[BUG] Slash command/skill menu does not open mid-input in the terminal (regression; #44447 closed without a fix)

## Summary

**Still broken in `2.1.275` (2026-09-18). Users have reported this since at least December 2025, 10 times, and a bot closed every report without a fix. See [Report history](#report-history).**

In the terminal CLI, the slash command / skill menu opens only when `/` is the first character of the input. If you type `/` after other text, no menu opens. You cannot find or pick a skill while you write a longer prompt.

This bug was reported in #44447 (terminal). That issue was closed for inactivity, not fixed. #89771 reports the same bug in the Desktop app and is still open.

## Environment

- Claude Code: `2.1.275`
- OS: Ubuntu on WSL2 (Windows 11 host)
- Surface: terminal CLI

## Steps to reproduce

1. Start `claude` in a terminal.
2. Type `/com`. The menu opens. (Expected.)
3. Clear the input. Type `fix the tests then /com`.
4. No menu opens. `/com` stays plain text.
5. Type a new line, then `/com` at the start of the second line. No menu opens.
6. Start the input with a skill, e.g. `/review then `, then type a second `/com`. No menu opens for the second one (see #26975).

## Expected behavior

Typing `/` after whitespace, anywhere in the input, opens the same menu with the same matching and ranking as at the start of the input. This must work for more than one `/skill` in one message. `@` mentions already work anywhere in the input.

Other agent CLIs (Codex CLI, Grok Build, Cursor agent CLI) support this.

## Why this matters

Prompts often chain steps, for example: `after you fix the lint errors, /commit with a clear message`, or `/plan the auth feature, then /test it`. The current workaround is to type the skill first, accept it, go back to the start of the line, and add the text before it. With many custom skills, you cannot remember the exact names without the menu.

## Design notes

A possible reason not to do this is conflict with file paths, URLs, and text such as `and/or` or `1/2`. Other CLIs avoid this with these rules:

- Open the menu only when `/` comes at the start of the input or after whitespace, and is followed by a letter.
- Close the menu when the token contains a second `/` (e.g. `/home/user/...`) or when no command matches.
- When the menu opened in the middle of the input, accept with Tab. Enter still sends the message, so a user who types a path and presses Enter does not run a command by accident.

## Question

When a `/skill` token is not at the start of the message, does Claude Code expand it the same way as a leading `/skill`? Or does the model only see it as plain text? If the behavior is different, please document it. Also, how are arguments passed to a skill that is not at the start of the message?

## Report history

Users reported this at least 10 times in about 9.5 months. A bot closed each issue as a duplicate or for inactivity. None of the reports shows a maintainer response.

| Opened | Issue | Result |
|---|---|---|
| 2025-12-04 | #13073 | Closed as duplicate the same day |
| 2026-01-09 | #17150 | Closed 2026-02-23 |
| 2026-01-22 | #20047 | Closed as duplicate of #17150. Says the CHANGELOG-listed behavior had stopped working |
| 2026-02-13 | #25477 | Closed as duplicate of #19107 (includes a crash on mid-sentence `/`) |
| 2026-02-19 | #26975 | Closed for inactivity 2026-03-20 |
| 2026-03-01 | #29752 | Closed as duplicate of #13073 |
| 2026-03-18 | #35589 | Closed as duplicate of #26975 |
| 2026-04-07 | #44447 | Closed for inactivity 2026-06-08 |
| 2026-05-01 | #55173 | Closed for inactivity 2026-06-20 |
| 2026-08-26 | #89771 | Open (Desktop app) |

Please do not close this issue as a duplicate of a closed issue. The earlier issues were closed without a fix, and the bug is still present in `2.1.275`.

No source files or tests are named. Start by reproducing the behavior in the terminal CLI on the stated WSL2 environment, then trace the slash-command menu handling and compare leading and mid-input tokens. Done means matching skills open after whitespace anywhere in the input, multiple skills work, and paths or unmatched tokens do not trigger the menu.

Written by the indexing model from the issue text.