Claude Code Engine

The Claude Code engine uses the Claude Agent SDK for complex, multi-step tasks. It provides richer output than the Built-in engine including file edits, file creates, and shell commands.

Requirements

Requires Claude Code CLI installed and authenticated. Conduit detects availability automatically and shows status in the engine selector.

Rich Output

The Claude Code engine produces structured output blocks, each with distinct visual styling:

• Text — Markdown-rendered prose
• Tool calls — Compact chips with live status
• File edits — Inline diffs showing changes
• File creates — New file contents with syntax highlighting
• Shell commands — Command and output display
• Approval requests — Inline action prompts

Approval Flow

Operations that require confirmation show inline approve/deny buttons in the chat. You control what the agent can do.

Session Persistence

Sessions are stored in SQLite and survive app restarts. Resume previous sessions from the conversation history panel. Stale sessions are automatically cleaned up.

Slash Commands

• /model — switch the active model
• /clear — reset context and start fresh
• /cost — view input and output token usage
• /help — list available commands

Edit & Retry

Edit previous user messages or regenerate assistant responses. Claude Code clears SDK session context and resends from the edit point.

Working Directory

Configure a default working directory in Settings for agent sessions. The agent operates within this directory context.