8 /learn Superpowers: Skills That Compound Forever

8 /learn Superpowers: Skills That Compound Forever
OWN THE EDGE
July 24, 2026  ·  AI & Automation for Builders

Hi everyone,

Hermes Agent has two commands most people scroll right past — /learn and /goal. Together they solve the three problems that make most AI agents useless after the first session: amnesia, supervision, and portability. This week I built eight real-world examples — from brand slide decks to support tone to deploy procedures — and at the end, showed how to make every single one permanent across any agent you ever spin up.

One table in Supabase. That's all it takes to stop your skills from dying with your session.

🎥 Video Deep Dive

Hermes Agent Skills Tutorial: 8 /learn Superpowers

The Build: /learn teaches Hermes something once — a folder of documents, a URL, a YouTube video, even a conversation you just had — and writes a permanent, reusable skill file. No manual coding. That solves the amnesia problem: no more spending the first 10 minutes of every session re-explaining your business, your voice, or your process. /goal is the other half: you provide a standing objective, and Hermes keeps working turn after turn until that objective is met and proven. Hermes just added completion contracts — you define what "done" looks like, and the agent checks its own work against those criteria before stopping. Not "I think I'm finished" — "here is proof I'm finished."

🔧 8 Real-World /learn Superpowers

1. Brand Slide Deck Factory. /learn from a reference deck — colors, layouts, fonts, everything. Then /goal any PDF into a branded slide deck. Hermes reads the content, iterates through slides, verifies correctness, and outputs the finished PDF. The layout matches the original branding perfectly because it learned the style, not just copied templates.

2. Learn from Any YouTube Video. Point /learn at a tutorial URL. Hermes transcribes it, analyzes the method, and distills the technique into a working skill — not a summary, an executable process. The demo uses a Japanese learning method: /goal then builds a personalized 30-day study plan with daily goals and resources.

3. Meta: Learn How to Write Better Goals. /learn a guide on writing effective /goal conditions. It extracts the formula: measurable end state, proof it happened, constraints that must hold, and a limit so it doesn't run forever. Then "make the deck better" becomes a proper completion contract the agent can verify itself against.

4. Your Writing Voice. Feed it years of your own writing — blog posts, emails, whatever. It doesn't just sound like you. It starts noticing things you never pointed out: every conference you've mentioned, patterns in how you structure arguments. Brand risk is real here — keep a human glancing at anything before it goes out.

5. Support Tone from Real Logs. Point it at actual support conversations, not the policy doc nobody reads. It picks up the tone your best team member uses when de-escalating — the phrases that actually work. Not a script, a calibrated voice.

6. Outreach That Gets Replies. Feed it cold emails that actually generated responses. It drafts new messages that sound like you on your best day — not like a template that landed in spam. The pattern recognition is the unlock: it sees what worked, not what you wish worked.

7. Monthly Reports on Autopilot. Show it one month's report. Next month, it builds the same format from raw numbers. It's copying structure, not judgment — a genuinely off number will still land in the same box unless you teach it to flag outliers. But 90% of the formatting labor disappears.

8. Deploy Procedures as Slash Commands. Walk Hermes through one deploy. Just once. That whole procedure — every check, every command, every verification step — becomes a skill any teammate can invoke. Same way you did it, every time.

🗄️ The Part That Makes Everything Compound: Supabase Persistence

Here's the problem: none of these eight skills travel past your machine. New agent? Doesn't know any of it. New teammate? Back to zero. The fix is the whole reason this video exists: one Supabase table.

Every skill gets written to a persistent skills table using Supabase's official agent skill — the same one that plugs into Hermes natively. Two steps, no custom code, no terminal:

1. /learn the Supabase skill from its docs URL — one command, knows how to work with the database.

2. Connect the Supabase MCP server with a personal access token — paste it when prompted, connection made.

Then one command: "Save all my learned skills into Supabase. One row per skill. Don't skip any." Everything migrates in one continuous atomic operation.

On a brand new agent 30 seconds old: "Connect to Supabase and load all skills from the skills table, especially any I don't already have." It starts answering questions using skills it never actually learned — because they're not stuck on your laptop anymore.

Why It Matters: The three-problem stack is what kills most agent setups. /learn solves amnesia — no more re-explaining yourself every session. /goal with completion contracts solves supervision — the agent proves it's done instead of quitting when it "feels satisfied." Supabase solves portability — skills compound across agents, machines, and teammates instead of resetting. Any one of the eight examples can be handed to /goal the same way. None of it disappears when the session closes.

📰 AI & Automation This Week

🔤 NDTV PROFIT

OpenAI Poaches xAI Veteran Brent Mayo As Sam Altman Heats Up Rivalry With Elon Musk

The AI talent war escalated as OpenAI hired a key engineering leader from xAI, signaling the rivalry between Altman and Musk is moving from rhetoric to direct talent raids. Expect this to accelerate as both companies race toward the next generation of frontier models.

🔤 ANTHROPIC

The Anthropic Economic Index Connector

Anthropic released a connector mapping real-world AI adoption data across industries — showing which sectors are actually deploying AI in production vs. just experimenting. Valuable signal for anyone building automation products or deciding where to focus.

🔤 SEARCH ENGINE JOURNAL

Anthropic's Claude Can Now Watch A Video And Learn Your Job

Claude can now watch screen recordings and learn workflows by observation — see a task performed once, replicate it. This is a direct parallel to the /learn pattern in this week's video: teach once, reuse forever. The industry is converging on skill permanence as the next frontier.

🔤 EGAMERS.IO

WAX Launches n8n-wax-kit, Bringing Blockchain Into No-Code Automation Flows

WAX blockchain released an n8n integration kit, making on-chain actions as simple as dragging nodes in a workflow. No-code automation platforms are becoming the universal interface for everything — APIs, blockchains, AI agents — eliminating the need to write glue code.

🔤 PCMAG UK

Zapier Review: Boost Your Business With Easy Automation

PCMag's latest review highlights how Zapier has evolved beyond simple triggers into a legitimate business automation layer — with AI-powered steps, multi-step logic, and enterprise features. The line between "no-code tool" and "operations platform" continues to blur.

🔤 TRIBUNE ONLINE

How Nigerian Freelancers Make More Money With Make.com Automation

Freelancers are using Make.com to automate client delivery — report generation, data processing, scheduling — effectively productizing their services. The pattern is universal: learn an automation tool once, sell the output indefinitely. Same economics as this week's video: one /learn, infinite output.

▶  Watch all 8 superpowers on YouTube

Happy building,

Derek

How was this week's newsletter?

⭐️⭐️⭐️⭐️⭐️

Reply with feedback or more on YouTube

AI for Non Coders

Sign up AI Automation newsletter