Composer, details sidebar and CLI disagree about which model a chat runs next #618

Which model a chat will run on next has no single answer: the composer, the details sidebar, Settings and the CLI each show something different.

What was seen

• The composer's pick lives in the browser until something is sent. After picking claude-opus-5-5 on a new channel, ace info showed no model on the chat, ace ls still said "automatic" and the details panel had no Model row (^378).
• After a reload the composer showed claude-sonnet-5-5 while the details sidebar said "Model: claude-opus-5-5", "and nothing said which one the next run would use" (^543).
• Switching models silently resets the reasoning effort to Default. After switching to Automatic, details still read "Model claude-opus-5-5 · Low" from the previous run (^517).
• Automatic never says what it resolves to, in the picker or in ace model ("Automatic").
• Settings → General → Default model lists about 60 anthropic/ and openai/ models with no grouping and no mark for providers without a key, and the Dashboard's "Start a new channel" composer shows no model at all (^46).
• New channels start on Automatic rather than the model last used.

Steps

1. New channel → composer → pick a model, don't send. Compare the details sidebar, ace info <channel> and ace ls.
2. Pick Automatic and send.

Expected: the composer, details and CLI agree on what the next run uses, and Automatic shows the model it will pick.

Not filed: the CLI refusing Anthropic models ("The Anthropic API key is missing…") while the app ran Opus. The rig's hosts had the key only in their own environment (from .env), so CLI-started channel workers didn't.

Rig: web (loopback source host from origin/main, 0.0.53 code).

Map: ^378, ^543, ^517, ^46.

Found in the 2026-10-22 map study.