Skip to content
ML2
Multi-Step Agent SetupsTool#Plugin / extension#Open source

Workflow Record: teach Hermes Agent a browser chore by doing it once

Added NewCopyable prompt below

Published Last verified Released

Our verdict

A quick way to automate a repeat browser chore (an order, a portal login and status check, a form you fill every week) without writing a long prompt: you do it once in Chrome while hermes record watches, then /learn writes a skill that replays it with new details. Passwords, one-time codes and card fields are masked inside the page and filled at replay time from Hermes' browser vault or an env var. It is a Hermes Labs plugin, which Nous Research says works but does not support or maintain, so a Hermes update can break it. It records one tab only, misses scrolling, drag and drop, uploads and iframes, and masks only password-type fields, so check the JSON before you run /learn. It is days old: so far there are demos (including a pizza order), not long-term user reports.

Who it's for
Hermes Agent users who repeat the same browser task and would rather show it once than describe it
Leverage
One demonstration becomes a reusable, parameterised browser skill
Effort
Medium
Date added

What it does

Workflow Record (NousResearch/hermes-labs-plugins, MIT, community tier in the Hermes plugin catalog) adds one CLI command and one skill. hermes record attaches to a Chromium-family browser (Chrome, Chromium, Brave, Edge) over the DevTools protocol, by default at 127.0.0.1:9222, and saves clicks (CSS selector plus visible label), final field values, Enter presses and page navigations as JSON in <HERMES_HOME>/recordings until you press Ctrl-C. Recordings stay on your machine until you run /learn on one, which hands it to your active model. The bundled workflow-record:replay-recording skill tells /learn to rebuild the steps in plain terms, treat recorded page text as untrusted data, map each {SECRET:*} placeholder to browser_vault or a named env var, and replay with Hermes' browser tools, using the recorded selectors only as hints. A --manual mode lets you type the steps instead (for Firefox, Safari or sensitive flows). It came out of hermes-agent PR #72199 and was added to the plugin catalog on 21 October 2026.

How builders use it

  • Install it with hermes plugins install workflow-record.
  • Start a browser Hermes can reach (/browser connect in a chat, or Chrome with --remote-debugging-port=9222), close other tabs, open the start page and run hermes record --slug <name>.
  • Do the task once, press Tab out of the last field, then Ctrl-C.
  • Open the JSON and remove anything sensitive (an API key in a plain text field, a token in a URL), then run the /learn line it prints.
  • Use the new skill with fresh details, for example: log into the shop and check my order status.

What to expect

A Hermes skill that repeats a browser workflow you demonstrated, with secrets kept out of the recording and the chat.

Copy this: Record, review, learn

# Official source: github.com/NousResearch/hermes-labs-plugins/tree/main/workflow-record
# Labs plugin: works, but unsupported by Nous Research.
hermes plugins install workflow-record

# Start a browser Hermes can reach over CDP (or use /browser connect in a chat):
google-chrome --remote-debugging-port=9222 --user-data-dir=/tmp/hermes-rec

# Close other tabs, open the start page, then:
hermes record --slug shop-login
# ...do the task, press Tab, then Ctrl-C

hermes record --list     # find the file; review the JSON first

# In a Hermes chat:
/learn the browser workflow recorded in ~/.hermes/recordings/<file>.json (follow the workflow-record:replay-recording skill)

# Firefox/Safari or sensitive flows: hermes record --manual

Related discoveries

See the newest picks →
  • ToolMulti-Step Agent Setups

    browser-use

    Open-source library for agents that use the browser.

    Effort: Medium#Open source

  • ToolMulti-Step Agent Setups

    Stagehand

    Mix A.I. actions with code for more reliable web automation.

    Effort: Medium#Open source

Want more like this?

Practical discoveries, without the noise.

Get the newsletterMore in Multi-Step Agent Setups