Hermes /learn: Give It a Folder, It Generates a Skill — AI That Learns How to Work

Hermes Agent (by Nous Research) now has a /learn command that reads a folder of documents or a web URL and automatically generates a reusable skill card — no vector databases, no RAG setup, no fine-tuning required.

💡 What You Will Learn

Hermes Agent (by Nous Research) now has a /learn command that reads a folder of documents or a web URL and automatically generates a reusable skill card — no vector databases, no RAG setup, no fine-tu

AI Agent has an awkward reality: you ask it to do a job, it does it well, but next time you mention the same thing, it asks you from scratch again. This is fundamentally because the Agent has no mechanism to 'save its work experience.'

Hermes Agent (Nous Research's open-source Agent framework) recently introduced the /learn command. You throw a folder or a web link at it, it reads through everything, and automatically generates a 'skill card.' Next time you need something similar, it follows the workflow directly — no need to teach it step by step again.

Use cases highlighted:

1. Company expense policy PDF: The Agent reads the PDF, extracts key points — submission portal, invoice requirements, approval thresholds (under 10k = department level, over 10k = VP level), monthly deadline of the 15th — and generates an 'Expense Reimbursement' skill card. Next time you say 'submit a reimbursement for last week's business trip to Shenzhen,' it follows the process without repeating instructions.

2. Operations SOP folder: /learn a folder of SOP documents (Tmall listing rules, Douyin livestream procedures, return processing standards). The Agent reads all files and generates a 'Tmall Listing' skill card that knows image specs (800x800 white background), title limits (30 chars), SKU naming rules, and forbidden keywords.

3. Personal trainer workout PDF: /learn reads the training plan PDF, memorizes the exercise combinations for each body part, sets/reps, and rest intervals. At the gym entrance, just say 'leg day today' and it tells you the exact routine.

4. Travel guide webpage: /learn a Xiaohongshu travel guide URL, the Agent extracts daily routes, restaurant recommendations, and pitfalls. Next time you're at Chunxi Road, ask 'what's good nearby' and it answers from the learned guide.

The most interesting part: /learn uses zero new infrastructure. No vector databases, no RAG, no fine-tuning. The Agent uses its own existing toolset — read_file, search_files, web_extract — to consume the source material, then writes a plain markdown skill card based on its understanding. The skill card lives locally, can be shared with colleagues, and can be auto-cleaned. Zero additional cost.

The real vision: company SOPs that used to be 'for humans only' can now become 'capability packs' that Agents also understand. Onboard a new colleague? Give them a PDF and run /learn for the Agent — human and machine get trained simultaneously.