[ahrazzle]

What happened

A dispatch brief carried a model line copied from earlier in the session. By the time the launch was attempted, Frida's live config had already changed, so the brief named a model her profile no longer ran. She stopped rather than launch under a stale model.

What this adds

gates/check-brief-model.py reads model.provider and model.default from the named profile's live config.yaml and compares them against the brief's provider: and model: lines. On any mismatch, a missing field, or an unreadable config it exits 1 and refuses the launch. It remembers no model and edits neither the brief nor the config.

• exit 0: brief matches the live config
• exit 1: mismatch, missing field, or unreadable config
• exit 2: usage error

tests/test_brief_model.py covers the stale-model refusal, the live-model pass, and the missing-model-line refusal.

The checker is copied unchanged from the team-skills source. The test's script path now resolves gates/check-brief-model.py from the repo root instead of the author-local team-skills tree; the assertions are unchanged.

Verification

python3 -m pytest tests/test_brief_model.py -> 3 passed, exit 0. Full suite: 154 passed, exit 0.

[ahrazzle]

Requested changes

The script works, but it is not installed. It lands at gates/check-brief-model.py, while install.sh (PAYLOAD/TARGETS, lines 10-18) and protean-ingredient.json (payload/installTargets) only ship gates/protean-ops. I ran bash install.sh --target on this branch: the installed tree has no gates/check-brief-model.py.

The SKILL.md patch in protean-control-plane#10 tells the coordinator to run exactly that path from the protean-ops install, so the documented command fails on an installed kit.

Either move the script into gates/protean-ops/ (and update the SKILL.md command to match), or add the file to the ingredient payload and the install.sh arrays. The gate logic itself and the three tests are correct.

sparky-sai