Slash Commands¶
Inside Claude Code, /z commands let you interact with the reviewer and control your monitoring session.
Command reference¶
/z spy¶
View the reviewer's transcript for the current session. Shows what was reviewed, what was blocked, and the reviewer's reasoning for each determination.
By default, shows the latest user turn. You can specify a turn number to view earlier turns.
/z ask¶
Ask the reviewer a question about the current session without changing its monitoring behavior. Use this to understand why the reviewer made a specific decision, or to get the reviewer's opinion on the agent's work.
/z direct¶
Send a high-priority directive that overrides the reviewer's behavior for the rest of the session. Use this when you need the reviewer to stand down on a specific issue or change its approach.
Directives are non-enforcing – the reviewer acknowledges and continues monitoring under the new instructions.
/z agent [agent_name]¶
Switch Claude Code's active agent persona mid-session. Presents a picker of available agents from ~/.claude/agents/, resolves the corresponding reviewer profile from settings.yaml, and cycles the reviewer to a fresh session with the new agent's definition as its rubric. Claude Code then reads and follows the selected agent's instructions.
The reviewer always cycles on an agent switch - it receives a fresh session where its rubric contains the new agent's .md file (and CLAUDE.md if include_claude_md is enabled for the target profile). The agent must have a matching profile in settings.yaml with an agents: list that includes it.
/z adversary [profile] [--cycle]¶
Switch the reviewer's behavioral profile mid-session. All settings update immediately -- operating rules, tool visibility, scrutiny level. By default, the reviewer keeps its conversation context so it retains everything it has observed. Pass --cycle to tear down the reviewer's session and start a new one with a handoff summary.
If you provide a profile name directly (e.g. /z adversary code-review), the switch happens immediately without the profile picker wizard.
/z model [model_id]¶
Switch the reviewer's model mid-session. This always cycles the reviewer's session – the current reviewer writes a handoff summary, then a new session starts on the target model with that summary plus a tail of recent events. The handoff and tail behavior are governed by your cycle settings (tail_depth, etc.).
If you provide a model ID directly, the switch happens immediately. If you omit it, the skill shows the current model, checks which providers have usable credentials, and presents a filtered model list grouped by provider. Only providers with configured credentials appear. For providers that support multiple auth modes (currently OpenAI with both API key and OAuth), you're prompted to choose.
Credentials for the target model must be available. If they aren't, the command tells you to run zwischen auth in your terminal.
/z cycle¶
Reset the reviewer's context window while keeping the same profile. The reviewer writes a handoff summary, a new session starts with that summary plus recent events, and monitoring continues with a fresh context.
Useful when the session has gone long and the reviewer seems stale or disoriented.
/z enable / /z disable¶
Enable or disable supervision for the current session or all sessions.
/z disable # This session only
/z disable --all # All sessions until re-enabled
/z enable # Re-enable
/z unblock¶
Clear all active blocks for the current session without disabling supervision. The reviewer keeps watching; the current block is just released.
Use this when message_adversary is unavailable (MCP glitch, lost session ID) and you need to clear a block. Unlike /z disable, monitoring continues after unblocking.
/z session-id¶
Show the current session ID. Mostly useful for diagnostics.
/z oauth¶
Authenticate or re-authenticate with a model provider (currently OpenAI only).
/z update¶
Update the Zwischen client package and restart the daemon.
/z restart¶
Restart the Zwischen daemon without updating.