Hermes Agent adds /learn command to turn docs and workflows into reusable skills

Nous Research has rolled out a handy new feature for Hermes Agent: a single /learn command that turns codebases, documentation, or recent terminal workflows into standardized, reusable skills on the fly.

Anyone who uses AI agents knows the frustration of walking an assistant through a tricky task, only to have it forget everything the second you start a clean session. This command aims to fix that retention problem without adding unnecessary complexity.

What it can do

The command is completely open-ended. You can point it at a local directory to extract an authentication flow, or hand it a documentation URL so it can parse the API endpoints on its own.

Even better, if you just guided the agent through an intricate debugging session, you can simply type /learn followed by a request to save what it just did. The agent will inspect its own conversation history and distill the workflow into a permanent skill.

Everything gets written to a standardized SKILL.md file, strictly formatted with short descriptions, clean sections, and associated scripts stored in a dedicated folder for future runs.

How it works under the hood

The engineering choice here is what makes it fun. Rather than building a bulky background distillation engine, the team leaned entirely into an engine-free, agent-driven architecture.

An earlier pull request actually tried using a dedicated backend parser for local directories, but it was tossed out because it was too rigid. Instead, the new /learn command simply constructs a standards-guided prompt and injects it into the agent's normal turn queue.

The agent then completes the task using the everyday tools it already possesses: file readers for directories, web scrapers for URLs, and session memory for past actions. Because there is no host-side ingestion step, it behaves identically whether you run it locally, inside Docker, or over a remote terminal.

How to take it for a spin

You can invoke the command across virtually all Hermes Agent interfaces. In the CLI or TUI, just type /learn with your argument to get started.

If you prefer the web dashboard, the Skills page now includes a dedicated Learn a skill button. Clicking it opens a form where you can paste a directory, drop a link, or write custom notes, which then fires the task straight into your active terminal.

Why it matters

Most agent frameworks treat skills as static plugins that humans have to code by hand. By borrowing from the record-and-replay playbook, Nous Research lets the agent turn its own live sessions into permanent modular capabilities.

It is a refreshingly lightweight pattern that avoids backend bloat while making personal agents meaningfully more capable over time. If you use Hermes Agent to manage code or servers, this is an easy way to build a tailored skill library with zero friction.