NEW ERA - OS Agent

Discussion & Solutions
1

Hello,

I’ve been following RoboNuggets’ work on agent systems and recently watched the detailed analysis of the new “Agentic OS” standard for Claude 5 models. The ARMS framework (Applications, Routines, Memory, Skills) really resonated with me, and I’d love to implement it in my own workflow.

I tried accessing the guide via the Skool members’ area link mentioned in the video, but I noticed there is a $100 fee for access.

Could you help me get access to the Agentic OS standard package or point me to where I can find it? I searched GitHub but couldn’t find it.

I am particularly interested in:

The router files and workspace structure

The Obsidian vault integration setup

The prompts for the mission control dashboard

Thanks in advance for your help—I’m excited to put this into practice!

2 Likes
2

:bullseye: You’re looking at the wrong link in that video

The $100 Skool buys the community + masterclass + agents-as-a-service course.

The ARMS setup guide is not in there — Jay published it free. In that same video he says it plainly:

“I’ve also published this nine-page PDF guide… you can just grab that in the description below.”

  ┌──────────────────────────────────────────┐
  │  $100 Skool   ▸ community · masterclass  │
  │               ▸ agents-as-a-service      │
  ├──────────────────────────────────────────┤
  │  FREE PDF     ▸ the ARMS setup guide     │  ◀ what you want
  │  (video desc) ▸ feed it to Claude Code   │
  └──────────────────────────────────────────┘

Drop that PDF straight into Claude Code and it walks you through the build.

:wrapped_gift: Everything on your list — free, from him

You asked for Where it already is
:puzzle_piece: Router files + workspace structure github.com/robonuggets/skills — calibrate · personalise · tweak. His note: “No licence file is included. Adapt freely.”
:bullseye: The /goal skill github.com/robonuggets/plan-for-goal — he publishes the install prompt openly
:books: The whole curriculum Agentic AI Foundations — Skills 101 · Memory 101 · Routines · Skill Patterns · Memory Layers · Command Centre
:toolbox: Templates + prompts robonuggets.us/tools — “all free to start”

You said GitHub came up empty — it’s under robonuggets, not the course name.

:wrench: The three pieces that need no course at all

▸ Router + workspace skeleton → the open-source agentic-os skill is the same 4-layer architecture, already written:

  CLAUDE.md          ▸ kernel — identity, routing rules, agent registry
  agents/            ▸ specialist agents (one .md each)
  .claude/commands/  ▸ your slash commands
  scripts/           ▸ scheduled tasks (external cron, not session cron)
  data/              ▸ file-based memory — no database

▸ Obsidian integration → Obsidian is free. Pick a folder as your vault, then open Claude Code inside that folder. That’s the entire integration. The real trick is an index.md at every level so the agent knows where to go before it reads anything.

▸ Skills layer → Claude Desktop → Customize → Skills ships Anthropic’s own, including a skill-creator skill that writes new ones for you.

:mirror: About the dashboard — his own words

“This visual interface captures only around 20 to 30% of the value… the remaining 70% really lies on what’s underneath.”

Build the four layers first. The dashboard is a wrapper you point Claude at afterwards — hand it a screenshot of a layout you like, your vault path, and your skill list.

:warning: The 40% token saving (30k vs 50k) is his own single demo, not an independent benchmark. Treat it as a direction, not a number.

Free guide, free skills, free curriculum — all from the same creator. The paywall was never in front of the thing you wanted.

3 Likes
3

I’ve been running a similar layered setup for a few weeks now, so here’s what actually worked for me beyond the theory — maybe it saves you some trial and error.

Start with a single router file, not all four layers. I kept one CLAUDE.md at the workspace root with exactly three sections: who the agent is, which folder owns which job, and what it must never do (for me: never touch credentials files, never push to remotes). Only when the root file got too long to read at a glance did I split the specialists into an agents/ folder with one short file per role. The routing map really is the whole game — the agent reads the router before anything else, so I keep every entry to one line: name, what it owns, what triggers it.

For the Obsidian part: I just opened my workspace folder as the vault, no plugins. The index.md-in-every-folder trick SRZ mentioned is real — I add one short index.md per project folder listing what’s inside and where its notes live, and the agent suddenly stops guessing paths. I also keep one daily log note where the agent appends what it did each session; that became my “memory” layer without any database at all.

Dashboard: skip building the UI first. I use one missions.md file with a plain table — mission, status, next step, blocked on — and the agent updates it at the end of every session. Ten minutes to set up, and honestly it gives me most of the value a fancy mission control would.

One caution on the token-saving claims: route narrowly and you’ll feel the difference regardless. I tell the router to send file-reading and lookup jobs to the smallest model that can handle them and reserve the big model for planning. That alone cut my usage noticeably before any framework-level optimization did.