mirror of
https://github.com/openclaw/openclaw.git
synced 2026-06-04 23:02:55 +00:00
75e0053cf9
Warn operators when message_tool_only produces unusually substantive private final text without a delivered source reply. Keeps short/NO_REPLY silence quiet, avoids logging response bodies, and distinguishes unrelated side effects from source-reply delivery.
22 KiB
22 KiB
title, sidebarTitle, summary, read_when
| title | sidebarTitle | summary | read_when | |||
|---|---|---|---|---|---|---|
| Slash commands | Slash commands | All available slash commands, directives, and inline shortcuts — configuration, routing, and per-surface behavior. |
|
The Gateway handles commands sent as standalone messages starting with /.
Host-only bash commands use ! <cmd> (with /bash <cmd> as an alias).
When a conversation is bound to an ACP session, normal text routes to the ACP
harness. Gateway management commands remain local: /acp ... always reaches
the OpenClaw command handler, and /status plus /unfocus stay local whenever
command handling is enabled for the surface.
Three command types
Standalone `/...` messages handled by the Gateway. Must be sent as the only content in the message. `/think`, `/fast`, `/verbose`, `/trace`, `/reasoning`, `/elevated`, `/exec`, `/model`, `/queue` — stripped from the message before the model sees it. Persist session settings when sent alone; act as inline hints when sent with other text. `/help`, `/commands`, `/status`, `/whoami` — run immediately and are stripped before the model sees the remaining text. Authorized senders only. - Directives are stripped from the message before the model sees it. - In **directive-only** messages (the message is only directives), they persist to the session and reply with an acknowledgement. - In **normal chat** messages with other text, they act as inline hints and do **not** persist session settings. - Directives only apply for **authorized senders**. If `commands.allowFrom` is set, it is the only allowlist used; otherwise authorization comes from channel allowlists/pairing plus `commands.useAccessGroups`. Unauthorized senders see directives treated as plain text.Configuration
{
commands: {
native: "auto",
nativeSkills: "auto",
text: true,
bash: false,
bashForegroundMs: 2000,
config: false,
mcp: false,
plugins: false,
debug: false,
restart: true,
ownerAllowFrom: ["discord:123456789012345678"],
ownerDisplay: "raw",
ownerDisplaySecret: "${OWNER_ID_HASH_SECRET}",
allowFrom: {
"*": ["user1"],
discord: ["user:123"],
},
useAccessGroups: true,
},
}
Command list
Commands come from three sources:
- Core built-ins:
src/auto-reply/commands-registry.shared.ts - Generated dock commands:
src/auto-reply/commands-registry.data.ts - Plugin commands: plugin
registerCommand()calls
Availability depends on config flags, channel surface, and installed/enabled plugins.
Core commands
| Command | Description | | --- | --- | | `/new [model]` | Archive the current session and start a fresh one | | `/reset [soft [message]]` | Reset the current session in place. `soft` keeps the transcript, drops reused CLI backend session ids, and reruns startup | | `/compact [instructions]` | Compact the session context. See [Compaction](/concepts/compaction) | | `/stop` | Abort the current run | | `/session idle ` | Manage thread-binding idle expiry | | `/session max-age ` | Manage thread-binding max-age expiry | | `/export-session [path]` | Export the current session to HTML. Alias: `/export` | | `/export-trajectory [path]` | Export a JSONL trajectory bundle for the current session. Alias: `/trajectory` |<Note>
Control UI intercepts typed `/new` to create and switch to a fresh
dashboard session, except when `session.dmScope: "main"` is configured
and the current parent is the agent's main session — in that case `/new`
resets the main session in place. Typed `/reset` still runs the Gateway's
in-place reset.
</Note>
<AccordionGroup>
<Accordion title="verbose / trace / fast / reasoning safety">
- `/verbose` is for debugging — keep it **off** in normal use.
- `/trace` reveals only plugin-owned trace/debug lines; normal verbose chatter stays off.
- `/fast on|off` persists a session override; use the Sessions UI `inherit` option to clear it.
- `/fast` is provider-specific: OpenAI/Codex map it to `service_tier=priority`; direct Anthropic requests map it to `service_tier=auto` or `standard_only`.
- `/reasoning`, `/verbose`, and `/trace` are risky in group settings — they may reveal internal reasoning or plugin diagnostics. Keep them off in group chats.
</Accordion>
<Accordion title="Model switching details">
- `/model` persists the new model immediately to the session.
- If the agent is idle, the next run uses it right away.
- If a run is active, the switch is marked pending and applied at the next clean retry point.
</Accordion>
</AccordionGroup>
Dock commands
Dock commands switch the active session's reply route to another linked channel. See Channel docking for setup and troubleshooting.
Generated from channel plugins with native-command support:
/dock-discord(alias:/dock_discord)/dock-mattermost(alias:/dock_mattermost)/dock-slack(alias:/dock_slack)/dock-telegram(alias:/dock_telegram)
Dock commands require session.identityLinks. The source sender and target peer
must be in the same identity group.
Bundled plugin commands
| Command | Description |
|---|---|
/dreaming [on|off|status|help] |
Toggle memory dreaming. See Dreaming |
/pair [qr|status|pending|approve|cleanup|notify] |
Manage device pairing. See Pairing |
/phone status|arm ...|disarm |
Temporarily arm high-risk phone node commands |
/voice status|list|set <voiceId> |
Manage Talk voice config. Discord native name: /talkvoice |
/card ... |
Send LINE rich card presets. See LINE |
/codex status|models|threads|resume|compact|review|diagnostics|account|mcp|skills |
Control the Codex app-server harness. See Codex harness |
QQBot-only: /bot-ping, /bot-version, /bot-help, /bot-upgrade, /bot-logs
Skill commands
User-invocable skills are exposed as slash commands:
/skill <name> [input]always works as the generic entrypoint.- Skills may register as direct commands (e.g.
/prosefor OpenProse). - Native skill-command registration is controlled by
commands.nativeSkillsandchannels.<provider>.commands.nativeSkills. - Names are sanitized to
a-z0-9_(max 32 chars); collisions get numeric suffixes.
Skills can declare `command-dispatch: tool` to route directly to a tool
(deterministic, no model involvement). Example: `/prose` (OpenProse plugin)
— see [OpenProse](/prose).
/tools — what the agent can use now
/tools answers a runtime question: what this agent can use right now in this
conversation — not a static config catalog.
/tools # compact view
/tools verbose # with short descriptions
Results are session-scoped. Changing agent, channel, thread, sender authorization, or model can change the output. For profile and override editing, use the Control UI Tools panel or config surfaces.
/model — model selection
/model # show model picker
/model list # same
/model 3 # select by number from picker
/model openai/gpt-5.4
/model opus@anthropic:default
/model status # detailed view with endpoint and API mode
On Discord, /model and /models open an interactive picker with provider and
model dropdowns. The picker respects agents.defaults.models, including
provider/* entries.
/config — on-disk config writes
Owner-only. Disabled by default — enable with `commands.config: true`.
/config show
/config show messages.responsePrefix
/config get messages.responsePrefix
/config set messages.responsePrefix="[openclaw]"
/config unset messages.responsePrefix
Config is validated before write. Invalid changes are rejected. /config
updates persist across restarts.
/mcp — MCP server config
Owner-only. Disabled by default — enable with `commands.mcp: true`.
/mcp show
/mcp show context7
/mcp set context7={"command":"uvx","args":["context7-mcp"]}
/mcp unset context7
/mcp stores config in OpenClaw config, not embedded-agent project settings.
/debug — runtime-only overrides
Owner-only. Disabled by default — enable with `commands.debug: true`.
Overrides apply immediately to new config reads but do **not** write to disk.
/debug show
/debug set messages.responsePrefix="[openclaw]"
/debug set channels.whatsapp.allowFrom=["+1555","+4477"]
/debug unset messages.responsePrefix
/debug reset
/plugins — plugin management
Owner-only for writes. Disabled by default — enable with `commands.plugins: true`.
/plugins
/plugins list
/plugin show context7
/plugins enable context7
/plugins disable context7
/plugins install ./path/to/plugin
/plugins enable|disable updates plugin config and hot-reloads the Gateway
plugin runtime for new agent turns. /plugins install restarts managed
Gateways automatically because plugin source modules changed.
/trace — plugin trace output
/trace # show current trace state
/trace on
/trace off
/trace reveals session-scoped plugin trace/debug lines without full verbose
mode. It does not replace /debug (runtime overrides) or /verbose (normal
tool output).
/btw — side questions
/btw is a quick side question about the current session context. Alias: /side.
/btw what are we doing right now?
/side what changed while the main run continued?
Unlike a normal message:
- Uses the current session as background context.
- In Codex harness sessions, runs as an ephemeral Codex side thread.
- Does not change future session context.
- Is not written to transcript history.
See BTW side questions for the full behavior.
Surface notes
- **Text commands:** run in the normal chat session (DMs share `main`, groups have their own session). - **Native Discord commands:** `agent::discord:slash:` - **Native Slack commands:** `agent::slack:slash:` (prefix configurable via `channels.slack.slashCommand.sessionPrefix`) - **Native Telegram commands:** `telegram:slash:` (targets the chat session via `CommandTargetSessionKey`) - **`/stop`** targets the active chat session to abort the current run. `channels.slack.slashCommand` supports a single `/openclaw`-style command. With `commands.native: true`, create one Slack slash command per built-in command. Register `/agentstatus` (not `/status`) because Slack reserves `/status`. Text `/status` still works in Slack messages. - Command-only messages from allowlisted senders are handled immediately (bypass queue + model). - Inline shortcuts (`/help`, `/commands`, `/status`, `/whoami`) also work embedded in normal messages and are stripped before the model sees the remaining text. - Unauthorized command-only messages are silently ignored; inline `/...` tokens are treated as plain text. - Commands accept an optional `:` between the command and args (`/think: high`, `/send: on`). - `/new ` accepts a model alias, `provider/model`, or a provider name (fuzzy match); if no match, the text is treated as the message body. - `/allowlist add|remove` requires `commands.config: true` and honors channel `configWrites`.Provider usage and status
- Provider usage/quota (e.g., "Claude 80% left") shows in
/statusfor the current model provider when usage tracking is enabled. - Token/cache lines in
/statuscan fall back to the latest transcript usage entry when the live session snapshot is sparse. - Execution vs runtime:
/statusreportsExecutionfor the effective sandbox path andRuntimefor who is running the session:OpenClaw Default,OpenAI Codex, a CLI backend, or an ACP backend. - Per-response tokens/cost: controlled by
/usage off|tokens|full. /model statusis about models/auth/endpoints, not usage.