every session.
Include the selected `# budget` label and target effort.
Use the mode's exact role labels below.
Replace the whole file on each run so repeated runs stay idempotent.

```
# pstack model configuration. astack skills read these lines; nothing else does.
# One line per role. Delete a line to fall back to the skill default.
# `inherit-parent` or `auto` as a value: the role runs on the parent chat model (omit Agent `model`). Alias entries in a panel list still count toward its fan-out.
# External seats (codex, codex:<model>, cursor-agent:<model>) are valid only on the panel lines and the cross-judge pool.
# budget: large (xhigh)
feature, refactoring: sonnet
bug-fix: sonnet
perf-issue: sonnet
hillclimb: sonnet
judgment and prose: opus
hardest tasks: opus
how explorer: haiku
how explainer: opus
why investigators: sonnet
why synthesizer: opus
reflect tooling: sonnet
reflect judgment, divergent, synthesizer: opus
arena runners: opus, codex
arena cross-judge pool: codex, opus
swarm workers: sonnet
architect runners: opus, codex
interrogate reviewers: opus, codex
```

## 6. Confirm the result

Report that the rule was written and applies to new sessions.
A later run of this skill updates it.

If `CLAUDE_CODE_EFFORT_LEVEL` is set, Claude Code runs all subagents at that level.
The Agent tool then has no `effort` parameter.
The budget affects external seats only.
Report that limit and name the variable.
`echo $CLAUDE_CODE_EFFORT_LEVEL` shows its value.

## 7. Offer verification when useful

Check for a `.claude/skills/verify-*` skill or an existing real-app driver.
If none exists, offer `/astack:create-verification-skill` once.
Explain that it creates a local skill to drive the app as a user and prove changes.

On acceptance, read `${CLAUDE_SKILL_DIR}/../create-verification-skill/SKILL.md` and follow it.
On refusal, continue without another offer.

Versions

The database has no other skills named setup-models.

Database snapshot: 13.11.2026 · package version d12d1b7d748e (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100setup-pstack
    Configure which models pstack uses per role and at what reasoning budget. Offers every agent Orca orchestration can launch, not only the CLIs on this machine, then writes one agent, model, and effort per role. Use for /setup-pstack, "configure pstack models", "pstack budget", or…
  • 100setup-pstack
    Configure which models pstack uses per role and at what reasoning budget. Detects your available models and writes an always-applied rule that overrides the skill defaults. Use for /setup-pstack, "configure pstack models", "pstack budget", or changing pstack's model choices.
  • 100setup-pstack
    Configure which models pstack uses per role and at what reasoning budget. Detects your available models and writes an always-applied rule that overrides the skill defaults. Use for /setup-pstack, "configure pstack models", "pstack budget", or changing pstack's model choices.
  • 100setup-pstack
    Configure which models pstack uses per role and at what reasoning budget. Detects your available models and writes an always-applied rule that overrides the skill defaults. Use for /setup-pstack, "configure pstack models", "pstack budget", or changing pstack's model choices.
  • 100setup-pstack
    Configure which models pstack uses per role and at what reasoning budget. Detects your available models and writes an always-applied rule that overrides the skill defaults. Use for /setup-pstack, "configure pstack models", "pstack budget", or changing pstack's model choices.
  • 100setup-pstack
    Configure which T3 provider and model pstack uses per role and at what reasoning budget. Detects the models T3 can run and writes the model sheet that overrides the skill defaults. Use for /setup-pstack, "configure pstack models", "pstack budget", or changing pstack's model choices.

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.