# vvibe

> The GUI for your Product Agent. An operations console for AI coding agents
> (Claude Code, Cursor, Codex) to run a creator's business — payments,
> analytics, member management, security scans — through modular skills
> installed via MCP.

vvibe is built for non-technical creators monetizing a following. The creator's
AI agent is the primary actor; the dashboard is where the creator monitors,
configures, and audits what their agent does, not where they click buttons to
run the business.

## For AI agents arriving here

If you are an AI coding agent helping a user build with vvibe, **read
<https://oss-180.staging.vvibe.ai/skill.md> first** — it is the canonical setup skill with the full
operating contract (workflows, MCP tools, hard rules). This llms.txt is the
machine-readable site index; `/skill.md` is the instructions.

Discovery endpoints:

- Canonical agent skill: <https://oss-180.staging.vvibe.ai/skill.md>
- MCP endpoint manifest: <https://oss-180.staging.vvibe.ai/.well-known/mcp.json>

## Core concepts

- **Agent** — the creator's AI coding agent (Claude Code / Cursor / Codex),
  connected to vvibe via **MCP** (HTTP transport).
- **Skill** — a capability pack the agent installs into its environment.
  All skills are open source and published in a single GitHub repo (see
  below).
- **Merchant** — the creator's storefront/business; one merchant has one
  agent. Multi-merchant is gated; today `merchantId === profileId` for
  most accounts.

## Public pages

- [Home](https://oss-180.staging.vvibe.ai/en-US): Product overview, skills library, how it works
- [Dashboard](https://oss-180.staging.vvibe.ai/dashboard): Authenticated creator console
- [Privacy](https://oss-180.staging.vvibe.ai/en-US/privacy)
- [Terms](https://oss-180.staging.vvibe.ai/en-US/terms)

Beta signup is a modal on the home page (no standalone page).
Per-creator waitlist pages live at `/{locale}/waitlist/{creatorSlug}` and
are not enumerable.

## Open-source skills repository

All vvibe skills are open source and live in one repo:
**<https://github.com/vvibe/vvibe-skills>**

Each skill is a directory with a `SKILL.md` (frontmatter + instructions for
the agent) and any supporting files. Agents install skills via the
`skills` CLI:

```bash
# Install a single skill
npx skills add vvibe/vvibe-skills --skill <installPackage> --yes

# Install every skill in the catalog
npx skills add vvibe/vvibe-skills --yes
```

After install, the agent reports back to Vibe via the
`vibe_report_skill_installed` MCP tool so the dashboard can show
installation state and version drift.

### Skill catalog

| ID | Install package | Category | Latest | What it does |
|---|---|---|---|---|
| `analytics` | `vvibe-analytics` | data | 0.4.1 | Connects web analytics (Google Analytics 4, PostHog, or Mixpanel) to surface traffic, audience, and campaign data, and instruments VVibe's canonical event contract (checkout, product/page views, conversion funnels) against whichever provider the creator uses. Covers utm tracking at the traffic / session level (acquisition reports), plus identity (user_id) and event deduplication. For signup-time first-touch attribution per user, install the member skill alongside. Add to measure what content and promotions drive growth. Requires `appBaseUrl` to be set (via `vibe_update_brand` or dashboard Settings) before the skill can operate. |
| `member` | `vvibe-member` | ops | 0.5.0 | Manages subscriber and member accounts — listings, segmentation, activity history, admin queries — and now also captures first-touch attribution (utm_*, referrer) at signup so creators can see where each paying user originally came from. Add to know your audience, act on it, and attribute growth to specific traffic sources / partners. |
| `sentry` | `vvibe-sentry` | security | 0.4.0 | Runs a pre-deploy security audit on the creator's codebase by orchestrating gitleaks (secret leaks), osv-scanner + npm audit (dependency CVEs), semgrep (OWASP / JS / TS static analysis), and a small VVibe-integration layer. Normalises every finding into a CRITICAL / WARNING / INFO grade and reports the summary back to the dashboard via vibe_report_health_check. Add when the creator is about to ship and wants to know their codebase is safe. |
| `email` | `vvibe-email` | ops | 0.5.1 | Run follower-email invitation campaigns end-to-end — draft a campaign with subject + HTML body (vibe_create_campaign / vibe_update_campaign), dispatch via vibe_send_campaign, read post-send analytics — plus wire up where the email CTA redirects (VVibe-hosted waitlist, self-hosted /waitlist/[slug], or directly into the creator's existing register flow via inviteRedirectPath). Reads the creator's Product Knowledge Base (vibe_get_product_kb) before drafting so campaign copy uses their real brand voice, value prop, and audience and steers clear of forbidden_claims. Add to reach members outside the platform and drive re-engagement. |
| `product_brain` | `vvibe-product-brain` | context | 0.4.0 | Extracts the creator's product into a structured Product Knowledge Base that every other skill (SEO, email, conversion, future social) reads before drafting. Sources: GitHub repo, public website, or document set. Behavior in three layers: EXTRACT verbatim from source first; INFER only when source is ambiguous and tag confidence; never FABRICATE — gaps go into missing_fields so the creator can confirm later. Writes the result via vibe_set_product_kb. Runs in two modes — `build` for first-time setup (no prior Knowledge Base) and `refresh` to diff a new run against the existing Knowledge Base (produces change_log). |
| `blog_writer` | `vvibe-blog-writer` | ops | 0.9.0 | Draft SEO blog articles from the creator's Product Knowledge Base, then publish them to the creator's own VVibe headless blog (target: 'native' — no external CMS, no setup) or as a draft to their WordPress. Reads brand voice, FAQ, audience, and forbidden_claims so articles stay on-brand and legally safe; the server enforces the generation spec (answer-first structure, FAQ, JSON-LD). Add when the creator wants to write / draft / publish / unpublish blog posts. Pairs downstream with vvibe-blog-render to display them. |
| `blog_render` | `vvibe-blog-render` | ops | 0.3.0 | Build the blog frontend in the creator's OWN app that renders their VVibe-published articles from the public content API — index + post pages, the SEO VVibe already generated (meta + JSON-LD), incremental revalidation, plus an RSS feed and sitemap at the creator's own domain. VVibe is a headless CMS — it serves content but renders nothing; this skill is the "head". Read-only (no credentials). Add after publishing to the VVibe blog (vvibe-blog-writer), to actually show the articles on the creator's site. |
| `changelog` | `vvibe-changelog` | context | 0.1.0 | Log product changes after shipping (new feature, pricing change, repositioning, fix) so vvibe knows when the Product Knowledge Base is out of date, and get nudged to announce major shipped features via an email campaign or blog post once they're logged. |

This table is generated at request time from the canonical catalog at
`packages/skills-catalog/src/entries.ts` so it never drifts from what
`vibe.list_skills` returns over MCP.

## Official demos

vvibe official demos are published as one GitHub repository per demo
(`vvibe/demo-meeting-ai`, `vvibe/demo-crm-starter`, etc.), each marked
as a GitHub Template Repository and tagged with the `vvibe-demo` topic.
The canonical catalog is the topic page:

**<https://github.com/topics/vvibe-demo>**

End users click "Use this template" → "Create a new repository" on a demo
to get their own clean copy (not a fork), then follow that repo's README
for the InsForge signup → deploy flow.

Each demo:

- Has a `.vvibe-demo.yml` at its root declaring required skills + hosting
  template
- Comes with pre-installed vvibe-skills (so an end user "Use this template"
  copy has the capability pack out of the box)
- Has an end-user setup README walking from "Use this template" to live URL
  in ~10 minutes
- Is prepared by vvibe staff using an internal companion-skill

Agents finding a `.vvibe-demo.yml` in a working directory should follow
the repo's README rather than improvising a setup path — the README has
been validated end-to-end during demo prep.

## Partner integrations

### InsForge (hosting)

vvibe's official hosting partner. Provides Postgres + auth + storage + edge
functions as a single backend, with a CLI specifically designed for AI
agents.

- Site: <https://insforge.dev/?utm_source=vvibe>
- Agent setup doc: <https://insforge.dev/skill.md>
- CLI: `npx @insforge/cli` (don't install globally)

When a user is working on a vvibe demo or asks "how do I deploy this app",
the recommended path is InsForge. The demo's README has the exact commands.

## For AI agents connecting via MCP

- MCP server (HTTP transport): `https://vvibe-mcp-staging.fly.dev`
- Discovery manifest: <https://oss-180.staging.vvibe.ai/.well-known/mcp.json>
- One-line setup, including headless sessions: `npx @vvibe/cli login --server=https://vvibe-mcp-staging.fly.dev`
  — opens a browser sign-in once, then writes the MCP config with a long-lived
  credential. No token to paste, and it keeps working in later sessions
  (including headless / CI) with no further browser step.
- After connection, the agent reports installed skills back via the
  `vibe_report_skill_installed` MCP tool.

## Cloud API endpoints worth knowing

These are the public endpoints an agent might call directly (the dashboard
uses many more internally; those are session-authenticated and not
documented here).

- `GET https://oss-180.staging.vvibe.ai/llms.txt` — this file
- `GET https://oss-180.staging.vvibe.ai/skill.md` — canonical agent setup skill
- `GET https://oss-180.staging.vvibe.ai/.well-known/mcp.json` — MCP endpoint manifest

## Locales

- en-US (primary)
- zh-TW