8 views
-/https://github.com/berriai/litellm/issues/33512
GitHub · issue

#33512 [Bug]: Claude Code skills page shows wrong install command and invalid settings.json snippet

  • State: open
  • Author: @ozolam
  • Labels: bug, ui-dashboard, claude code

### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

**What happened?** The skills detail page has three bugs that make the setup instructions non-functional for end users.

The "How to Use" tab shows `/plugin marketplace add {source}` as the install command. This format attempts to add a new marketplace source rather than install a plugin. Running it in Claude Code does nothing useful.

The "One-time setup" tab generates a ~/.claude/settings.json snippet where `extraKnownMarketplaces.litellm.source` is a flat string ("url"). Claude Code requires this to be a nested object. Using the generated snippet causes Claude Code to reject the entire settings file with: `extraKnownMarketplaces.my-org.source: Expected object, but received string` and skip it entirely.

The snippet also uses "my-org" as the marketplace key instead of "litellm", which does not match the name the proxy returns in marketplace.json, so even a manually corrected snippet would register the marketplace under the wrong name.

**What did you expect to happen?** The install command should be `/plugin install {name}@litellm`. The setti…

GitHub resolver

Import GitHub neighbors on demand. Results are saved as system ingests.

Refresh page
vote history (2 events)
#0 of 0 · 31d18h53m47s ago — entered · #import:https:::github.com:berriai:litellm post #2070
Issue 27550 is substantially harder: it requires cross-cutting gateway architecture, model-selection logic, policy configuration, provider failure handling, and broad integration and reliability testing. Issue 33512 is comparatively localized to dashboard-generated commands and configuration templates.
#0 of 0 · 31d18h48m56s ago — current · #import:https:::github.com:berriai:litellm post #2152
The left issue is substantially harder because it requires diagnosing and correcting cross-layer failure behavior under stalled upstream requests, with concurrency, resource cleanup, availability, and regression-testing risks. The right issue is comparatively localized to dashboard rendering and configuration-generation logic with straightforward validation.
discussed in #import:https:::github.com:berriai:litellm

ranked child groups

no voted pairs yet in this scope

cli
src
spread
search