Type / in the interactive TUI to open the command picker, or type a command name directly. Commands marked hidden are accessible by name but do not appear in /help.
Clear the transcript and conversation history. Keeps system prompt, todos, plan-mode, and skills cache.
/init
[--force]
Generate a CLAUDE.draft.md from available context sources (AGENTS.md, .cursorrules, .windsurfrules, README.md, git status). Always writes the draft and never touches CLAUDE.md; review and rename it yourself. --force suppresses the "CLAUDE.md already exists" notice.
/resume
[query]
List persisted sessions sorted by most-recently-updated, with an optional case-insensitive substring filter.
/recap
—
Summarize the conversation so far without mutating history.
/export
[path] [--format json]
Export the session transcript to a file. Default format: Markdown. Default filename: caliban-session-<date>.md in the CWD. Pass --format json for JSON output.
/btw
<question>
One-shot ephemeral question to a fast model (routed as FastClassifier); result inlined to transcript without touching the main session.
With no args: list the active provider's known model ids and the currently-selected one. With an id: switch the active model at runtime (same-provider in v1).
/effort
<level>
Set reasoning effort for the next turn. Values: low, medium, high, max, auto.
/think
<on|off|auto|budget>
Toggle extended thinking independently of /effort. auto derives the budget from /effort; a bare <budget> (or on <budget>) sets an explicit token budget. No args reports the current setting.
/status
—
Show the active provider name. Full auth/subscription status is pending the Auth spec.
/login
—
Run the active provider's auth flow (full browser OAuth implementation pending the Auth spec).
/logout
—
Clear cached credentials for the active provider (pending the Auth spec).
/setup-token
—
Generate a long-lived Anthropic OAuth token for CI use (pending the Auth spec).
View or edit memory tiers and auto-memory topic files. No args: show tier summary. delete (alias rm) only previews the deletion unless --force is given.
Show the active output style and the available list. Change the style with CALIBAN_OUTPUT_STYLE and a restart (the output_style setting is not read yet).
Open the checkpoint overlay (ADR 0028); also opened by Esc Esc on empty input. Pick a prompt with ↑/↓, then c restore code, v conversation, b both, s/S summarize after/up to, or f fork a new session. See Checkpoints & Rewind.
/statusline
—
Show the active statusLine.command configuration (command, timeout, padding) or how to set one.
/loop
[--n=<count>] [--interval=<seconds>]
Stub. Reports the planned repeat count (default 3 every 15 s, bounded by --max-turns); the repeat execution is not implemented yet.
/feedback
—
Stub. Reports that no feedback endpoint is configured.
/heapdump
—
Capture a heap profile (requires caliban to be rebuilt with --features=jemalloc-prof).
/tui
—
Toggle fullscreen vs. default TUI mode (pending TUI ergonomics spec).
/exit, /plugin (alias for /plugins), /system (view active system prompt), and /voice (reserved for future voice dictation) are registered but do not appear in /help output.
Custom slash commands
You can add your own slash commands by placing skill files under .caliban/skills/<name>/SKILL.md. See Custom Slash Commands.