Desktop app (2.26454.0): Code sessions refuse to delete Slack messages even when the user explicitly asks

What happened

Since the Claude desktop app updated on Oct 6 (now 2.26454.0), Claude Code sessions in the Code tab refuse to delete Slack messages, even when I explicitly tell them to. The session cites a built-in rule against permanently deleting messages, says it applies even when the user asks, and says nothing in CLAUDE.md, memory or project settings can change it. Up until the update, the same sessions did this for me routinely: dozens of confirmed deletes in the two weeks before Oct 6 alone.

These are messages posted by my own Slack bots, in my own workspace, using chat.delete scopes I granted to those bots on purpose so they could clean up after themselves.

Why this breaks real work

Deleting bot output is not an edge case when you build and run Slack bots. It's part of the normal loop:

• Testing. Every new card layout, alert or workflow gets posted to a test channel, checked, and removed. Without deletes, every test run leaves debris that someone has to clean up by hand, and test channels fill up with junk until they're useless for the next test.
• Fixing mistakes. When a bot posts something wrong (a duplicate card, a malformed message, a card for a ticket that was merged or deleted), the fix is to remove it. Leaving wrong information up in a team channel is worse than deleting it.
• Only the bot can delete its own posts. In Slack, a user can't delete a message another app posted. So when Claude refuses, I can't just click delete myself either. The bot's token is the only thing that can remove it, and Claude is the thing I use to drive the bot's tooling.
• Channel cleanup and migrations. Retiring a channel's old cards, clearing stale status posts, or redoing a set of posts after a design change all need bulk deletes of the bot's own messages.
• Undoing Claude's own output. If Claude posts a test message for me, I need it to be able to remove that same message. Right now it can post but not clean up, which is the worst combination.

Why the current rule doesn't fit

A blanket "never permanently delete messages" rule makes sense for someone's email inbox or a person's DMs, where a deletion could destroy something irreplaceable that belongs to someone else. It doesn't make sense for a developer's own bot output in their own workspace, deleted at their explicit instruction, using permissions they deliberately granted. The session even agreed it was routine and technically possible, and still couldn't do it.

It also behaves inconsistently. Sessions that started before the update could still delete, which made it look like something on my side broke.

What would fix it

Any of these would work:

• A setting (per user or per project) that allows message deletion when the user explicitly asks.
• Treat a delete the user explicitly requested like any other tool action that goes through the normal permission prompt, instead of an absolute ban.
• At minimum, allow deleting messages posted by the user's own bots or apps, or messages Claude itself posted in the same session.

Environment

• Claude desktop app 2.26454.0, macOS, Code tab
• Started after the Oct 6 app updates; worked before them