claude-code - 💡(How to fix) Fix VSCode extension: "Unknown skill" errors from slash commands are not displayed in the chat UI [1 participants]

When typing an unknown slash command (e.g. /status) into the VSCode extension's chat input, the command is silently swallowed. No error is shown to the user. The error message Unknown skill: <name> does reach the model as a <local-command-caveat> block, but the user sees nothing and assumes the command ran (or was ignored intentionally). In the terminal CLI the same input produces a visible Unknown skill: <name> error in the output — so this is a UI regression specific to the VSCode extension. 4. Ask the model "did you see an error?" — it will report having received Unknown skill: status via a local-command-caveat The Unknown skill: <name> error should be rendered visibly in the chat transcript, matching terminal CLI behavior, so users know their command was rejected.

TL;DR

The VSCode extension should be updated to display the "Unknown skill: <name>" error message in the chat transcript when an unknown slash command is entered.

Guidance

• Investigate the difference in error handling between the terminal CLI and the VSCode extension to identify why the error message is not being displayed in the extension.
• Check the code that handles the <local-command-caveat> block in the VSCode extension to ensure it is properly rendering the error message.
• Verify that the error message is being received by the VSCode extension by logging or debugging the extension's code.
• Consider adding a try-catch block around the code that executes the slash command to catch and display any errors that occur.

Example

No code example is provided as the issue does not include specific code snippets.

Notes

The issue seems to be specific to the VSCode extension and not a problem with the underlying model or command handling. The fact that the error message is received by the model but not displayed in the extension suggests a UI or rendering issue.

Recommendation

Apply a workaround to display the error message in the chat transcript, such as modifying the code that handles the <local-command-caveat> block to render the error message. This will provide a better user experience until a full fix can be implemented.