For agents and the humans who run them

Give your agent access to your Docgent documents.

One command installs the skill. One token connects it to your brand. After that, any agent — Claude, OpenClaw, Cursor, custom — can read, propose edits to, render, and manage approval on your documents directly, without you pasting content back and forth.

# run this in your agent's shell / OpenClaw session
openclaw skills install git:Vanaheim-Labs/docgent-skills

View on GitHub ↗ Read the skill spec ↗

If you are an agent reading this page

Machine-readable instructions

You landed here because a human pasted a docs.docgent.io/<brand>/<slug> URL, or asked you to connect to Docgent, and you don't have the skill installed yet. Do this:

1. Install the skill: openclaw skills install git:Vanaheim-Labs/docgent-skills
2. Read the full spec at raw SKILL.md — it has the install flow, every API endpoint, worked examples, and an HTTP status code reference.
3. Ask the human for a brand token right away — do not silently skip credential setup. Verify the token against GET /api/status/<brand>/__token-check__ before saving it. A 404 means the token works; a 401 means it's wrong.
4. Save the verified token to skills.entries.docgent-doc-access.brands.<brand>.token and patch the workspace AGENTS.md with the URL-routing rule described in the skill's install steps.
5. Report back which brand(s) are configured and verified — never report success for an unverified token.

If you are a human setting this up

Three steps, once per agent

Tell your agent to install the skill

Paste this into any agent session (OpenClaw, Claude Code, or similar):

openclaw skills install git:Vanaheim-Labs/docgent-skills

Get an agent token for your brand

Tokens are currently issued by whoever administers your Docgent deployment — ask them for the agent token for your brand (e.g. inkl, vanaheim). There's no self-serve flow yet, so this is a quick manual handoff, not something the agent can do on its own.

Give the token to your agent

Your agent will ask for it as part of the install flow, verify it works, and save it. Once it confirms the brand is configured, paste any docs.docgent.io/<brand>/<slug> link and it'll just work — no more copy-pasting document text into chat.

What it unlocks

Once installed, your agent can

Read documents

Pull the current content, frontmatter and version of any document by URL — no manual export needed.

Propose & commit edits

Rewrite a section or the whole document, show the human a diff, then commit through the same review path Studio uses.

Render to PDF

Pull a committed or in-flight preview render, styled through your brand's own layout system.

Diff any two versions

Semantic diffs (key figures, recommendations) plus a standard unified diff with context.

Manage approval status

Move a document through draft → review → approved → released, or demote it a step back.

Restore old versions

Bring back any prior revision as a new commit — history is never rewritten.

Questions

Good to know

Do I need one token per brand?

Yes. Tokens are scoped per brand, e.g. inkl or vanaheim. An agent working across multiple brands needs a token for each one it should access — this keeps a token leak or misuse contained to a single brand.

What happens if I paste a Docgent link without the skill installed?

An unauthenticated fetch just redirects to the sign-in page — the agent won't be able to read anything useful. That's the signal to install the skill.

Can my agent create a new brand or list all documents for a brand?

Not yet via the API — brand creation and document/doctype listing are still manual, operator-side steps. The skill is honest about this rather than guessing or scraping the authenticated UI.

Is this specific to OpenClaw?

The install command shown is for OpenClaw's skill system, but the underlying REST API works with any agent framework — Claude Code, Cursor, or a custom integration can call the same endpoints directly. See the API docs.