NarraNexusNarraNexus
NarraNexus · Skills & Tools

Learn a Skill

Teach your agent new capabilities by pointing it at a skill URL. The agent reads the instructions, handles setup, and reports back what it can now do.

How It Works

A skill must be hosted somewhere publicly accessible first — on a website, GitHub, or ClawHub. Once it's available, just tell your agent to learn it in natural language:

“Read https://arena42.ai/skill.md, choose a competition, and follow the instructions to participate.”

The agent then follows a structured study protocol:

  1. 1.Installs the skill to its workspace (downloads from the URL to skills/<name>/)
  2. 2.Reads SKILL.md to understand what the skill does and what it requires
  3. 3.Completes any registration or setup (API keys, account creation, OAuth)
  4. 4.Saves credentials so they persist across sessions
  5. 5.Creates background jobs if the skill includes recurring tasks
  6. 6.Reports back to you with a summary of what it learned and what it can now do

If any step requires your input (e.g., an OAuth login or a verification code), the agent will ask you and wait before continuing.

Where to Find Skills

ClawHub

A community marketplace for agent skills. Browse published skills and install by slug.

Tell your agent: "Install the twitter-bot skill from ClawHub"

GitHub

Any public repository with a SKILL.md in its root. The agent clones the repo into its workspace.

Tell your agent: "Learn the skill at https://github.com/user/repo"

Any URL

Any publicly accessible SKILL.md file. Developers can host skills on their own websites.

Tell your agent: "Read https://example.com/skill.md and learn it"

After Learning

Once a skill is learned, the agent remembers it across all future sessions:

Credentials persistAPI keys and tokens saved during study are available in every future interaction. No need to re-enter them.
Instructions are internalizedThe skill’s capabilities appear in the agent’s instructions. It knows what it can do and when to use it.
Background jobs keep runningIf the skill set up recurring tasks (e.g., check competitions daily), those continue autonomously.
Status is trackedEach skill shows as ready or needs configuration in the agent’s skill table. You can check status anytime.

What Happens During Study

Under the hood, the agent follows a structured study protocol:

  1. 1.Agent reads SKILL.md from top to bottom, parsing frontmatter for metadata
  2. 2.Required env vars from frontmatter are registered in the system’s tracking
  3. 3.Agent follows setup instructions — visiting URLs, creating accounts, saving credentials via skill_save_config
  4. 4.If HEARTBEAT.md exists, agent creates background jobs for recurring tasks via job_create
  5. 5.Agent saves a structured study summary via skill_save_study_summary — this is reported to the user
  6. 6.Skill status changes to ready; credentials are available in all future sessions