Zernio Library Skills

A library of Claude Code skills for Zernio. It ships:

• clip-machine — one long video in, native TikTok + Reels + Shorts posts out. The four stations: CUT (word-level transcript → best-moments framework → 9:16 reframe), CAPTION (burned word captions + per-platform caption text), COVER (per-platform cover rules), POST (one Zernio request per clip). Includes the copy-paste best-moments selection prompt and the per-platform spec cheat-sheet (TikTok / Reels / Shorts: dimensions, lengths, captions, covers, native rules). New here? Copy-paste the initiation prompt to Claude Code in an empty folder and the whole system sets itself up.
• white-label-blueprint — build your own white-label social media tool: YOUR brand, your clients’ accounts, one API underneath. The five checks: embedded OAuth under your brand (headless connect — the client never sees the vendor), one isolated workspace per client (profiles, 403-enforced), full-capability posting on their behalf, webhooks for token health (“reconnect needed” before the client notices), and the maintenance you never build. Deliberately thin: no database (the API is the record), one ~200-line server, two pages. Ships the stage prompts (paste one per check, each a visible win), depth stages (unified inbox, per-client analytics + printable report, the posting queue), THE WHITE-LABEL CHECKLIST one-pager, and THE POLISH LOOP — a scored screenshot→score→fix loop (10-category SaaS standard, exit only on two consecutive 9+ passes) that takes the dashboard from working to product-grade. New here? Copy-paste the initiation prompt to Claude Code in an empty folder.
• zernio-publish — end-to-end social publishing across 13 platforms.
• zernio-comment-to-dm — build Instagram/Facebook comment-to-DM automations + DM sequences on command. Tell Claude a post, a keyword, and the message — it creates the Zernio automation and orchestrates the follow-up DMs, so you never click through a dashboard or clone workflows. Reuse it for any new post/keyword in seconds.
• meta-ads-launch — launch Meta (Facebook/Instagram) ad batches through Zernio’s ads API. You decide the offer, copy, audience, and budget; Claude launches ALL your creative variations in one call (1 campaign + 1 ad set + N ads), then immediately pauses the campaign so nothing spends until you review and go live. Ships with the complete verified Zernio ads API reference (multi-creative shapes, idempotency, pause/resume, targeting, lead forms, click-to-WhatsApp). Requires the Zernio Ads add-on (included on usage-based plans).
• zernio-workflow-creator — build Zernio conversation Workflows, including a 24/7 WhatsApp customer-service agent, on command. Tell Claude what the agent should do and it builds the whole node graph (trigger → human-escape → AI node with memory → reply → wait → loop) via Zernio’s Workflow API, activates it, and edits it later when you ask — no dragging nodes one by one. Pairs with zernio-comment-to-dm (comment → DM → this AI agent runs the conversation). Ships the full Workflow API reference (all 16 node types) + a ready, fill-in-the-blank customer-service agent template.
• zernio-voice-agent — give ONE WhatsApp number a full AI front desk: text it and an AI replies (memory-enabled Zernio workflow); call it and an AI voice picks up (Zernio WhatsApp Calling routes inbound calls via forwardTo to a Retell AI voice agent). Live-proven inbound-first: the text agent detects call intent and sends the number’s tap-to-call deep link; the tapped call lands on the voice agent. Ships the verified calling endpoints + gates (plan + messaging-tier), the Retell wiring reference, the proven front-desk workflow template, an outbound start_call variant, and fill-in-the-blank voice prompts. Pairs with zernio-workflow-creator.

More Zernio skills (analytics, calendaring, account management) will land over time.

You hand Claude an asset (video file, Drive link, image folder, URL) and tell it where to post. Claude analyzes it, drafts platform-tailored captions and titles and hashtags, converts formats if needed, uploads the media, shows you the full package for approval, schedules the post via the Zernio API, verifies it landed, and logs the result.

Install

Everything lives in one hidden folder: .claude/ (note the dot — your file explorer may hide it). Claude Code automatically discovers any skill placed at .claude/skills// in your project and loads it — that’s the whole install mechanism.

Option A — use this repo as your project (simplest):

git clone https://github.com/Enriquemarq1/zernio-library-skills.git
cd zernio-library-skills
claude

Option B — you already have a project (and probably your own .claude/ folder): MERGE, don’t replace. Your project’s .claude/ and this repo’s .claude/ are the same kind of folder. Do NOT overwrite yours — copy the skill folders you want into it, side by side with your existing skills:

# from inside your project
cp -r /path/to/zernio-library-skills/.claude/skills/* .claude/skills/

# Windows PowerShell
Copy-Item -Recurse \path\to\zernio-library-skills\.claude\skills\* .claude\skills\

Result: your-project/.claude/skills/ now contains your own skills plus zernio-publish, zernio-comment-to-dm, meta-ads-launch, zernio-workflow-creator, zernio-voice-agent, …

Verify it worked: restart Claude Code in the project and the skills appear as slash commands — type / and you’ll see them (e.g. /zernio-publish) — and they also auto-trigger when you just describe the task (“post this video to TikTok”, “make my WhatsApp answer calls with AI”). If you don’t see them, the folders are probably nested one level too deep: the check is that .claude/skills/zernio-publish/SKILL.md exists at exactly that path in YOUR project.

For claude.ai web (Skills upload UI), zip the skill folder yourself:

cd .claude/skills/zernio-publish && zip -r ../../../zernio-publish.skill.zip .

Then upload that ZIP on claude.ai → Capabilities → Skills.

Configure

Edit .env at the repo root, replace the placeholder with your real key from https://zernio.com/dashboard/api-keys, and run:

git update-index --skip-worktree .env

That keeps your local key edits from being staged into commits while the file stays tracked.

Use

In Claude Code, just say what you want to ship:

“Publish this video to my YouTube, TikTok, and Reels: ./media/launch.mp4”

“Drop a carousel post to Instagram and LinkedIn from this Drive folder: ``. Angle: three lessons from launching v1.”

Claude reads the skill, runs through the workflow (intake → analyze → draft → convert → upload → package → approve → post → verify → log), shows you the full package, and ships on your approval.

What’s in the box

zernio-library-skills/
├── .claude/
│   └── skills/
│       ├── zernio-publish/                    ← the publishing skill
│       │   ├── SKILL.md
│       │   ├── reference/                       deep API + per-platform docs
│       │   └── templates/manifest.json
│       ├── zernio-comment-to-dm/              ← comment-to-DM automations + sequences
│       │   ├── SKILL.md
│       │   ├── reference/zernio-automations-api.md
│       │   └── templates/automation.json
│       ├── meta-ads-launch/                   ← Meta ad batches via Zernio's ads API
│       │   ├── SKILL.md
│       │   ├── reference/zernio-ads-api.md      the complete verified ads API reference
│       │   └── templates/campaign-plan.json
│       ├── zernio-workflow-creator/           ← Zernio Workflows + WhatsApp AI agents
│       │   ├── SKILL.md
│       │   ├── reference/zernio-workflows-api.md  full node/edge contract (16 node types)
│       │   └── templates/whatsapp-customer-service-agent.json
│       ├── zernio-voice-agent/                ← WhatsApp AI FRONT DESK (Zernio + Retell AI)
│       │   ├── SKILL.md                          text + voice on one number, inbound-first
│       │   ├── reference/whatsapp-calling-api.md  verified calling endpoints + gates + start_call
│       │   ├── reference/retell-voice-api.md      Retell wiring (inbound_agents / SIP / wss)
│       │   └── templates/
│       │       ├── whatsapp-ai-front-desk.workflow.json  THE proven graph (text + tap-to-call)
│       │       ├── whatsapp-voice-agent.workflow.json    outbound start_call variant
│       │       ├── retell-voice-agent.base.md            fill-in-the-blank voice prompt
│       │       └── retell-voice-agent.example.md         a filled voice prompt (booking)
│       └── skill-creator/                     ← Anthropic's official skill-creator
│                                                  bundled for when you add more skills
├── .env                                       ← ZERNIO_API_KEY (placeholder until you fill it)
├── CLAUDE.md                                  ← architectural map of the repo
├── README.md
├── LICENSE                                    ← MIT
└── .gitignore

No build scripts, no helper scripts, no auto-generated artifacts. The agent has its native Claude Code toolkit (bash, curl, ffmpeg, vision, transcript extraction, web fetching) and uses those directly.

Adding more skills to the library

Drop a new folder under .claude/skills/:

mkdir .claude/skills/my-new-zernio-skill
# Create SKILL.md with YAML frontmatter (name + description) and the skill's instructions

The bundled skill-creator at .claude/skills/skill-creator/ is Anthropic’s official tool for designing and improving skills — invoke it when building a new one.

Design principles

• End-to-end agency. Claude handles intake, analysis, drafting, conversion, upload, posting, verification, logging. Not transport-only.
• The skill informs; the agent acts. SKILL.md is reference knowledge plus a suggested workflow — not a script. The agent uses its native tools to execute.
• Approval before publish. Every post passes through an explicit gate. No bypass, no auto-publish.
• Verify, don’t trust. Zernio’s 200 OK ≠ landed on the platform. Claude hits each platform’s public URL or oEmbed after scheduledFor + 60s and surfaces any field drift.
• Schedule, don’t push. Multi-platform posts always use scheduledFor 2-3 minutes ahead.