ScriptedAlchemy Before following this workflow, read the Codex runtime contract. It defines plugin paths, model configuration, delegation, and persistence.
Setup pstack for Codex
Write ~/.codex/pstack-models.md, the override file every delegating pstack skill reads before it spawns subagents. Missing roles keep the defaults below. This file is configuration for pstack, not a Codex model-provider configuration file.
1. Load current state
Read the existing override file when present. Otherwise start from the defaults in step 4. Determine which Codex models the current host exposes from the current tool metadata or official local model picker. Never invent a model slug. inherit-parent and auto are always valid and both mean to omit the model override.
2. Choose a budget
Use request_user_input when it is available. Otherwise ask one concise question. Offer:
unlimiteduses each role's listed effort.largecaps explicit role efforts atxhigh.mediumcaps them athigh.smallcaps them atmedium.
Preserve the current choice on a rerun unless the user changes it. A model must support its selected effort. Fall back to the nearest supported effort at or below the requested cap.
3. Confirm role assignments
Show every role, model, and reasoning effort. Allow inherit-parent or auto for any role. Panel roles are ordered lists; one subagent runs per list entry. List length controls panel size. Keep a user's existing override unless it is unavailable.
4. Write the override
Overwrite ~/.codex/pstack-models.md atomically so reruns are idempotent. Use this shape:
# pstack model configuration for Codex. Delete a role line to use its skill default.
# `inherit-parent` or `auto` omits explicit model and reasoning overrides.
# budget: unlimited
feature, refactoring: gpt-5.6-luna @ medium
bug-fix: gpt-5.6-luna @ high
perf-issue: gpt-5.6-luna @ high
hillclimb: gpt-5.6-luna @ high
judgment and prose: gpt-6-astra @ high
hardest tasks: gpt-6-astra @ xhigh
how explorer: gpt-5.6-terra @ medium
how explainer: gpt-6-astra @ high
why investigators: gpt-5.6-terra @ medium
why synthesizer: gpt-6-astra @ high
reflect tooling: gpt-5.6-sol @ high
reflect judgment, divergent, synthesizer: gpt-6-astra @ high
arena runners: gpt-6-astra @ high, gpt-5.6-sol @ high, gpt-5.6-terra @ medium, gpt-5.6-luna @ medium
arena cross-judge pool: gpt-6-astra @ high, gpt-5.6-sol @ high, gpt-5.6-terra @ medium, gpt-5.6-luna @ medium
swarm workers: gpt-5.6-luna @ medium
architect runners: gpt-6-astra @ high, gpt-5.6-sol @ high, gpt-5.6-terra @ medium, gpt-5.6-luna @ medium
interrogate reviewers: gpt-6-astra @ high, gpt-5.6-sol @ high, gpt-5.6-terra @ medium, gpt-5.6-luna @ medium
Apply the budget cap to the effort tokens before writing. Validate every real model against the host's available-model list. Stop for a replacement choice if a requested slug is unavailable.
5. Install or refresh custom agents
Locate this installed plugin's root. Copy codex-agents/poteto-agent.toml and codex-agents/comment-sicko.toml into ~/.codex/agents/, replacing only those two pstack-owned files. Do not alter unrelated agents. Tell the user that new Codex sessions pick them up.
6. Confirm
Report the chosen budget, overrides, config path, and installed agent files. Re-running this skill updates the same files.
Optionally offer to create a project-local verification skill when the repository has no real-surface test harness.
If the user wants poteto-mode to persist across tasks in this repository, use the optional scoped AGENTS.md method. Do not alter project instructions unless that persistence was requested.