Claude Code is Anthropic's ACP-compatible coding agent. Once enabled, it becomes available as a selectable agent for new sessions just like any other ACP agent.
Once the status shows Running, Claude Code is available in the agent selector when creating a session.
Claude Code reads options from the _meta field that Braide attaches to the newSession request. The recommended starting configuration enables adaptive thinking with a summarized display:
{
"claudeCode": {
"options": {
"thinking": {
"type": "adaptive",
"display": "summarized"
}
}
}
}
type — one of:
"adaptive" (recommended) — Claude Code decides when to think based on the difficulty of the turn, rather than thinking on every prompt or never."enabled" — thinking is always on."disabled" — thinking is always off.display — one of:
"summarized" (recommended) — the thinking stream is condensed into a digestible summary rather than surfacing the full chain-of-thought. This keeps the session view tidy without hiding the agent's reasoning entirely."omitted" — the thinking stream is suppressed entirely and not shown in the session view.See Managing Agents > Additional Settings for the full editor behaviour and restart semantics, and Configuration > Agent Meta for where the JSON is persisted on disk.
Claude Code honours CLAUDE.md files within your project root and subdirectories. These files are a good place for per-project conventions — package manager choice, build commands, testing preferences, and architectural constraints. The agent loads them on session start, so changes take effect on the next new session.