setup-gpt56-model-router

name	setup-gpt56-model-router
description	Install, verify, or uninstall the eight schema-7 GPT-5.6 roles and report scoped v0.6 installation health.

Set up the GPT-5.6 model router

Python 3.9 or newer and stable enabled Codex multi_agent support are required.

python3 <skill-directory>/scripts/setup_router.py install --json
python3 <skill-directory>/scripts/setup_router.py check --json
python3 <skill-directory>/scripts/setup_router.py uninstall --json

check is read-only. It reports runtime capability, exact plugin installation/enabled state, app-server plugin/read discovery of both explicit skills, schema-7 role files, and pending legacy cleanup separately. Its overall result covers only those installation surfaces. It does not launch a fresh task or prove that a host has injected the skills into that task's startup catalog; record that as separate live acceptance evidence.

v0.6 is a clean break. It does not migrate old roles or payloads. Earlier role files are divergent; install --force backs them up before replacement. Uninstall removes only exact v0.6 roles unless forced.

Install performs one compatibility cleanup only: an intact router-owned v0.5 schema-3 depth marker/state pair is restored to its recorded prior value and removed transactionally. Edited, incomplete, or older state fails closed. v0.6 never creates, manages, or claims enforcement from agents.max_depth.

Role installation and legacy cleanup roll back together on failure. Backups remain outside agent discovery. Start a fresh Codex session after installation so role and skill discovery reload.

Subagents inherit active runtime permissions. Role sandbox_mode = "read-only" is behavioral guidance unless persisted evidence proves actual isolation.