Feature request: /train for single corrections, and /feedback without session history

Two related ideas from a daily Claude Code user (Max plan).

1. A /train command for single corrections
When I notice the model got something wrong about current software, I'd like to type /train. The agent writes a short, machine-friendly correction (what the model believed, the correct answer, app and version, how it was confirmed). I review and approve it, and only that snippet is sent, never the whole chat.

• Why: software changes every release, and users find new menus and features first. Thousands of small verified corrections would keep models current.
• Example: Snagit 2026 for Mac. Claude said "Copy to Clipboard" was in Settings > Capture; it is actually in the Snagit menu-bar icon menu.
• Safeguards: agent and human review before anything is used for training.
• Fairness: no tokens charged to submit; credit (usage or tokens) for people who contribute regularly.

2. /feedback should allow sending no session history
/feedback asks "How much session history should we include?" with only: This session only / + last 24 hours / + last 7 days. There is no "None, just my message" option. Sessions often contain private information (health, family, business), so people cancel instead of sending useful feedback.

Claude Code 2.1.288, macOS (Apple Silicon).