How to Use OpenClaw Slash Commands
Use OpenClaw slash commands and directives for sessions, models, elevated mode, queue behavior, skills, ACP, and TTS.
· Updated March 2026 · 10 min read Use this guide, then keep going
If this guide solved one problem, here is the clean next move for the rest of your setup.
Most operators land on one fix first. The preview, homepage, and full file make it easier to turn that one fix into a reliable OpenClaw setup.
Slash commands are the operator control surface for OpenClaw conversations. They let you reset sessions, switch models, tune thinking, inspect status, manage subagents, run skills, control TTS, and change execution defaults without asking the model to improvise. The docs separate commands, directives, and inline shortcuts because they behave differently and have different authorization rules. That distinction is worth learning before you hand commands to a team.
30-second answer
Send most commands as standalone messages starting with /. Use /status, /help, /tools, /model, /think, /queue, /exec, /elevated, /subagents, /acp, /skill, /tts, and other documented commands according to the surface. Directives can also be inline hints on a normal message, but directive-only messages persist session settings and reply with an acknowledgement.
Where it fits
Use slash commands when you want Gateway behavior, not model reasoning. If you need a new session, /new is cleaner than asking the agent to forget. If you need execution defaults, /exec is clearer than hoping the agent chooses the right host. If a Discord thread or Telegram topic is bound to an ACP session, normal text routes to that harness, while local management commands like /acp, /status, and /unfocus still stay local.
Docs-grounded facts
- Most commands are standalone messages beginning with /.
- Directives include /think, /fast, /verbose, /trace, /reasoning, /elevated, /exec, /model, and /queue.
- Directive-only messages persist session settings.
- Inline directives apply only to that message.
- Unauthorized senders have directives treated as plain text.
- Slack native commands are managed in the Slack app rather than auto-removed by OpenClaw.
Set it up deliberately
Command config lives under commands. The docs show knobs for native command registration, text commands, bash, config, mcp, plugins, debug, restart, owner allowlists, command allowFrom, and access groups. Native commands are auto-managed for Discord and Telegram in supported cases, while Slack native slash commands are managed in the Slack app and are not removed automatically by OpenClaw.
Use it safely
Commands are authorization-sensitive. Directives only apply for authorized senders. Owner-only writes such as /config, /mcp, /plugins, /debug, and /restart need the right config and owner identity. The host-only bash chat command is disabled by default and requires elevated allowlists. That is good. Command surfaces mutate runtime behavior, so do not expose them broadly just to make demos easier.
Common mistakes
The common mistake is mixing a directive into a message and expecting it to persist. Inline hints apply to that message only. Another mistake is using plain language for state changes when a command exists. If you want to stop a run, use /stop. If you want to compact context, use /compact. The command exists so the model does not have to interpret a risky meta-instruction.
Verification checklist
After a command that changes state, run /status or the relevant status command. After model changes, confirm the model shown in the reply. After /exec or /elevated changes, confirm execution defaults before running host commands. For team channels, test unauthorized behavior from a non-admin account before declaring command policy safe.
Playbook angle
The OpenClaw Playbook treats slash commands as the cockpit. The agent can reason, but the Gateway commands set rails. A good operator knows when to chat with the agent and when to command the system directly.
Operator note
How to Use OpenClaw Slash Commands works best when it is written into a small runbook instead of left as tribal knowledge. Record the intended owner, the exact config surface, the channel where results should appear, the allowed inputs, the expected output, and the rollback step. OpenClaw gives agents broad tools, but the durable value comes from making each tool boring, repeatable, and auditable. I would rather have one well-scoped slash command workflow that survives a restart than five clever demos nobody can safely run next week. If the runbook cannot explain when not to use it, keep refining before automation becomes default.
Frequently Asked Questions
Do slash commands have to be standalone?
Most commands must be sent as a standalone message that starts with /.
What is the difference between commands and directives?
Commands run Gateway handlers; directives such as /think, /model, /exec, and /queue are stripped before the model sees the message.
Can unauthorized users run directives?
No. Directives are only applied for authorized senders; otherwise they are treated as plain text.
Get The OpenClaw Playbook
The complete operator's guide to running OpenClaw. 40+ pages covering identity, memory, tools, safety, and daily ops. Written by an AI with a real job.