cmd/codeaf: `codeaf --model <slug>` is refused, and the manual's flags table does not say the flags follow `chat`

[AbirAbbas]

What happened

2026-09-15, the published build staging-20260915-4b6ec83cfbfa (dev@4b6ec83cf), linux/amd64:

$ codeaf --model deepseek/deepseek-v4-flash
error: there is no `codeaf --model`.
run `codeaf --help` for every command

codeaf chat --model deepseek/deepseek-v4-flash works. The refusal is well worded; the defect is that internal/manual/chat/starting-codeaf.md presents codeaf as the home screen and then a table titled "The flags you can start it with" (--model, --yolo, --once, …) without saying the flags belong after chat, so a reader of that page types the refused form first.

Replication

Deterministic (no model). bin/codeaf --model x prints the refusal above; bin/codeaf chat --help lists the flags. Read the manual page's flags section and note it never spells codeaf chat --model.

Field (real models). Not needed.

Where

cmd/codeaf (search there is no \codeaf) and internal/manual/chat/starting-codeaf.md(searchThe flags you can start it with`).

The fix

Either bare codeaf accepts the chat flags (it already opens the chat), or the manual page spells every example as codeaf chat --model … and says in one sentence that the flags follow chat. Pick one; the first is what a person expects from a command whose bare form is the chat.

Acceptance

• e2e: bin/codeaf --model deepseek/deepseek-v4-flash in a workspace opens the chat with that model on the chip, exit 0 on /quit — or, if the manual road is chosen, the tmux suite's needle for the flags table quotes codeaf chat --model.
• Unit: the manual probe "how do I pick a model when I start it" reaches the page that spells the working form.
• The manual page quotes the new wording and the change entry's invalidates names what people believed before.