NeuralCodeDOCS
Core workflow

The agent chat

The chat is where you talk to Claude Code inside a workspace. Output streams in live with tool-use chips, and the composer gives you model selection, extended thinking, slash commands and @mentions.

NeuralCode
2xlabs / fix-test-bug code-agents/fix-test-bug Ready to review Commit
Y
You
Add a section called ‘Cipher’ to the landing page. Leave it blank for now.
Claude Code
6 tool calls
Readapp/page.tsx
Editapp/page.tsx
Bashnpm run typecheck
Done. Added a blank section-shell with id="cipher"so it's anchor-linkable.
2m 22s
Ask to make changes, @mention files, run /commands
Opus 4.8 Thinking off
A workspace chat: streamed agent turns with tool chips and timing, and the composer with model + thinking controls.

The transcript#

The conversation renders three kinds of message:

MessageRendering
YouYour prompts, in a plain bubble marked “You”.
Claude CodeAgent turns, rendered as GitHub-flavored markdown (code blocks, tables, lists) with a brain avatar.
SystemOccasional system notes (e.g. the seeded task prompt).

A finished agent turn shows a collapsible “N tool calls” timeline and a duration footer like 2m 22s.

Tool-use chips

As the agent works, each tool it invokes appears as a chip — the tool name (Read, Edit, Bash, …) plus a short detail summarized from its input (a file path, a command, a search pattern, a URL). This lets you follow exactly what the agent is touching.

Live streaming

Runs use claude --output-format stream-json. NeuralCode parses the stream as it arrives, showing text and tool chips immediately, with a braille spinner reading Thinking… then Working…. When the run completes, the final summary, token usage and cost are recorded (see Usage).

The composer#

The input box at the bottom is where you drive the agent.

ControlWhat it does
Model pickerChoose the Claude model for new runs (Opus / Sonnet / Haiku). See Models.
Thinking toggleSet the extended-thinking budget: Off, High, XHigh, Max.
Add (+)Create a custom command or skill on the fly.
Send / StopThe arrow sends; while a run is active it becomes a Stop square that interrupts the agent.

Slash commands

Type / to open the command menu. It lists two groups:

  • Your commands & skills — your saved custom prompts. Selecting one expands its full prompt text into the box. Skills are tagged with a skill badge.
  • Claude Code — the built-in slash commands available in the worktree (discovered from the CLI). Selecting one inserts /name verbatim to pass straight through to Claude Code.

@mention files

Type @ to autocomplete files from the workspace. Pick one to insert its path, so you can point the agent at exact files.

Keyboard#

KeysAction
Send the message.
Insert a newline.
Move through the slash / @mention menu.
/ TabAccept the highlighted menu item.
EscClose the mention menu / clear the slash query.
Continuing a thread
After a workspace's first run, follow-up messages continue the same Claude Code session, so the agent keeps full context of everything it has already done in that worktree.