agentos.guide › blog

Hermes Agent Inbox: Five Prompts That Turn The Read-Only Status View Into A Daily Routine In Your Agent OS

By Julian Goldie · 8 November 2026 · agentos.guide

The Hermes Agent inbox gives your Agent OS a morning check-in: one read-only snapshot of every job, delegation and session the agent still has open, which you can hand straight to the agent to sort.

It is the inbox labs plugin from Nous Research, version 0.1.0, community tier, and it is not an email client.

This guide gives you a quick start, five prompts to paste, and a daily routine built around them.

If you would like my team to help plan the wider setup, book a free SEO strategy session.

Quick Start In Three Commands

hermes plugins install inbox
hermes inbox
hermes inbox --json

The first adds the plugin from the catalog, the second prints the text view, and the third prints the same snapshot as JSON.

For another profile, install the plugin inside that profile as well, then run hermes -p work inbox.

There are no settings to change; the plugin README lists none.

The plugin adds no model tools, so the agent reaches it the plain way: by running the command in its terminal.

Three checks before the first prompt

1. If the installer asked whether to enable the plugin, make sure you answered yes; hermes plugins list shows its status, per the plugins guide.
2. Run hermes inbox yourself once, so you know what a normal morning looks like before the agent starts summarising it.
3. Write down every profile you use, because each one needs its own install and its own check.

Two empty headings, Needs attention and In progress, are the all-clear signal you are looking for.

If a Not read line appears at the bottom, fix that first: one of the stores could not be opened, and the prompts below will be working from a partial picture.

What Your Agent Will Be Reading

The top two blocks always print; the rest vanish when empty.

The catalog entry confirms it opens its stores read-only and sends nothing anywhere, so these prompts are safe to run at any time.

🔥 Want the full Agent OS with Hermes already wired in?

Inside the AI Profit Boardroom you get the installable Agent OS, step-by-step video trainings, four coaching calls a week and 3,400+ members sharing the prompts they run each morning.

→ Join the AI Profit Boardroom

Prompt 1: The Morning Triage

Run `hermes inbox --json` in the terminal.
First check the sources field. If any store is "absent" or
"unreadable", tell me which one before anything else.
Then list every item under Needs attention, one line each:
what it is, why it is flagged, and the single next step.
Do not change, rerun or delete anything. Report only.

The sources check matters: an empty block from a store that could not be read is not good news.

Prompt 2: Diagnose A Failed Cron Job Without Touching It

The inbox shows cron job '[job name]' under Needs attention.
Run `hermes cron list` and find that job.
Explain in plain English why its last run failed or why it is paused.
Suggest the smallest fix, and the exact command to rerun it once.
Wait for my yes before you edit, resume or run anything.

The cron docs cover listing, pausing, resuming, editing and triggering jobs, so the agent has everything it needs once you approve.

For the scheduling side in depth, see our guide to the Hermes Agent cron job.

Prompt 3: Chase Results That Never Arrived

From `hermes inbox --json`, list every delegation under
"Finished, result not yet delivered" and every stalled delegation.
For each one, tell me which conversation it belongs to and
what it was asked to do, so I can open that chat and collect it.

This catches delegated research that finished overnight but never reached your chat.

The agent cannot deliver those results for you from here, because the inbox only reads; the point is to tell you exactly which chat to reopen.

Prompt 4: Sweep Every Profile

I use these Hermes profiles: [default, work, clients].
For each one, run `hermes -p [profile] inbox --json`.
If the inbox command is missing in a profile, say so and stop for that profile.
Give me one combined table: profile, block, item, next step.
Note that kanban is shared across profiles, so list each kanban card once.

The kanban note is there because the README says Hermes shares one board across profiles.

📞 Want a hand mapping your agents and profiles?

Book a free SEO strategy session and my team will review your workflows, your content and where each profile fits.

→ Book a free SEO strategy session

Prompt 5: Build A Script-Only Morning Report

Write a script at ~/.hermes/scripts/morning-inbox.sh that runs `hermes inbox`.
Then create a no-agent cron job named morning-inbox that runs it
every day at 07:30 and delivers to telegram.
Show me the script and the cron command before you create anything.
After I approve, run the job once and paste what it delivered.

Cron's no-agent mode runs a script and delivers its output as it is, with no model call, and the script must sit inside the Hermes scripts folder.

Pairing it with the inbox is my own combination of two documented features, which is why the prompt asks for a test run.

A Daily Routine For A Small Business

Here is how a two-person bookkeeping practice could run it.

1. At 07:30 the script-only job sends the inbox to the shared Telegram chat.
2. If Needs attention has anything in it, one partner pastes Prompt 1 into the Agent OS.
3. For a failed cron job, they run Prompt 2 and approve the fix.
4. For waiting results, they run Prompt 3 and open the listed chats.
5. On Fridays they run Prompt 4 across both client profiles.

A salon or a plumbing firm can use the same loop with one profile and fewer steps.

The salon version

1. An overnight cron job writes the next day's appointment reminders into a file as drafts.
2. The morning report arrives on the owner's phone before opening time.
3. If the reminder job failed, the owner opens the Agent OS and pastes Prompt 2 with the job name filled in.
4. The agent explains the failure and proposes a fix, and the owner approves it.
5. Reception sends the reminders by hand once the drafts are there.

Keep the draft-only rule in the job itself, because Hermes approvals are built for risky commands rather than outgoing messages.

Limits To Brief Your Agent On

• Pending approvals and clarify prompts never appear; answer them in the chat that asked.
• Each run covers one profile.
• Processes inside a sandbox backend show as sandboxed, because their PIDs cannot be probed.
• It is a labs plugin: Nous built it, does not maintain it, and makes no promise of fixes, per the labs repository.
• Cron jobs deny dangerous commands by default under approvals.cron_mode, per the security guide.

Searching For An Email Inbox Instead?

That is a separate feature.

Hermes reads and answers mail through its Email gateway over IMAP and SMTP, replying in the same thread.

Give it a dedicated account and set EMAIL_ALLOWED_USERS, and write "draft only" into any job that sends.

🔥 Get the prompts members run with Hermes every day.

Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS and four live coaching calls a week.

→ Join the AI Profit Boardroom

Where The Plugin Came From

It started as core PR #81507 and was moved into the labs repo instead of core.

It reached the catalog with the v0.23.0 release on 5 November 2026.

Re-run Prompt 1 after each Hermes update to make sure it still works.

What Not To Ask The Agent

Some requests sound reasonable but will not work, so leave them out of your prompts.

• Do not ask it to list pending approvals from the inbox; they sit in the memory of the chat that asked and cannot be read from outside.
• Do not ask the inbox to fix, rerun or clear anything; it never writes to the stores it reads, so any fix has to go through cron or the original chat.
• Do not ask it to check a profile where the plugin is not installed; install it there first.
• Do not ask it for email; that belongs to the gateway covered above.

Briefing the agent on these four points up front avoids a round of confused replies.

FAQ

Can my agent read the Hermes Agent inbox itself?

Yes. Ask it to run hermes inbox --json in its terminal. The plugin adds no model tools, so the terminal command is the way in.

Will the triage prompt change anything?

No. The inbox is read-only, and the prompt tells the agent to report only. Fixes wait for your approval in Prompt 2.

How do I check several profiles?

Install the plugin in each profile, then run hermes -p <profile> inbox for each one, as Prompt 4 does.

Can the morning report run without a model?

Yes. A no-agent cron job runs a script and delivers its output with no model call.

Where do I approve risky commands?

In the chat that asked. Approvals are not shown in the inbox.

Final word on Hermes Agent Inbox

Paste Prompt 1 tomorrow morning, and book a free SEO strategy session or join the AI Profit Boardroom when you want help building your routine around the Hermes Agent inbox.

Related guides on agentos.guide

→ Hermes Agent Cron Job: Schedule, Stop And Recover Jobs In Your Agent OS With 5 Copy-Paste Prompts

→ Hermes Agent Bubblewrap: Sandbox Commands In Your Agent OS

→ Hermes Agent Hooks: Add Guardrails, Turn Blocks And Dashboard Events To Your Agent OS With Five Copy-Paste Prompts

→ Hermes Agent For Business: Start With One Job, Then Grow It With Five Copy-Paste Prompts In Your Agent OS

📺 Video notes + links to the tools 👉

🎥 Learn how I make these videos 👉

🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉

← all guides on agentos.guide