`-p` + `--dangerously-skip-permissions` still emits interactive-style prompts in output

Command

claude --dangerously-skip-permissions -p '<prompt asking to edit a file>'

Expected

Non-interactive: model performs tool calls without approval, prints a final result, exits. No questions to the user (there's no next turn to answer them).

Actual

Output ends with a question directed at the user:

"want me to commit this to ~/.claude?"

Exit code 0. Unclear whether the edit actually landed, and no way to answer.

Impact

• Ambiguous completion state in scripted/headless flows.
• Model wastes tokens on rhetorical follow-ups that can never be answered.
• Users can't tell if the CLI blocked-then-timed-out vs. just printed a dead-end question.

Ask

In -p mode, either:
(a) system-prompt the model to never end with a question / never propose follow-up actions, or
(b) auto-execute the proposed follow-up when --dangerously-skip-permissions is set, or
(c) document that -p output is a monologue and questions are cosmetic.

Version

• claude --version: 2.1.207 (Claude Code)
• OS: macOS Darwin 25.5.0