setup-pstack

Setup pstack

Write ~/.config/pstack/pstack-models.md, the file that sets pstack's model per role. Skills read it when they delegate.

Steps

1. Detect available models

Call orchestrator_capabilities and keep the providers whose canRunChildTask is true. Their models are what you can pass to a delegate_task subagent in this session. That is the only source. If it returns no eligible model, say so and offer inherit-parent for every role. An entry is <providerInstanceId>/<model>, plus (<effort>) when the model lists an effort option (effort on Claude, reasoningEffort on Codex). Never write an entry it did not return. The aliases inherit-parent and auto are always valid even though they are not detected models.

2. Load current state

The default role-to-model mapping is the rule in step 3(b), over the roles shown in step 5. If ~/.config/pstack/pstack-models.md already exists, read it and treat its # budget line and its role values as the current choices. Otherwise start from those defaults. A line whose role is not in step 5, such as how critics, is from a retired role. Drop it.

3. Budget, map, and confirm

(a) Ask for a budget. Ask plainly in the thread. Offer these four options with these exact labels, and name the current budget when the file records one.

• unlimited - max reasoning
• large - xhigh reasoning
• medium - high reasoning
• small - medium reasoning

(b) Apply it. Build the working table from the skill defaults, and on a re-run keep any role you changed by model, list, or alias (inherit-parent, auto). By default, code roles (feature, refactoring, bug-fix, perf-issue, hillclimb) and the how explorer, why investigators, swarm workers, and reflect tooling roles get the fastest strong coding model detected. judgment and prose, hardest tasks, how explainer, why synthesizer, and reflect judgment, divergent, synthesizer get the most capable model detected. Each panel role (arena runners, arena cross-judge pool, architect runners, interrogate reviewers) gets the most capable model of each distinct provider detected, up to three entries. unlimited, large, medium, and small set the effort of every real entry, panel entries included, to max, xhigh, high, or medium, on the ladder max > xhigh > high > medium > low. If the model does not list the target, use the highest ladder option at or below it that the model lists, else mark the role as needing a choice. Never pick an option off the ladder, such as ultra, ultracode, or ultrathink. A model with no effort option gets no (<effort>). inherit-parent and auto do not change. So under large, a model that lists low, medium, high, and max gets high.

(c) Show the roles and confirm. Show every role with its model, marking any real entry not in the detected set as needing a choice. Also list each line step 2 dropped. Ask whether to accept as-is or change specific roles, offering the detected models plus inherit-parent and auto (both mean: this role runs on the parent thread's model) as the options. Ask plainly in the thread. For panel roles (arena runners, architect runners, interrogate reviewers) the value is a list, and one subagent runs per entry, alias entries included, so the list length sets the count. arena cross-judge pool is also a list, but Arena selects one value from it whose model family differs from the parent's when possible. swarm workers is the default model for every worker unless a race or comparison assigns another model per arm.

4. Validate

Every real entry written must be in the orchestrator_capabilities result. Its provider has canRunChildTask true, its model is listed under that provider, and its effort, when it has one, is one of that model's effort options. A model with no effort option has no (<effort>). inherit-parent and auto always pass. If a chosen real entry is not available, stop and ask again.

5. Write the file

Write ~/.config/pstack/pstack-models.md (create ~/.config/pstack/ if it is missing) as a plain file with the header comments, a # budget line with the chosen label and its target effort, and one line per role, using the same labels poteto-mode uses. Overwrite the whole file so re-runs stay idempotent. Shape:

# pstack model configuration. 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 thread's model (omit delegate_task `target`). Alias entries in a panel list still count toward its fan-out.
# budget: large (xhigh)
feature, refactoring: <providerInstanceId>/<model> (<effort>)
bug-fix: <entry>
perf-issue: <entry>
hillclimb: <entry>
judgment and prose: <entry>
hardest tasks: <entry>
how explorer: <entry>
how explainer: <entry>
why investigators: <entry>
why synthesizer: <entry>
reflect tooling: <entry>
reflect judgment, divergent, synthesizer: <entry>
arena runners: <entry>, <entry>, <entry>
arena cross-judge pool: <entry>, <entry>, <entry>
swarm workers: <entry>
architect runners: <entry>, <entry>, <entry>
interrogate reviewers: <entry>, <entry>, <entry>

6. Confirm

Tell the user the file was written and that it applies the next time a skill delegates, since skills read it then. Re-running this skill updates it.

7. Offer a verification skill (optional)

Check whether the project has a way to drive the real app for proof (a verify-* skill, or an existing harness). If not, offer once: "want a project-local verification skill, so agents can drive the app the way a user does and prove changes work? I can generate one with /create-verification-skill." On yes, read ../create-verification-skill/SKILL.md from this skill's directory (the one your harness gave when it loaded this file, else ~/.claude/skills/setup-pstack/ on Claude Code and ~/.agents/skills/setup-pstack/ on Codex) and follow it. On no, move on without pushing.

8. Offer a formal verification skill (optional)

Check whether the project already has a formal verification skill (a project skill in .agents/skills/ or .claude/skills/ that runs a model checker, a proof, or another verifier against the code). If not, offer once: "want a formal verification skill, so agents can prove a key property of your code holds in every case in scope, not only the cases tests reach? I can generate one with /create-formal-verification-skill." On yes, read ../create-formal-verification-skill/SKILL.md from this skill's directory and follow it. On no, move on without pushing.