Type /plan, with or without a task after it, and Hermes agent /plan writes a markdown implementation plan into your active workspace — planning only, nothing gets executed. I rate it as the most quietly important command in the Hermes toolkit, because it forces the agent to put its thinking on disk where I can read it, edit it and sign it off before anything actually happens. Everything below sticks to what the official slash-commands reference documents, layered with how I actually run the command day to day on my own projects.
📺 Watch: Hermes Agent OS Just Changed AI Agents Forever!
🔥 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
What Hermes agent /plan actually does
The documented behaviour is deliberately narrow. Run the command and Hermes writes a markdown implementation plan to .hermes/plans/ in the active workspace — and that is the entire job. No execution, no edits to your project, no actions fired off in the background. Planning and doing are split into two distinct moments, and Hermes agent /plan owns the first one completely.
That separation is the whole point, and it matters more than it looks. The failure mode I see most often with agents is not bad planning — it is invisible planning. An agent that thinks and acts in the same breath gives you nothing to inspect until the work is already done, and by then unwinding a wrong turn costs far more than preventing it would have. With /plan, the reasoning lands as a file first. I read it, I disagree with step four, I rewrite step four, and only then does any execution begin. The command works in the CLI and through the messaging gateway alike, so the same discipline holds whether I am at my desk or steering Hermes from my phone between calls.
Syntax: a task if you have one, silence if you do not
The documented syntax is /plan followed by an optional task. Describe the job after the command and the plan gets written for that job, scoped to your wording rather than the agent's interpretation of a long chat. That is the version I use when I arrive with a clear brief — a migration, a rebuild, a content push — and I want the plan anchored to exactly what I said.
Leave the argument empty, though, and /plan infers the task from the conversation so far. This is the mode I reach for most. I talk a problem through with Hermes the way I would with a colleague — constraints, dead ends, half-decisions — and once the discussion has real shape, a bare /plan turns the whole thread into one structured document. Nothing gets lost between the talking and the doing, because the plan is distilled from the talking. Two modes, one command, and both finish the same way: a markdown file waiting for your eyes, not an agent already halfway through the work.
Want the exact plan-first Hermes workflows I use on client projects every week? Get my Hermes planning playbooks inside AI Profit Lab — and if you would rather map your setup one to one, book a free SEO strategy session and we will build your workflow together.
📺 Watch: Hermes3D Is INSANE: AI Agents Now Have a 3D Office
Where plans live: files you read before anything runs
Every plan is written to .hermes/plans/ inside the active workspace, and that location is a philosophy statement as much as a path. A plan in chat scrollback is a suggestion; a plan in a file is an artefact. You can open it, edit it, version it, share it and keep it sitting next to the project it describes. It is the same file-based thinking I unpacked in my Hermes memory system guide — readable, editable, portable files beat opaque state every single time.
My habit is simple: never let execution start from a plan I have not opened. The file in .hermes/plans/ is markdown, so editing takes seconds — cut the speculative steps, tighten the order, add the constraint the agent could not have known about. And because plans accumulate in the workspace, I get a paper trail of how each project was supposed to unfold, which is worth more than any changelog when I audit weeks later what an agent did and why.
From bundled skill to built-in command
The docs carry a small but telling note: /plan formerly shipped as the bundled plan skill, and it is now built in so that it survives the Telegram and Discord command-menu caps. If you only ever drive Hermes from the CLI, that reads like trivia. If you run it through the messaging gateway, it is the difference between the command dependably being there and it falling off a platform-limited menu because too many skills were competing for the slots.
I read the promotion as a signal of priority: when something graduates from optional skill to built-in command, the developers are telling you it belongs in everyone's core loop. For attribution's sake, the unofficial Hermes Release Watch cheat sheet from 5 September 2026 lists /plan among the built-ins, which matches the official line. And if you like keeping score of what ships when, my Hermes v0.21 write-up shows how I track each release without drowning in changelogs.
📺 Watch: Hermes AI Agents Just Went Portable
The plan-then-execute workflow I actually run
Plan-first discipline is what separates agents you trust from agents that wander — that is the core argument of my Hermes agent best practices guide, and /plan is the command that makes the discipline effortless. My loop is short: run /plan, open the markdown, edit ruthlessly, then instruct Hermes to execute the approved version. The edit step is where the value hides, because a two-minute read regularly catches a wrong assumption that would have cost an hour to unwind.
Two supporting habits round it out. When a planning pattern keeps proving itself, I bank the lesson so it becomes standing behaviour — my /learn command guide covers exactly how I do that. Inside my wider Agent OS setup, /plan acts as the gate every non-trivial task passes through before it touches anything real. I have hammered this loop on Goldie Bench, my own testing rig, and the result is consistent: an edited plan beats a straight-to-execution run on accuracy almost every time.
/plan vs /goal vs /review
Keep the three straight by their position in the timeline. /plan is the before: it writes the implementation route and deliberately executes none of it. /goal sits above the individual task, at the objective level — I break down how I use it in my /goal command guide. /review is the after: the checking pass once work exists, which I cover in my /review command breakdown. Plan the route, hold the destination, inspect the result — three commands, three moments, no overlap.
Documented behaviour at a glance
| Aspect | Documented behaviour |
|---|---|
| Output | A markdown implementation plan |
| Location | .hermes/plans/ in the active workspace |
| Execution | None — planning only |
| Syntax | /plan followed by an optional task |
| Empty argument | Infers the task from the conversation |
| Surfaces | CLI and the messaging gateway |
| History | Formerly the bundled plan skill, now built in |
| Why built in | Survives the Telegram and Discord command-menu caps |
Questions I keep getting about /plan
Does Hermes agent /plan execute any part of the plan?
No. The documented behaviour is planning only: the command writes a markdown implementation plan to .hermes/plans/ and stops there. Execution is a separate decision you take afterwards — ideally once you have read and edited the file rather than waved it through.
What happens if I run /plan without describing a task?
The task argument is optional. Left empty, /plan infers the task from the conversation so far, turning your discussion with Hermes into the brief. It is the fastest way I know to convert a messy thinking session into a clean, ordered document you can act on.
Can I change a plan before anything runs?
Yes, and you should. Plans are plain markdown files sitting in your workspace, so you edit them like any other document. My rule from years of agent work: never approve execution on a plan you have not opened, because the two-minute read is the cheapest insurance you will ever buy.
Does /plan work outside the CLI?
Yes. The docs state it works in the CLI and through the messaging gateway, and because it is now built in rather than a bundled skill, it survives the Telegram and Discord command-menu caps instead of dropping off a crowded platform menu.
My verdict on Hermes agent /plan
It is a small command carrying a big principle: thinking and doing belong in separate steps, with a human read in between. It writes the plan, saves it where you can see it, and leaves the execution decision with you. That ordering — file first, action second — is why my agents ship work I trust rather than surprises I have to debug. If you adopt one habit from this page, make it this: no execution without an opened plan.
If plan-first agents are how you want to work, join me inside AI Profit Lab and take my full Hermes command system — or book a free SEO strategy session and I will show you exactly where /plan fits in your stack.
