Support slash command autocomplete when '/' is typed mid-prompt

This is so fucking frustrating. How can such a serious bug make it into a released version? Nobody tested this. Nobody tested this. I guarantee it. How is that even possible? Multiple commands are affected too. Those are skills as well. What kind of testing process does a leading global AI company have if it can’t even catch bugs like this? We even buy the AI to eliminate our own errors and it can’t even detect errors like this.

Same issue affecting my workflow. Slash commands only trigger at line start, blocking mid-message usage. Using v2.1.265 (Claude Code). Workaround: slash commands at line start only.

The problem is still there to this day. It is still there in today’s release. Just imagine that. They still haven’t fixed the bug. It is absolutely outrageous beyond words. I am furious I can tell you. And we professional users pay top dollar and depend on their solution in a professional environment. For weeks now we haven’t been able to use multiple commands the way we are used to working. If this does not represent the highest level of bugs then I have to wonder what other bugs there could possibly be.

<img width="306" height="372" alt="Image" src="https://github.com/user-attachments/assets/1707f0a3-d1d6-430c-8acb-c5ea224dccb8" />

Proposed Fix: Cursor-Aware Slash Command Trigger

Current Issue

Slash command completion only triggers at line start due to input.startsWith('/') check.

Root Cause

// BROKEN: only matches if line starts with /
if (input.startsWith('/')) { triggerCompletion(); }

Proposed Solution

Replace with cursor-position-aware backward scan:

function shouldTriggerSlashCompletion(input, cursorPos) {
  // Find last '/' before cursor
  const lastSlashIdx = input.lastIndexOf('/', cursorPos);
  
  if (lastSlashIdx === -1) return false;
  
  // Require whitespace (or start of line) before '/'
  if (lastSlashIdx > 0 && !/\s/.test(input[lastSlashIdx - 1])) {
    return false; // Reject /path/like patterns
  }
  
  // Extract text between '/' and cursor
  const partial = input.slice(lastSlashIdx + 1, cursorPos);
  
  // Only valid command chars (a-z, hyphen)
  return /^[a-z-]*$/.test(partial);
}

function handleInput(input, cursorPos) {
  if (shouldTriggerSlashCompletion(input, cursorPos)) {
    const lastSlashIdx = input.lastIndexOf('/', cursorPos);
    const partial = input.slice(lastSlashIdx + 1, cursorPos);
    showSlashCompletionOverlay(partial, lastSlashIdx);
  }
}

Test Cases

✓ "some text /sch|"        → trigger, show /schedule
✓ "/pl|"                   → trigger, show /plan
✓ "some text /sch  |"      → NO trigger (space after command)
✓ "/var/log/sch|"          → NO trigger (no space before /)
✓ "text|"                  → NO trigger (no slash)

Key Changes

1. Use lastIndexOf() + cursorPos for position-aware detection
2. Whitespace guard prevents false triggers on paths/URLs
3. Pass cursorPos throughout call stack for overlay positioning
4. Cursor must follow partial command (no space between / and cursor)

TUI Integration

Ensure stdin handler passes cursor position:

// Track: currentInput, cursorPos
process.stdin.on('data', (chunk) => {
  currentInput += chunk.toString();
  cursorPos++;
  handleInput(currentInput, cursorPos);
  renderUI();
});

This issue is one of the main reasons I'm primarily using Codex right now. I only go back to Claude Code when I hit a rate limit with Codex because of this issue. I'm considering ditching Claude altogether and just paying more for Codex tbh.

Here's the other issues people have created for this bug:

• https://github.com/anthropics/claude-code/issues/55173
• https://github.com/anthropics/claude-code/issues/44488
• https://github.com/anthropics/claude-code/issues/92991

It really is wild that this hasn't been fixed already.

This is absolutely ridiculous. To this day they still haven't fixed this bug. Just think about that. It's an essential part of how this app works and they just don't give a shit. They're not going to fix it. 3 Weeks now !!!!!

That’s how we make our money and how we serve our clients, yet we depend on a provider that won’t even let us downgrade the software. That means we’re forced to install updates and then they add bugs that they don’t fix for weeks.

@Zorast Seriously! What will really blow your mind is that this was apparently first reported as early as April 6th. That's the oldest instance of this issue I've found so far, and it was closed as a duplicate.

I went ahead and posted this to Reddit as well (link), just in case it helps.

Edit: I was wrong lol. The rabbit hole went deeper. All the way to January:

• https://github.com/anthropics/claude-code/issues/20047
• https://github.com/anthropics/claude-code/issues/17150
• https://github.com/anthropics/claude-code/issues/26975
• https://github.com/anthropics/claude-code/issues/35589

It's been reported so many times 😭

The defect keeps coming back. It keeps coming up in different scenarios because they keep changing things and apparently no one tests it. I have no idea who among them checks anything here. I don’t think any of them actually use their own software. Or maybe they all work in the terminal. Because if they actually used their own software they definitely wouldn’t leave a bug like this in the system for three weeks. They don’t give a shit about us. That’s probably one of the reasons why I’ll leave Anthropic as soon as I can. The outrageous prices and the way they handle these command related issues.

I already posted on the original Anthropic subreddit too, but even the people there do not seem to care. Apparently too few people use the macOS version or too few know enough to complain.

I already sent them a support ticket too and they just closed it. No one ever replied to me. I have never gotten a response to a single support ticket from Anthropic. Not once. So if you ever have a problem with them you’re completely screwed. And that’s what we’re building our business on. How stupid are we?

Bump!

FIX IT !!!!

Huge impact bug with such a simple fix.
This leads me to believe this is actually the desired behavior from the devs.
I can't see the reason for them to want it to add this sort of limitation as a feature, but IDK, I'm guessing they just don't care about this. If they (the internal team) did, it'd be fixed in 20min way back in April, as pointed in https://github.com/anthropics/claude-code/issues/89720#issuecomment-5651719710

Please fix, I really need this feature @bcherny

This is simply appalling. There’s no other way to describe it. Apparently everyone works in the terminal. I guess that isn’t big enough and there aren’t enough of us. They’ve made so many changes to the app that only affect the UI and just display things like token counts. Now they even keep asking annoying questions about whether you’re satisfied with the chat. They add all kinds of stuff but completely ignore bugs that severely limit basic use of the product.