Setup pstack

For a request scoped to PR templates, run step 0e only. A full setup runs all applicable steps below.

Write the pstack models config, a file that sets pstack's model and reasoning effort per role. Agents resolve it per role with this skill's resolver. A workspace .agents/pstack-models.md sits over the user's ~/.agents/pstack-models.md, a harness section beats a flat line in either file, and a role neither file names falls back to the shipped default in examples/pstack-models.md. So this is an override layer, not a requirement.

The inline defaults are written as short model aliases (fable, opus, sonnet, haiku), which translate to gpt-6 models on Codex and to grok-4.7 on Grok Build. Any other value your harness does not accept for subagents means inherit-parent: the role runs on the session model, and multi-model panels become same-model panels. An entry may pin a reasoning effort as model@effort, and ## codex, ## claude-code, ## grok, or ## hermes sections hold lines that apply to one harness only. The grammar, layering, and effort policy are defined once, in references/models-config.md, and scripts/check-models-config.py --resolve applies them.

Steps

0a. Check personas and optional native registration

Read the pstack-harness skill's registration reference. Resolve its logical installed skills root through portable-paths.md, then run:

python3 "${PSTACK_SKILLS_ROOT:?}/pstack-harness/scripts/subagents.py" check

Failure signal: nonzero exit. Repair missing payloads or sibling skills before reporting persona readiness. A ready payload lets a generic delegate receive the full persona without native registration.