Coven Documentation

Command reference for Coven Code, including CLI flags, headless print mode, provider login commands, the complete slash-command catalog, and daemon-aware /coven workflows.

Shell Commands

Open the TUI:

coven tui
coven-code

Run a one-shot prompt and print the result:

coven-code --print "Explain the failing test"
coven-code -p --provider codex "Summarize this repository"

Resume a previous session:

coven-code --resume
coven-code --resume <session-id>
coven-code --continue

Choose provider, model, and effort:

coven-code --provider anthropic --model claude-sonnet-4-6
coven-code --provider codex --effort high
coven-code --fallback-model claude-haiku-4-6

Common Flags

All flags below are defined by the coven-code binary itself. Run coven-code --help for the authoritative list in your installed version.

Launch & Session

Flag	Purpose
`[PROMPT]`	Positional initial prompt. Combined with `--print` or `--headless` it runs one-shot.
`--print`, `-p`	Print mode: send the prompt and exit (non-interactive).
`--continue`, `-c`	Continue the most recent conversation.
`--resume [id]`	Resume a previous session by ID, or the most recent session when the ID is omitted.
`--session-id <id>`	Tag a headless run with a session ID for tracking in logs and hooks.
`--cwd <dir>`	Set the working directory.
`--add-dir <dir>`	Grant Coven access to an additional directory. Repeatable.
`--model <model>`, `-m`	Choose the active model.
`--fallback-model <model>`	Fallback model if the primary model is overloaded or unavailable.
`--provider <provider>`	Select `anthropic` or `codex`. Env: `COVEN_CODE_PROVIDER`.
`--api-base <url>`	Override the API base URL for the selected provider. Env: `COVEN_CODE_API_BASE`.
`--api-key <key>`	API key for the active provider; overrides provider-specific env vars.
`--agent <name>`, `-A`	Use a named agent profile such as `build`, `plan`, or `explore`.
`--effort <level>`	Effort level for extended thinking: `low`, `medium`, `high`, or `max`.
`--thinking <tokens>`	Extended-thinking budget in tokens (enables extended thinking).

Permissions & Safety

Flag	Purpose
`--permission-mode <mode>`	Use `default`, `accept-edits`, `bypass-permissions`, or `plan`.
`--dangerously-skip-permissions`	Bypass all permission checks. Alias: `--yolo`.
`--allowed-tools <tools>`	Comma-separated list of tools to allow (default: all).
`--disallowed-tools <tools>`	Comma-separated list of tools to disallow.
`--max-turns <n>`	Stop after a maximum number of agentic turns (default: 10).
`--max-tokens <n>`	Maximum tokens per response.
`--max-budget-usd <usd>`	Maximum spend in USD before aborting the query loop.

Input, Output & Automation

Flag	Purpose
`--output-format <format>`	Use `text`, `json`, or `stream-json`.
`--input-format <format>`	Input format for `--print` mode: `text` or `stream-json` (newline-delimited JSON messages).
`--prefill <text>`	Prefill the first assistant turn with this text.
`--headless`	Run without the TUI: non-interactive, structured output. Equivalent to `--print` for input and output handling.
`--context <brief.json>`	Load a session brief (JSON) for fully automated headless runs; implies headless mode.
`--output <result.json>`	Write a structured result envelope (status, branch, commits, files changed, summary) on exit.
`--hosted-review`	Run with hosted review memory isolation. Env: `COVEN_CODE_HOSTED_REVIEW`.
`--hosted-repair`	Hosted repair mode: allow repository file edits but no command execution, plugins, MCP servers, sub-agents, or network tools. Requires `--headless`, `--context`, and `--output`.
`--workload <tag>`	Billing workload tag.
`--verbose`, `-v`	Enable verbose logging.

Config & Prompt

Flag	Purpose
`--system-prompt <text>`, `-s`	Replace the system prompt.
`--append-system-prompt <text>`	Append extra system instructions.
`--system-prompt-file <path>`	Override the system prompt from a file.
`--no-claude-md`	Disable AGENTS.md memory file injection.
`--mcp-config <json>`	Inline MCP server definitions as a JSON string.
`--no-auto-compact`	Disable auto-compaction.
`--auto-commits`	Enable shadow-git auto-commit snapshots, which power `/rewind list`, `/rewind diff`, and turn rollback.
`--betas <headers>`	Extra beta feature headers to send (comma-separated).
`--disable-slash-commands`	Disable all slash commands.
`--bare`	Run in bare mode: no hooks, no plugins, no AGENTS.md.

Provider Commands

coven-code auth login
coven-code codex login
coven-code models
coven-code models anthropic --refresh
coven-code models codex --json

Use /connect in the TUI when you prefer an interactive provider picker.

Key Slash Commands

The commands you will reach for most often:

Command	Purpose
`/connect`	Connect an AI provider: connect, import, or switch provider accounts.
`/model`	Show or change the current model with the provider's model picker.
`/effort LEVEL`	Set the model's thinking effort to `low`, `normal`, or `high`. `/effort fast` toggles fast mode (smaller, quicker model).
`/coven`	Drive the local Coven daemon (sessions, harness runs, rituals) from inside the TUI.
`/help`	Show available commands and usage information. Aliases: `/h`, `/?`.
`/export share`	Upload the current session as a secret GitHub gist and return a shareable URL.
`/goal OBJECTIVE`	Set or manage a durable long-running goal for autonomous work across turns.

Slash Command Catalog

The full built-in registry, grouped by task. Aliases appear in parentheses.

Session & Context

Command	Purpose
`/clear` (`/c`, `/reset`, `/new`)	Clear the conversation history.
`/compact`	Compact the conversation to reduce token usage.
`/resume` (`/r`, `/continue`)	Resume a previous conversation.
`/rewind [list\|diff [n]\|last\|<n>]`	Rewind the conversation or roll back a turn's file changes recorded by the shadow-git snapshot system.
`/session` (`/remote`)	Show or manage conversation sessions. Subcommands: `rename`, `fork`, `branch`, `tag`, `add-dir`.
`/search`	Search across all sessions.
`/export`	Export the conversation to markdown or JSON. Subcommands: `copy [n]`, `share`.
`/status`	Show comprehensive system and session status. `/status doctor` runs the full diagnostic suite.
`/usage`	Show API usage, quotas, and rate limit status. Subcommands: `cost`, `context`, `stats`.
`/memory`	View, edit, or clear AGENTS.md memory files.
`/init`	Initialize a new project with AGENTS.md.
`/tasks` (`/bashes`)	List and manage background tasks.
`/goal`	Set or manage a durable long-running goal for autonomous work.

Config & Auth

Command	Purpose
`/config` (`/settings`)	Show or modify configuration settings. Also hosts `theme`, `keybindings`, `output-style`, `color`, `vim`, `voice`, `statusline`, `terminal-setup`, `import`, and `advisor` subcommands.
`/login`	Authenticate with Anthropic or Codex (multi-account). `/login switch [name]` switches the active account; `/login refresh` clears saved provider auth and model caches.
`/logout`	Clear credentials for the active account.
`/accounts`	List stored Anthropic and Codex accounts.
`/providers`	List available AI providers and their status.
`/connect`	Connect an AI provider.
`/model`	Show or change the current model.
`/effort`	Set the model's thinking effort (`low`, `normal`, `high`).
`/thinking` (`/think`)	Toggle extended thinking mode. `/thinking back [play]` shows or replays extended-thinking traces.
`/permissions`	View or change tool permission settings.
`/config import`	Import CLAUDE.md and settings.json from `~/.claude` (Claude CLI import).
`/status doctor`	Check system health and diagnose issues.

Review & Safety

Command	Purpose
`/review`	Review code changes via LLM and optionally post to a GitHub PR.
`/review security`	Run a security review of the current project (absorbs the former `/security-review`).
`/review ultra`	Run an exhaustive multi-dimensional code review (absorbs the former `/ultrareview`).
`/review comments`	Get comments from a GitHub pull request.
`/diff`	Show the git diff of changes in the working directory.
`/commit`	Ask Coven to commit staged changes.
`/plan`	Enter plan mode: the model outputs a plan for approval before acting.
`/sandbox`	Enable or disable sandboxed execution of shell commands.
`/rewind list`	List assistant turns that have recorded file changes.
`/rewind diff [n]`	Show the shadow-git diff of a turn's file changes without reverting.
`/rewind last`	Revert file changes from the most recent assistant turn back to pre-turn state.

Editor & UX

Command	Purpose
`/config theme`	Show or change the current theme.
`/config keybindings`	Create or open `~/.coven-code/keybindings.json`.
`/config output-style`	Show or switch the current output style.
`/config color`	Set or show the prompt bar color for this session.
`/config vim`	Toggle vim keybinding mode on or off.
`/config voice`	Toggle voice input mode on or off.
`/config statusline`	Configure what is shown in the status line.
`/config terminal-setup`	Help configure your terminal for optimal Coven use.
`/splash` (`/welcome-screen`)	Show, hide, or toggle the empty-session splash screen.
`/export copy [n]`	Copy the (n-th most recent) assistant response to the clipboard.

Integrations

Command	Purpose
`/mcp`	Show MCP server status and manage connections.
`/hooks`	Show configured event hooks.
`/chrome`	Browser automation via the Chrome DevTools Protocol (CDP).
`/plugin` (`/plugins`)	Manage plugins. `/plugin reload` reloads all plugins without restarting.
`/coven`	Drive the local Coven daemon (sessions, harness runs, rituals).
`/link` (`/links`)	Save and manage links in the structured stash.
`/attach` (`/attachments`)	Save and manage file attachments in the structured stash.

Agents & Experimental

Command	Purpose
`/familiar` (`/familiars`, `/agent`)	Your familiar and agents: switch persona, inspect, and manage. `/familiar managed` configures the manager-executor architecture (absorbs the former `/managed-agents` and `/agents`).
`/config advisor`	Set or unset the server-side advisor model.
`/effort fast`	Toggle fast mode, which uses a faster and cheaper model (absorbs the former `/fast`).
`/thinking back [play]`	Show thinking traces from previous responses, or replay one as an animated walkthrough (absorbs the former `/think-back` and `/thinkback-play`).
`/whisper` (`/btw`)	Whisper a side question to your familiar without adding it to history.
`/incant`	Cast a speech incantation: change the model's voice (caveman, rocky) or lift it.
`/learn` (`/scribe`)	Codify a script or workflow you just built into a reusable skill.
`/skills` (`/skill`)	List available skills in `.coven-code/commands/`.

General & System

Command	Purpose
`/help` (`/h`, `/?`)	Show available commands and usage information.
`/exit` (`/quit`, `/q`)	Exit Coven.
`/version` (`/v`)	Show version information.
`/release-notes` (`/whats-new`, `/changelog`)	Show recent Coven highlights.
`/update` (`/upgrade`)	Check for updates and download the latest release.
`/feedback` (`/bug`)	Submit feedback about Coven.

Legacy Names & Custom Commands

Several commands were folded into their parent surfaces. Old top-level names such as /doctor, /switch, /refresh, /copy, /share, /fork, /branch, /tag, /add-dir, /theme, /keybindings, /output-style, /import-config, /advisor, /fast, /think-back, /reload-plugins, /agents, /managed-agents, and /pr-comments remain hidden one-release compatibility aliases; prefer the subcommand forms above. Fully retired names (/cost, /context, /stats, /color, /vim, /voice, /statusline, /terminal-setup, /revert, /undo) now live only as subcommands of /usage, /config, and /rewind.

You can extend the registry yourself: user-defined command templates in ~/.coven-code/settings.json (commands), discovered skills, and plugins all contribute additional slash commands. See Configuration.

Daemon-Aware Commands

The /coven command talks to the local daemon over ~/.coven/coven.sock by default. Before relying on it, verify:

coven doctor
coven daemon status
coven daemon start

If the daemon is unreachable, Coven Code can still operate as a coding TUI, but /coven actions that list or manage daemon sessions will fail closed. For the daemon lifecycle, see CLI Daemon and Daemon Lifecycle.

Commands

On this page