Field notes · AI agents, Hermes & Claude Code By Julian Goldie

Hermes Agent For Content Creation: How I'd Set Up A Draft-Only Content Studio In 9 Steps

The exact setup order, commands and config keys for a Hermes content studio that researches, drafts and repurposes, then waits for you to publish.

Get The AI Profit Stack Join AIPB →
  • 01 1,000+ done-for-you AI agent workflows
  • 02 5 live coaching calls / week with me
  • 03 7-day refund, cancel anytime
  • 04 3,400+ AI operators inside

Hermes Agent for content creation works best as a small studio you build once: a house-style skill, a research step, a drafting step, a repurposing step and a review channel.

Below is the exact order I'd set it up in, with the commands and config keys from the official docs.

Every tool and plugin named here exists in the v0.23.0 docs or the plugin catalog, with its tier.

Want a second pair of eyes on your content plan? Book a free SEO strategy session with my team.

Step 1: Give Content Its Own Profile

Start with a separate profile so the studio's skills and memory never mix with your other agents.

hermes profile create studio --description "Writes drafts in our house style. Never publishes."

The profiles docs show that each profile gets its own command alias, so studio opens this one directly.

If you already run a marketing profile, keep it separate; my Hermes marketing desk walkthrough covers campaigns, while this page is about making the pieces themselves.

Step 2: Capture Your House Style With /learn

Don't hand-write the style guide.

Feed Hermes the material you already have and let /learn turn it into a skill.

/learn ~/content/style-guide.pdf, focus on tone, banned words, headline rules and sign-offs
/learn https://yoursite.com/blog/your-best-post

The skills docs say /learn accepts a local file, a docs URL, pasted notes or a workflow you just walked through.

The finished skill lands in ~/.hermes/skills/.

The same page warns that the agent can change or delete any skill, so commit a copy to git.

Then add one line to the skill by hand: "Every output is a draft for review."

Step 3: Set The Voice Layer In SOUL.md

A skill loads when the task needs it.

SOUL.md loads every time.

The personality docs call it the agent's primary identity, slot one in the system prompt, stored in HERMES_HOME.

Keep it short: who the studio writes for, and the one sentence of tone you never want broken.

Put project rules, such as folder names and file formats, in an AGENTS.md inside your content folder; the context files page explains the load order.

Step 4: Switch On The Research Tools

Research is two core tools in the tools reference.

  • web_search returns five results by default, and you can raise the limit up to 100.
  • web_extract pulls clean page text, works on PDF links and takes up to five URLs per call.
  • x_search is optional, read-only and off by default; it needs xAI credentials and never posts.

Pick your search backend in hermes tools.

My rule for the research step: every claim in a draft carries the link it came from.

Step 5: Build The Drafting And Repurposing Skills

The skills catalog already ships the two I lean on.

  • youtube-content turns a video transcript into a summary, a thread or a blog post.
  • humanizer strips AI-isms and adds real voice, so run it as the last pass on every draft.

Chain them in one prompt: transcript in, blog draft out, thread draft out, newsletter section out, each saved as its own file.

For a one-off landing page or slide mock-up, the bundled claude-design skill builds a single HTML artifact.

🔥 Want the house-style skill and the repurposing prompts already built? Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS, four coaching calls a week and 3,400+ members building content systems like this one. → Join the AI Profit Boardroom

Step 6: Add Media Tools One At A Time

Only install what a real job needs.

NeedToolTierSetup
Imagesimage_generate via FAL.aiCorePick one of 11 models in hermes tools
Video clipsvideo_generate in the video_gen toolsetCore, opt-in--toolsets video_gen or hermes tools → Video Generation
Voice-overtext_to_speechCoreEleven providers; Edge TTS is the default and needs no key
Music bedsElevenLabs MusicOfficial pluginhermes plugins install elevenlabs-music + ELEVENLABS_API_KEY
Sound effectsElevenLabs Sound EffectsOfficial pluginhermes plugins install elevenlabs-sound-effects
3D modelsMeshyOfficial pluginMESHY_API_KEY + Node.js 18+

Sources: the image generation page, the TTS page, the video_gen section and the plugin folders for ElevenLabs Music, Sound Effects and Meshy.

Two settings matter here.

The image and video model is yours to set; the agent can't switch it mid-job.

Meshy sends every credit-spending call through Hermes's approval prompt with an estimated cost, so leave that on.

Step 7: Connect The Design And Publishing Apps

These are Nous-approved MCP entries from the optional-mcps catalog, each installed by name.

hermes mcp install canva
hermes mcp login canva
  • Canva: create, search and manage designs.
  • Gamma: generate and edit presentations, docs and sites.
  • Figma: Figma's official remote server.
  • Notion: pages and databases, handy as the content calendar.
  • WordPress.com: posts, pages and drafts; this is WordPress.com only, not a self-hosted site.
  • Webflow: sites, CMS collections and pages.

Google Docs has no MCP entry.

It works through the bundled google-workspace skill, which covers Gmail, Calendar, Drive, Docs and Sheets.

Step 8: Lock Publishing Behind "Draft Only"

This is the step people skip, and it's the one that matters.

Hermes approvals cover dangerous shell commands, per the security docs.

There's no built-in step that holds a post before it goes out.

So do three things.

  1. Write "Draft only. Do not publish, schedule or send." into every job prompt.
  2. Mark each publishing server untrusted in config.yaml, which per the MCP config reference makes every write tool ask first.
  3. Keep approvals.cron_mode: deny, the default, so a scheduled run can't push past a prompt.
mcp_servers:
  wordpress-com:
    trust: untrusted
  postel:
    trust: untrusted

That second entry matters: the catalog says the Postel sign-in replaces the package's untrusted setting, so add it back yourself.

Step 9: Route Drafts To A Review Chat

Pick the one app you read every day from the messaging gateway list: Telegram, Slack, Discord, WhatsApp, Signal, email and more.

Every job delivers there, you read, you edit, you publish.

X Chat is an official plugin too, but it's a private-message gateway for talking to Hermes in X's encrypted DMs.

It does not post to your timeline.

Posting to X goes through the xurl skill, or the community Postel and AdaptlyPost plugins, and only after you've read the draft.

Schedule The Content Calendar With Cron

Now make it run without you asking.

Every Tuesday at 9am, take the newest transcript in ~/content/inbox,
write a blog draft, a 6-post thread and a newsletter section
using the house-style skill, then run humanizer on all three.
Save them to ~/content/drafts/ and send me the file list on Telegram.
Draft only. Do not publish, schedule or send.

The cron docs accept plain language like this, attach skills to the job and deliver to the chat, local files or a platform target.

You can also pin one model to one job with hermes cron edit <job_id> --provider <provider> --model <model>.

Give a heavy long-form job a higher thinking level with --reasoning-effort high, and keep the quick caption job low.

When a job should fire on an event instead of a clock, a webhook route with cron_job set runs that existing job the moment something lands; the webhooks page calls these event-triggered cron jobs.

A form submission or a new file in your storage service can be the trigger, as long as it can send a webhook.

Split Big Pieces With Subagents

A long guide is really four jobs: research, outline, draft and edit.

The delegate_task tool in the delegation docs spawns child agents with their own fresh context, and only each child's final summary returns to the parent.

That keeps the drafting agent's context clean of forty pages of raw research.

Delegate three researchers: one on competitor articles, one on
recent questions people ask, one on official stats with links.
Wait for all three, then write the outline using the house-style skill.

You can point subagents at a lighter model under delegation.model in config.yaml, while the main agent keeps the stronger one for the final draft.

If you leave it blank, children use the same model as the parent.

📞 Want help deciding which pieces should rank in search? Book a free SEO strategy session and my team will map your drafts to the keywords worth chasing. → Grab a free strategy call

Worked Example: A Bakery's Weekly Content Run

A two-person bakery records one short video a week about a new bake.

  1. The owner drops the video file into the inbox folder.
  2. The Tuesday cron job reads the transcript and writes a blog recipe post, a thread and a newsletter blurb in the bakery's voice.
  3. image_generate drafts two header images from the recipe title.
  4. text_to_speech reads the newsletter blurb as a short audio clip.
  5. Everything lands in Telegram as files, and the owner posts what she likes.

No step in that list publishes anything.

If she wants a short jingle under the video, ElevenLabs Music can write a ten-second instrumental bed from a description, and the file arrives in the same chat.

The cost of that track goes to her own ElevenLabs account, which needs a paid ElevenLabs plan according to the plugin's catalog entry.

Worked Example: An Agency Making Client Decks

A small design agency turns each client's monthly notes into a short deck.

  1. A Notion database holds the notes, one row per client.
  2. A cron job reads new rows and drafts the outline with the client's style skill.
  3. Gamma builds the draft presentation, or Canva builds the social graphics.
  4. The account lead reviews the link in Slack before anything reaches the client.

One profile per client keeps each house style apart.

For a product client, Meshy can turn a product photo into a rough 3D model for a spinning mock-up, and the approval prompt shows the estimated credit spend before each call.

The team finishes the deck in Gamma by hand; Hermes only prepares the first version.

Gotchas I'd Check Before Going Live

  • Community video editing: the hermes-video-editor plugin is local FFmpeg editing, but its licence is PolyForm Noncommercial, so skip it for paid client work.
  • Silent spending: the community acedatacloud-media plugin says each generate call spends credits with no Hermes approval prompt.
  • AdaptlyPost: it marks its own server untrusted, so Hermes asks before every write tool and refuses them in unattended runs.
  • Long songs: ElevenLabs Music finishes long tracks in the background and sends them to the chat, so don't chain a step that expects the file instantly.
  • Blocked commands in cron: with cron_mode: deny, a scheduled run that hits a risky shell command is stopped and the agent has to find another route, so test new jobs by hand first with hermes cron run.
  • Figma sign-in: the Figma entry needs its own OAuth login through hermes mcp login figma before its tools load.
  • Version: the media plugins arrived with v0.23.0, so update before installing.

Want more add-ons? My best Hermes Agent plugins list covers the wider catalog.

🔥 Build the whole content studio alongside other operators Inside the AI Profit Boardroom you get the installable Agent OS, video trainings on skills and cron, four coaching calls a week and 3,400+ members to compare setups with. → Join the AI Profit Boardroom

Also On Our Network

FAQ: Hermes Agent For Content Creation

Can Hermes publish posts for me automatically?

It can through the xurl skill or community plugins like Postel and AdaptlyPost, but there is no built-in hold before a post goes out. Write "draft only" into every job and mark publishing servers trust: untrusted so write tools ask first.

How do I teach Hermes my house style?

Run /learn on your style guide, a PDF or your best posts, and Hermes writes a skill into ~/.hermes/skills/. Keep a copy in git, because the agent can edit or delete skills.

Does Hermes have a Google Docs connector?

No. Google Docs works through the bundled google-workspace skill, which also covers Gmail, Calendar, Drive and Sheets.

Which media plugins are official?

ElevenLabs Music, ElevenLabs Sound Effects and Meshy are official Nous plugins in the v0.23.0 catalog. Image generation, video generation and text-to-speech are core tools.

Is X Chat a way to post on X?

No. X Chat is a private-message gateway for talking to Hermes in encrypted X DMs. Posting goes through the xurl skill or the community Postel and AdaptlyPost plugins.

📺 Video notes + links to the tools 👉

🎥 Learn how I make these videos 👉

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

About Julian

I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.

I help business owners scale with AI agents, automation and SEO.

  • I've built a 7-figure SEO and link building agency, Goldie Agency, with a team of 70+ SEO experts.
  • I've grown a YouTube channel to 400,000+ subscribers.
  • I wrote "Link Building Mastery".
  • My Udemy courses have taught over 29,000 students.

→ Get my best AI training inside the AI Profit Boardroom

Set up the profile and the house-style skill today, then book a free SEO strategy session so your drafts target the right searches, and let Hermes Agent for content creation handle the rest of the drafting.

§ Field reports

Real wins from inside the AI Profit Boardroom

See all 3,400+ members →
AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot

Scroll sideways →

⚑ Next step

Ready to run AI agents that actually do the work?

Join 3,400+ operators inside the AI Profit Boardroom. Get 1,000+ plug-and-play AI agent workflows, 5 coaching calls a week, and a community that holds you accountable.

Join The AI Agent Community →

7-day no-questions refund · Cancel anytime

Found this useful? Get more of this site in your Google.

One tap adds us as a preferred source — Google then shows you our posts more often in Search, Top Stories, and AI Overviews.

← Back to all posts