Give your agent the Stele skill

Stele publishes an agent skill — stele-resources — that teaches a coding agent how a Stele-published repository works, before it touches your files. Install it once in a repository and an agent gets the conventions for free: where to put a new artifact, how .stele.yml behaves, which file types render and how, what the build environment supports, and how to wire up Ask.

What the skill teaches

The skill covers the same ground as this user guide, written for an agent rather than a person:

Folder conventions — organising artifacts by team or system, when to create a subfolder, naming and dating files, and how to tell a refresh from a new version.
The .stele.yml reference — every field, what merges between parent and child manifests, and what does not.
Supported formats — Markdown, HTML, and React (TSX/JSX), landing-page resolution, and intra-repo linking.
Build steps — the fixed tool set available at sync time (likec4, d2, dot), and the build:/output: pattern for diagram folders.
The ask: block — the schema, where a corpus is rooted, and where each kind of citation points.

It triggers automatically once installed: an agent that sees a .stele.yml, a .likec4/.d2 file, or a .tsx/.jsx artifact in your repository picks it up without you having to mention Stele by name.

Installing with the skills CLI

The one-line install:

npx skills add https://stele.axelerant.tech/skills/stele-resources.tgz

This downloads the packaged skill and installs it for the agents it detects in your project. Useful flags:

-a claude-code (or another agent name) — install for a specific agent instead of every agent the CLI finds.
-g — install globally, so the skill is available in every repository rather than just this one.
--list — see what the archive contains before installing anything.

Installing by hand

If you'd rather not run the CLI, fetch the archive yourself and untar it into your agent's skills directory:

curl -L -o stele-resources.tgz https://stele.axelerant.tech/skills/stele-resources.tgz
mkdir -p .agents/skills
tar -xzf stele-resources.tgz -C .agents/skills

That leaves a stele-resources/ folder containing SKILL.md and references/. Use .claude/skills/ instead of .agents/skills/ if your agent looks there specifically. Either layout is just files on disk — commit them or gitignore them as you would any other tool configuration.

Checking for updates

Every install carries an index.json alongside the skill files, at https://stele.axelerant.tech/skills/stele-resources/index.json. It lists the current version — a short hash of the skill's contents, so it only changes when the skill itself changes:

{
  "name": "stele-resources",
  "version": "57c33603ba54",
  "sha256": "…",
  "files": ["stele-resources/SKILL.md", "stele-resources/references/…"]
}

Compare that version against the copy in your installed SKILL.md (or just diff the files) to see whether you're behind. We tested npx skills update against a skill installed straight from this URL, and it does not pick it up — the update command only tracks skills it installed from a registered source (a GitHub repository it recognises), not a plain tarball URL. Until that changes upstream, treat updates as manual: re-run the install command above to pull the latest archive.

What the skill does not do

The skill is documentation, not automation. By itself it does not onboard a repository to Stele, talk to the portal, or reach the running site — installing it changes nothing about what gets published or how. It gives your agent the same reference a person would read in this guide.

To let the agent act on Stele — fetch the live link to a page it just pushed, or onboard a new repository — also connect it to Stele's MCP server. The skill tells the agent to use those tools when they are connected, and to point you at that guide when they are not.