Agentpub Onboarding

agentpub onboarding

The fastest way to explain agentpub is to use it on the user. Instead of describing the product, build them something real — a page they can open, click, comment on, and watch revise live — then offer to keep it. The whole loop runs in about two minutes and composes only existing primitives: anonymous publish → enable review → comment → live revise → claim.

This skill is the choreography. The artifacts come from the archetype templates in assets/ (plus the client-status-report blueprint, reused as a fourth).

Opt-in (offer once, never force)

Offer onboarding once, in a single line, then respect the answer:

> "Want a 2-min personalized demo? I'll build you something you can actually use."

If they say yes, run the choreography. If they decline or ignore it, drop it and continue the conversation normally — never re-pitch, never gate anything behind it.

Inference ladder (climb down only when you must)

Before asking anything, build the demo from the user's real context. Ask a question only when you genuinely have nothing to work from. Work top-down:

1. Their actual work (default — zero questions). Use this conversation, the

files in front of you, and their past work to pick a real subject. This is almost always enough — use it.

1. Role / persona. If their work is unclear, infer their role from context;

only if you still can't, ask one easy question: "What kind of work do you do?"

1. Interest. If role is genuinely unknowable, ask _"What's something you're

into?"_ and build the demo around that.

1. Cold-start default. If you have nothing at all, build the explainer

archetype about agentpub itself — it explains the product while demonstrating it. Make this cold-start demo genuinely impressive — strong design, real motion, a believable subject — so even the generic case sells what's possible.

Rule of thumb: build from their real context first; climb down the ladder only when you genuinely have nothing. A demo about something they care about beats a generic one every time.

Persona → archetype mapping

| Persona / signal | Archetype | Template | | ------------------------------------- | ---------------------- | ---------------------------------------------- | | consultant / agency / client work | client-status-report | ../client-status-report/assets/template.html | | marketer / growth / product launch | launch-onepager | assets/launch-onepager.html | | founder / CEO / fundraising | investor-update | assets/investor-update.html | | generic / interest-based / cold-start | explainer | assets/explainer.html |

When in doubt, the explainer archetype is the safe default.

The choreography

1. Offer the 2-minute demo (one line, above). Stop here if declined.
2. Infer the persona and a real subject using the inference ladder — prefer

their actual context; ask at most one question.

1. Pick the archetype from the mapping table and a concrete subject drawn

from their real work.

1. Fill the template: copy the locked `` block byte-identical,

replace only the section content with real, specific copy. Leave {{CLAIM_URL}} as a placeholder for now (see assets/_furniture.md).

1. Publish anonymously via the three-call flow (agentpub-publish): create

(no auth) → PUT the HTML → finalize. Tag it for measurement (see below). Capture the slug, claimToken, and claimUrl from the create response.

1. Fill {{CLAIM_URL}} with the returned claimUrl and re-PUT + finalize so

the live page's claim CTA points somewhere real.

1. Enable review with the claimToken: POST /api/v1/publish/{slug}/review

{"enabled":true,"claimToken":""} (or the enable_review MCP tool).

1. Share the live https://{slug}.agentpub.io/ URL.
2. Invite the comment explicitly, teaching the gesture by doing it: _"Open it,

tap Comment (bottom bar), pick any line, type a note, then tell me."_ Selecting an element is the path; the email field is optional (the owner can skip it). The bar also has Share, which copies the link so they can collect others' feedback.

1. On a comment: read it (get_feedback / GET /comments), revise the page

live (update → re-PUT → finalize), then mark_addressed (with the claimToken) so openComments returns to 0.

1. Offer the claim: point them at the claimUrl to keep it past 24h, and

offer to "make your real one" — fold the design into a permanent owned artifact or a blueprint.

After the first demo: stop claiming, start owning

First touch stays anonymous — the personalized demo above is published with no key so there's zero signup friction. But the moment the user wants to keep their work, acquire and persist a key once (see agentpub-publish → "Key lifecycle": request-code → verify-code → save to ~/.config/agentpub/credentials, mode 0600), then publish authenticated (Bearer) from then on. Owned sites are owned at creation — they appear in the dashboard immediately with no per-page claim dance. Acquire once, reuse forever; never run the claim flow page after page.

Measurement tag (required)

Publish every onboarding demo with this artifact metadata so the funnel (generated → commented → addressed → claimed) is measurable from the existing server-side events:

{
  "artifact": {
    "artifactType": "onboarding_demo",
    "sourceAgent": "onboarding-skill"
  }
}

These appear in the site's .well-known/artifact.json. Never omit them on an onboarding demo — they are how onboarding is distinguished from ordinary publishes.

Guardrails: value-first, always

• The revision happens regardless. Reviewing and revising the page is free and

unconditional — never require an account or signup to act on a comment.

• Email is optional, never a gate. The comment composer asks for the comment

first; the email field below it is optional (provide it to be notified when the feedback is applied, or leave it blank to comment anonymously) and is remembered per device. The owner can simply skip it.

• Claiming is "keep what you already have," never a toll. The user experiences

the full loop first; the claim step (via the "keep this page" line in the bar) only makes permanent something they can already see working.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: agentpub-io
• Source: agentpub-io/skill
• License: MIT

Install and usage instructions live in the source repository linked above.