run wraps space-containing prompt arguments in literal quotes #1560

altimate-code run "<prompt with spaces>" sends the prompt to the model wrapped in literal double quotes, with an added trailing newline when stdin is not a TTY.

Cause: src/cli/cmd/run.ts:488 joins the message arguments and wraps any argument containing a space in "...". That quoting is needed for --command: command templates split arguments on quotes for $1, $2. Plain prompts go straight into a text part, so the quotes reach the model and every consumer of the user message.

Impact:

• The model receives "Fix the failing test" instead of Fix the failing test.
• Learn capture doesn't classify corrections sent this way. run --session <id> "No, use to_utc." reaches the classifier as "No, use to_utc.", so the leading No, cue doesn't match and no signal is recorded. Found while running an end-to-end check of feat(learn): attribute lesson recurrence from session corrections #1556 on a GCP VM: zero corrections were captured until the correction was passed as separate words.

Expected: plain prompts are joined with spaces and not quoted. --command keeps today's quoting so $1/$2 still split correctly.