VS Code

The companion VS Code extension embeds an xcsh conversation in an editor workspace. Its extension manifest and source repository define the supported commands and distribution channel.

What carries over from the terminal?

Provider configuration, model selection, session concepts, and workspace file boundaries follow the xcsh runtime. Open the intended workspace before starting a conversation.

Which behavior should I validate in the terminal?

Editor menus and command palette entries belong to the extension host and change with its release. Reproduce runtime behavior with the command-line interface (CLI) first:

1. Run the same prompt with xcsh -p in the workspace directory.
2. Compare the model, provider, and tool set reported there with the editor session.
3. Inspect the persisted session file rather than the editor transcript when an exact sequence matters.

A difference between the two points at extension configuration, not at the agent loop.

Need	Use VS Code?	Alternative
Work with editor selection and workspace context	Yes	The terminal cannot infer the active editor selection.
Reproduce provider, tool, or session behavior	Usually no	Reproduce with the xcsh CLI first.
Run unattended automation	No	Use a headless protocol or the SDK.

The extension inherits the VS Code workspace trust and process environment but still uses xcsh provider and tool boundaries. Confirm the workspace root and selected file before applying an edit. Re-run the same prompt in the terminal when isolating whether a failure belongs to xcsh or the editor host.