> Agent-readable docs index: /llms.txt. Download /docs.zip to grep all markdown files locally.

---
title: Skill Discovery
description: Agent-skills spec and .well-known discovery for AI agents.
icon: radar
---

# Skill Discovery

Holocron implements the **agent-skills 0.2.0 discovery spec** (RFC 8615 extension) and a legacy format for backward compatibility. This lets AI agents discover your docs as a "skill" they can load.

## Discovery endpoints

| Endpoint                                    | Format              |
| ------------------------------------------- | ------------------- |
| `/.well-known/agent-skills/index.json`      | v0.2.0 JSON index   |
| `/.well-known/agent-skills/<name>/SKILL.md` | Skill markdown file |
| `/.well-known/skills/index.json`            | Legacy JSON index   |
| `/.well-known/skills/<name>/SKILL.md`       | Legacy skill file   |

The `<name>` is derived from your site's `name` field in `docs.json` (lowercased, kebab-cased).

## SKILL.md content

The generated `SKILL.md` includes:

* Frontmatter with `name` and `description` (from your config)
* Instructions for using `curl` to fetch `/docs.zip` or `/llms.txt`
* Links to individual page `.md` routes

## How agents use this

1. Agent fetches `/.well-known/agent-skills/index.json`
2. Discovers available skills with names and descriptions
3. Fetches the `SKILL.md` for a relevant skill
4. Follows instructions inside (usually downloading `/docs.zip`)

## Customizing the description

The skill description comes from the `description` field in your `docs.json`:

```json
{
  "name": "My SDK",
  "description": "SDK for building widgets with the Acme API"
}
```

If no description is set, Holocron generates a default from the site name.


---

*Powered by [holocron.so](https://holocron.so)*