Type /goal followed by what you want done and Hermes agent /goal sets a standing goal your agent works toward across turns — after each turn, a separate judge model checks whether the goal is genuinely finished, and if it is not, Hermes carries on by itself.
📺 Watch: Hermes Bot Mode DESTROYS Grok Bot? (FREE!)
🔥 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
That single command has changed how I run agents more than anything else this year. Instead of prompting, waiting, nudging and re-prompting, I hand Hermes a finish line and let it grind toward it whilst I get on with something more useful. In this guide I will break down what the goal loop actually does under the bonnet, walk through every subcommand, compare /goal with /loop, cover mid-loop steering with /subgoal, and flag the Telegram quirk that catches almost everyone out.
What the Hermes agent /goal command actually does
According to the official slash-commands reference, /goal sets a standing goal that Hermes works toward across turns — the docs describe it as their take on the Ralph loop. The clever part is what happens after each turn: an auxiliary judge model reviews the work and decides whether the goal is done. If the judge says no, Hermes auto-continues without you touching the keyboard. And the judge does not simply glance at your original request — it evaluates the original goal plus any attached subgoals before marking anything complete.
Two guardrails keep the goal loop sensible. The turn budget defaults to 20 turns, controlled by the goals.max_turns setting, so a runaway goal cannot eat your whole day. And any real message you send preempts the continuation loop, so a live goal never locks you out of your own agent.
I have been logging my goal runs on Goldie Bench, my testing resource, and one pattern is impossible to miss: goals written like acceptance criteria finish cleanly, whilst goals written like vibes burn the entire budget. Give the judge model something it can actually verify.
Want the exact goal prompts I use for client work — the ones with pass-fail criteria the judge model can tick off? I share them first inside AI Profit Boardroom, along with my full Hermes goal template library. Prefer to talk it through instead? Book a free SEO strategy session and I will map standing goals onto your workflow with you.
Syntax and subcommands: the full /goal walkthrough
The base syntax could not be simpler: /goal followed by the goal text. Something like /goal get every draft in the content folder ready to publish becomes the standing objective, and Hermes keeps working toward it turn after turn until the judge signs it off, the 20-turn budget runs out, or you step in. Once a goal is live, four subcommands control the loop.
/goal status
This shows progress on the current goal. I run it constantly — it is the quickest way to see how the loop is tracking without interrupting anything, and it is the first thing I check before deciding whether a goal needs extra criteria bolted on.
/goal pause
This halts auto-continuation. The goal itself stays in place; Hermes simply stops continuing on its own. I use it when I want to inspect the work mid-run, or when something urgent needs the agent and I do not want the goal loop competing for turns.
/goal resume
The opposite number: /goal resume restarts auto-continuation after a pause. Note that this is not the same as the session-level /resume command — I will come back to that distinction in the steering section, because it trips people up.
/goal clear
This removes the standing goal entirely — no judge, no auto-continuation, clean slate. I clear goals the moment they stop reflecting what I actually want, because a stale goal quietly steers every subsequent turn.
📺 Watch: Hermes Agent: NEW Goals Update is Insane (FREE)!
Hermes agent /goal vs /loop: standing goal or recurring loop?
These two get confused constantly, and the distinction is simple. /goal is a standing goal with a completion judge: it has a finish line, and the loop exists purely to reach it, after which it ends. /loop is the always-on recurring pattern — the agent doing something on repeat with no notion of done — and I have covered that fully in my Hermes /loop guide. Goals end; loops recur.
My rule of thumb: outcomes get a goal, rhythms get a loop. Getting a backlog to zero is a goal, because a judge can verify it happened. Watching an inbox all day is a loop, because it is never finished. And if what you really want is a scheduled pulse rather than either of those, that is the job of the Hermes heartbeat command instead.
📺 Watch: Hermes Agent: NEW Goals Update is Insane (FREE)!
/subgoal and steering the goal mid-loop
/subgoal appends criteria mid-loop, and this is where the Hermes agent /goal pattern goes from neat to genuinely powerful. Because the judge evaluates the original goal plus any attached subgoals before marking completion, you can tighten the definition of done whilst the agent is already running — no restart, no re-explaining the whole task from scratch.
Two documented behaviours make mid-loop steering safe. First, any real user message preempts the continuation loop, so you can interject whenever you like and Hermes deals with you before the goal. Second, goal state survives /resume — close the CLI tonight, resume the session tomorrow, and the standing goal is still standing, judge and all. That combination means you never have to choose between stepping in and keeping the goal alive.
Where /goal works — and the Telegram quirk
The /goal command works in both the CLI and the messaging gateway, so the same standing-goal pattern runs whether you are at your desk or steering an agent from your phone. Since Hermes v0.21 made rosters persistent, running goals through the gateway has felt like proper agent management rather than a novelty.
One heads-up from the community. The widely shared cheat sheet from the unofficial Hermes Release Watch account (5 September 2026) put /goal at the top of its save-this list, and one user reported a Telegram quirk worth knowing about: selecting /goal from the command menu executes it immediately without sending a message, and tapping the button after typing your goal clears the text you just wrote. The workaround is to type the command manually at the end of your message. That is a user report rather than official guidance, and bugs like it belong in the official Discord.
Business uses for standing goals
Where does /goal earn actual money? Outreach follow-ups that continue until every reply is drafted, content refreshes that run until each flagged page is updated, lead-list enrichment that keeps going until the sheet is complete. I keep the full breakdown in my Hermes agent goals guide, and inside my Agent OS setup a standing goal is usually the engine behind each delivery workflow. When a goal completes, I bank whatever worked — my Hermes /learn command guide covers how I handle that side.
| Subcommand | What it does |
|---|---|
| /goal followed by goal text | Sets the standing goal Hermes works toward across turns |
| /goal status | Shows progress on the current goal |
| /goal pause | Halts auto-continuation |
| /goal resume | Restarts auto-continuation after a pause |
| /goal clear | Removes the standing goal |
| /subgoal followed by criteria | Appends criteria mid-loop for the judge to evaluate |
Hermes agent /goal FAQ
How many turns will a goal run for?
The budget defaults to 20 turns via the goals.max_turns setting. In my experience that default is generous for well-scoped goals — if you are hitting the cap regularly, your goal is too big and wants splitting into smaller verifiable pieces.
How does Hermes decide a goal is finished?
An auxiliary judge model evaluates the original goal plus any attached subgoals after each turn. If everything is satisfied, the goal is marked complete; if not, Hermes auto-continues. Write goals the judge can verify and the whole system behaves beautifully.
What happens if I message the agent mid-goal?
Any real user message preempts the continuation loop. Hermes handles your message first, so a running goal never stops you steering, asking questions or redirecting the agent entirely.
Does a goal survive closing the session?
Yes — goal state survives /resume, so resuming a session brings the standing goal back with it. Within a session, /goal resume is the separate subcommand that restarts a paused loop. Two similar names, two different jobs, and that persistence is exactly what makes longer-running goals viable.
My verdict on /goal
/goal is the most useful command in the Hermes toolkit for anyone who wants outcomes rather than conversations. The judge-model loop turns keep-going-until-done from a wish into a mechanism, and the subcommands give you full control without babysitting anything. Start with small verifiable goals, steer with /subgoal, follow the habits in my Hermes agent best practices guide, and it quickly becomes the command you build everything else around.
Standing goals are the first brick; an agent-run business is the building. Join me inside AI Profit Boardroom for my complete standing-goal playbook and client-ready agent workflows — or book a free SEO strategy session and we will plan exactly where /goal fits in your business.
