Hermes Agent slash commands are the shortcuts you type into a chat, starting with a forward slash, to control the agent without leaving the conversation. /new starts fresh, /model switches the model, /queue lines up your next prompt, /stop kills background work. They work in the Hermes terminal chat, and most of them work in Telegram, Discord, Slack and the other messaging apps too.
📺 Watch: NEW Hermes /Review Workflow is SCARY Good
🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside
This is the hub. Below is every command the official Hermes docs confirm, grouped by the job it does, plus what's new for plugin commands in v0.22.5. Where I've written a deeper guide to one command, it's linked so you can go straight there.
How Hermes slash commands work
Per the Hermes command reference, there are two places slash commands run, and both come from one central registry:
- The interactive CLI. Type a forward slash and an autocomplete menu opens. Commands aren't case-sensitive.
- Messaging platforms. The gateway handles the same registry for Telegram, Discord, Slack, WhatsApp, Signal, email, Home Assistant and Teams, with help text and platform menus built from it.
Three things make them faster to use:
- Prefix matching. Type /h and you get /help. Type /mod and you get /model. Full names and aliases always win over a prefix.
- Skills become commands. Every installed skill shows up as its own command, named after the skill. If a skill shares a name with a built-in, the built-in wins.
- Slack threads use a bang. Slack blocks slash commands inside threads, so use !stop, !new or !status there and Hermes treats them the same.
Can't remember a name? Open the fuzzy command palette with /palette or Ctrl+P, type a few letters, and press Enter to drop the command into your message. It never runs it for you. My guide on how to use the Hermes Agent command palette covers the rest.
I run Hermes every day inside my Agent OS, with a profile per model, and slash commands are how I steer it. Want that setup ready to go? The AI Profit Boardroom ($69/mo) gives you the Agent OS, a 30-day AI agent roadmap, daily tutorials as new tools ship, 4 coaching calls a week and the prompt library. Over 3,400 members are in there.
Hermes Agent slash commands by job
Every command here is in the Hermes slash command reference. Aliases are in brackets. A few are CLI-only or messaging-only, and the notes say so.
| Job | Command | What it does |
|---|---|---|
| Start and manage sessions | /new (/reset) | Fresh session with a new ID and empty history. Add a name to title it. |
| Start and manage sessions | /resume, /sessions | Go back to an earlier session, or browse them |
| Start and manage sessions | /title, /branch (/fork) | Name the session, or branch it into a separate copy |
| Start and manage sessions | /retry, /undo | Resend the last message, or remove the last exchange |
| Control a running turn | /stop | Kills running background processes. On messaging it also interrupts the agent. |
| Control a running turn | /queue (/q) | Lines up a prompt for the next turn without interrupting this one |
| Control a running turn | /steer | Adds a note the agent picks up after its next tool call |
| Control a running turn | /busy | Sets what happens when you message mid-turn: queue, steer or interrupt |
| Side work | /btw | Asks a quick side question about the chat without interrupting it |
| Side work | /bg | Runs a prompt in a separate background session |
| Side work | /review | Spins up an independent reviewer subagent for the work you just did |
| Long-running work | /goal, /subgoal | Sets a standing goal Hermes keeps working toward, and adds criteria to it |
| Long-running work | /loop (/proactive), /heartbeat (/hb) | Re-runs a prompt on an interval, or nudges an idle session |
| Long-running work | /agents (/tasks) | Shows active agents and running tasks |
| Planning and project setup | /plan | Writes a markdown plan to .hermes/plans/ without running anything |
| Planning and project setup | /init | Creates or updates AGENTS.md from a scan of your repo |
| Models and behaviour | /model | Shows or changes the model, including provider switches and your own aliases |
| Models and behaviour | /reasoning, /fast | Sets reasoning effort and display, or fast mode |
| Models and behaviour | /personality, /skin | Changes the persona overlay, or the display theme |
| Models and behaviour | /approvals, /yolo | Sets the dangerous-command approval mode, or skips approvals |
| Context and cost | /compress, /context (/ctx) | Summarises old context, or shows what's filling the window |
| Context and cost | /usage, /insights | Token use, cost and account limits, or 30 days of analytics |
| Skills, memory and tools | /learn | Turns a folder, URL, workflow or notes into a reusable skill |
| Skills, memory and tools | /skills, /memory, /curator | Manages skills, reviews pending memory writes, runs skill maintenance |
| Skills, memory and tools | /tools, /plugins, /reload-mcp | Manages tools, lists plugins, reloads MCP servers |
| Automation | /cron, /blueprint (/bp), /suggestions | Scheduled tasks, automation templates and suggested automations |
| Automation | /kanban | Drives the multi-profile task board from chat |
| Files and changes | /diff, /rollback, /snapshot (/snap) | Shows git changes, restores file checkpoints, snapshots Hermes state |
| Account and help | /login, /topup, /subscription (/upgrade) | Nous sign-in, balance and billing, your plan (CLI only) |
| Account and help | /help, /status, /version, /update, /debug | Help, session info, version, update, debug report |
| Messaging only | /sethome, /approve, /deny, /restart, /pause | Home channel, approve or reject a risky command, restart or pause the gateway |
| CLI only | /worktree, /handoff, /copy, /quit (/exit) | Git worktrees, hand a session to a messaging app, copy a reply, exit |
Typed /new by accident? The CLI asks before running commands that throw away session state: /clear, /new, /reset, /undo and /exit with --delete. You get Approve Once, Always Approve or Cancel. Add now, --yes or -y to skip the prompt once, like /reset now.
📺 Watch: Hermes3D Is INSANE: AI Agents Now Have a 3D Office
The commands worth learning first
You don't need all of them. These are the ones that change how you work day to day, each with its own guide.
Keep the agent moving: /btw, /bg and /busy
The biggest time sink with any agent is waiting for it to finish. Three commands fix that. /btw answers a quick side question from a snapshot of the chat while the main turn keeps going (see the Hermes /btw command guide). /bg runs a whole prompt in its own background session and brings the result back when it's done (the Hermes /bg command guide). /busy decides what your messages do while Hermes is mid-turn: queue them, steer the current turn, or interrupt it (the Hermes /busy command guide).
Long jobs: /goal and /plan
/goal sets a standing goal. After each turn a judge model checks whether it's done, and if not, Hermes carries on, up to a turn budget that defaults to 20. You can pause, resume or clear it, and add extra criteria with /subgoal. Full walkthrough in the Hermes /goal command guide.
/plan does the opposite: it writes a markdown plan to .hermes/plans/ and runs nothing. It used to be a bundled skill and is now built in, so it survives the command-menu caps on Telegram and Discord. More in the Hermes /plan command guide.
Teach it once: /learn
/learn turns almost anything into a reusable skill: a folder, a URL, a workflow you just walked the agent through, or pasted notes. The agent gathers the sources itself and writes the SKILL.md. It works in the CLI and on messaging. See the Hermes /learn command guide.
Want help picking which commands fit your workflow? Book a free AI strategy session and we'll set it up.
📺 Watch: Hermes AI Agents Just Went Portable
Terminal commands people mix up with slash commands
Some of the most useful Hermes commands aren't typed into a chat at all. They run in your terminal, starting with hermes:
- hermes doctor checks your install for problems, and hermes doctor --fix tries automatic repairs. See the Hermes doctor command guide.
- hermes backup zips your Hermes config, skills, sessions and data, leaving out the code itself. See the Hermes backup command guide.
- hermes usage shows your account's rate-limit windows without opening a session.
Rule of thumb: if it starts with hermes, run it in the terminal. If it starts with a slash, type it into a chat.
New in v0.22.5: plugin commands that ask you to pick
Plugins can add their own slash commands, and as of v0.22.5 those commands can ask you to pick an option. The release notes list it as "plugin slash commands that can ask you to pick an option", and the plugin docs show how it works.
A plugin command calls ask_choice with a question and a list of choices. How you see it depends on where you are:
| Where you are | What you see |
|---|---|
| Telegram, Discord, Slack | A card with one button per choice, plus Other |
| Other messaging platforms | A numbered list. Reply with the number or the label. |
| Classic CLI, TUI, Desktop | The same picker Hermes uses when the agent asks you a question |
You can also type the value or label instead of clicking. If you cancel, type something that isn't a choice, or let the question time out, the command gets nothing back and should handle that.
The example in the docs is /delegation, from the official advanced-delegation plugin. Run it with no arguments and it asks which delegation mode you want: Default, Model routes or Profiles. Run /delegation default and it skips the question. That plugin landed alongside v0.22.5's routing of subagents by model map or profile.
Want your own? Quick commands in config.yaml map a short slash command to a shell command or another slash command, no plugin needed. For full commands with pickers, the plugin route is register_command plus ask_choice.
Slash commands for plans and imports
Two commands tie into the other Hermes changes this month. /usage and /subscription are how you keep an eye on the new daily and weekly windows, covered in my guide to Hermes Agent usage limits. And if you've just moved over and want the session commands to feel familiar, my guide on how to migrate OpenClaw to Hermes explains why /new and /reset replace OpenClaw's timed session resets.
Picking the model behind /model matters as much as the command, so check Goldie Bench for real agent results. The Agent OS guide shows how I run several profiles side by side. Prefer to set it up with me? Grab a free AI strategy session.
FAQ
How do I see every Hermes slash command?
Type /help. It groups commands by category, and /help skills lists every skill command. On messaging platforms, /commands pages through all commands and skills. /palette or Ctrl+P opens a searchable list in the CLI.
Do slash commands work in Telegram and Discord?
Most do. The reference lists which are CLI-only, like /worktree, /handoff and /copy, and which are messaging-only, like /sethome, /approve and /restart. In Slack threads, use the ! prefix.
What's the difference between /queue and /steer?
/queue holds your prompt for the next turn. /steer slips a note into the current turn after the next tool call, without starting a new one.
Is /stop the same as starting a new session?
No. /stop kills running background processes, and on messaging it also interrupts the agent. Your session and history stay. /new or /reset starts a fresh session.
Can I make my own slash commands?
Yes. Quick commands in config.yaml cover simple shortcuts, every installed skill becomes a command, and plugins can register full commands that ask you to pick an option.
Ready to run Hermes Agent slash commands like a pro? Join the AI Profit Boardroom for the Agent OS and daily Hermes tutorials, or book a free AI strategy session and we'll build your setup together.
