Terminal Commands Reference

Shell commands for managing the Zwischen installation. These are run in your terminal, not inside Claude Code.

Command	Description
zwischen install	Interactive setup wizard
zwischen install --dry-run	Preview changes without writing anything
zwischen install --print	Print hook and MCP config for manual application
zwischen install --update	Re-wire hooks, MCP, and skills without running the wizard
zwischen install --reconfigure	Replay the full setup with every answer pre-filled
zwischen install --host codex	Install into Codex only (see Codex)
zwischen config	Open the interactive settings editor
zwischen update	Update the package and restart the daemon
zwischen uninstall	Remove Zwischen from Claude Code
zwischen uninstall --purge	Also remove ~/.zwischen/settings.yaml
zwischen auth	Configure provider credentials (API key or OAuth)
zwischen providers	Show providers with usable credentials
zwischen models	List selectable reviewer models from the server catalog
zwischen adversaries	List your custom reviewer prompts
zwischen codex <agent>	Launch a monitored Codex session
zwischen restart	Restart the daemon
zwischen monitor --session-id <id>	Open a live window streaming the reviewer's verdicts

zwischen install

Runs the interactive setup wizard. Prompts for your Zwischen API key, model selection, and provider API key. Installs hooks into Claude Code settings, registers the MCP server, and installs the /z skill.

--dry-run – Shows what would be changed without writing anything. Use this to preview before committing.

--print – Prints the hook and MCP configuration as JSON for manual application. Useful if you want to integrate Zwischen into an existing Claude Code config without the wizard touching your files.

--update – Re-runs the hook, MCP, and skill installation without prompting for API keys or model selection. Use this after updating the package to ensure your Claude Code wiring matches the latest version.

--reconfigure – Replays the full setup on a machine that is already configured, with every step pre-filled from your current settings. Pressing enter at each prompt keeps the existing value, so this is the safe way to revisit one answer without retyping the rest.

--host – Which host to install into. Defaults to auto, which sets up every host it detects (Claude Code, Codex, or both). Pass claude or codex to target one. See Codex.

zwischen config

Opens the interactive settings editor in a new terminal window. This is the recommended way to change ~/.zwischen/settings.yaml if you would rather not hand-edit YAML.

zwischen config

The editor asks you to pick a scope first – defaults, a named profile, global settings, or credentials – then shows every setting in that scope with its current value and lets you change it from a menu. Profile scopes show what each unset key inherits from defaults.

Settings you have not set yet live under 'Add or change a setting'

A scope's opening view lists only the settings that scope currently overrides. Everything else – including model and adversary – is one level down under + Add or change a setting. If a setting looks unavailable for a profile, look there before concluding it can only be set in defaults.

Changes apply on the reviewer's next turn. Environment variable changes need zwischen restart.

The same editor is available inside a session as /z config.

zwischen update

Updates the zwischen package to the latest version from PyPI and restarts the daemon. Automatically detects whether Zwischen was installed with pipx or pip and uses the matching upgrade path. Equivalent to:

pipx upgrade zwischen   # or: pip install --upgrade zwischen
zwischen install --update
zwischen restart

zwischen uninstall

Removes all Zwischen hooks from Claude Code settings, unregisters the MCP server, and removes the /z skill. Preserves ~/.zwischen/settings.yaml.

--purge – Also deletes ~/.zwischen/settings.yaml.

--host – Which host to remove from. Defaults to claude. Pass codex to remove Zwischen's hooks and MCP entry from ~/.codex/config.toml instead.

zwischen auth

Interactive terminal credential setup for model providers. Supports Anthropic, Google, OpenAI, and DeepSeek.

zwischen auth                    # Show status, then pick a provider
zwischen auth --provider google  # Configure Google directly

For OpenAI, you're prompted to choose between API key and OAuth. Selecting OAuth opens a browser-based authentication flow via your ChatGPT account.

API keys are persisted as OS environment variables (Windows registry on Windows, managed shell files on macOS/Linux). Keys are not stored in ~/.zwischen/settings.yaml.

When a provider's environment variable is already set, zwischen auth reports where the existing credential came from and asks whether to keep or replace it. Keeping references the existing variable without reading the secret. Replacing overwrites the variable in place, which is how you rotate a changed or leaked key. Keep is the default, so pressing enter never clobbers an existing credential.

zwischen providers

Show which providers have usable credentials without revealing secrets.

zwischen providers

Output is space-separated provider:auth_mode pairs, e.g.:

anthropic:api_key google:api_key openai:oauth

Used by the /z model skill internally to filter the model picker, but also useful for manual diagnostics.

zwischen models

Lists the reviewer models you can actually select, grouped by provider.

zwischen models

The list comes from the Zwischen server's model catalog, filtered to providers you have usable credentials for. The client does not keep its own copy, so this is always current – and a model the server does not list is rejected when you try to switch to it, rather than failing later inside the provider call.

This is the source the /z model picker and the setup wizard both read.

zwischen adversaries

Lists the custom reviewer prompts you have written in ~/.zwischen/agents/, with the first line of each as its description.

zwischen adversaries

Only usable files appear – ones that are unreadable, empty, or named after a built-in reviewer are skipped, matching exactly what resolves at runtime. See Custom Prompts.

zwischen codex

Launches a Codex session monitored by Zwischen, as a named agent.

zwischen codex coder                 # interactive session
zwischen codex coder "fix the auth bug"   # with an initial prompt

The agent name matches a profile in settings.yaml and a role file at ~/.codex/agents/<name>.toml. See Codex.

zwischen restart

Restarts the local Zwischen daemon. The daemon will be re-started automatically on the next Claude Code session if it's not running.

zwischen monitor

Opens a separate terminal window that streams the reviewer's verdicts for a session in real time. Usually invoked as /z monitor inside a session ($z monitor in Codex), which fills in the session ID for you.

zwischen monitor --session-id <session_id>