Agents write most of the code; the operator inspects, verifies, and occasionally fixes. Desk’s editor is built for that loop — it lives one rail click from the terminals and shares its language servers with the agents.

​
Inspect what an agent wrote

Open the file the agent mentioned (channel file links open it directly, or Ctrl+P fuzzy-finds it). The explorer is git-aware — modified files carry badges, changed directories show dots, and the gutter marks exactly which lines differ from HEAD. Language intelligence is live as you read:
LSP hover tooltip over a TypeScript symbol

Hovering a symbol: real language-server intelligence, LSP ready in the status bar, zero problems

Hover types, jump to definitions and references, and watch the Problems panel aggregate diagnostics for your open files. The status bar tracks the language server lifecycle — warming, indexing with progress, ready — and the same server instance answers your agents’ MCP queries, so you and the agents see one consistent view of the project.

​
Find anything

The search panel runs filename and ripgrep-backed content search across the workspace root, opening results at the matching line:
Workspace search results panel

Workspace content search: every hit for a symbol, one click from its line

​
Edit safely while agents run

Saving is mtime-guarded: if an agent (or anything else) changed the file on disk since you opened it, a conflict banner offers reload or keep-mine — nothing is silently overwritten in either direction. Autosave modes (off / after-delay / on-focus-change) are strictly weaker than manual save and never write over a conflict. Renaming a file with language servers running becomes a transactional refactor: Desk previews which references the server would update, shows every affected file, and applies the whole edit server-side as one unit with rollback.

​
Close the loop with git and notes

  • The file context menu stages, unstages, discards, opens the diff, and shows per-file history — and “Open diff” lands in the source-control rail.
  • Select output in any terminal and Create note captures it into the notes subsystem, auto-named from its content — hypotheses and evidence stay attached to the workspace.
The full feature reference is in IDE and LSP.