Skip to main content
Agent Skills are packaged knowledge modules that teach an AI agent (Claude Code, Cursor, Codex, Gemini CLI, and others) how to operate Keebai end-to-end. Once the skill is installed, you can tell your agent “send ‘Hello’ to +1555… via WhatsApp” and it will run the right flow without you having to explain the SDK, the endpoints, or auth. Repo: github.com/keebai-ai/agent-skills — the open Agent Skills format (the same one used by Vercel, Kapso, Shopify, and others).

​
Why use the skills

Zero agent ramp-up

Without the skill, an agent has to figure out which SDK to use, where the PAT is, which method covers each case, and how to handle errors. With the skill: read SKILL.md, run the script, done.

Executable scripts included

25 CLI scripts (one per SDK method) that take --to, --text, --template, etc. The agent runs them directly, no code to write.

Bundled references

9 files in references/ (messages, media, templates, interactive, webhooks, errors, etc.). The agent lazy-loads them on demand.

Standard format

YAML frontmatter + Markdown. Works with Claude Code, Cursor, Codex, Gemini CLI, Aider, Continue, Cline, Roo, Warp, Goose, OpenHands, and 50+ other agents.

​
Installation

With Vercel Labs’ skills package (no prior install needed, comes via npx): 
npx skills add keebai-ai/agent-skills -g -y
Flags:
FlagWhat it does
-gGlobal (lands in ~/.claude/skills/ or the directory matching your agent).
-ySkip interactive confirmations.
-a <agent>Install for a specific agent only (e.g. claude-code).
--copyCopy files instead of symlinking.
Alternative modes: 
# Just list what skills the repo ships, without installing
npx skills add keebai-ai/agent-skills -l

# Project-level (only in the current cwd, not global)
cd ~/my-project && npx skills add keebai-ai/agent-skills -y

# View installed skills
npx skills list

# Update to the latest version
npx skills update integrate-keebai-whatsapp -y

# Remove
npx skills remove integrate-keebai-whatsapp -g -y

​
Available skill

​
integrate-keebai-whatsapp

End-to-end WhatsApp integrations. Covers the full cycle: sending messages (text, media, interactive, templates), uploading and downloading media, scheduling mass broadcasts, and receiving events via webhooks. Supports both SDK modes (Keebai PAT or direct Meta). Triggers that auto-invoke it in a conversation with the agent:
  • “@keebai/sdk”, “Keebai SDK”, “Keebai public-api”
  • “send WhatsApp via Keebai”, “Keebai WhatsApp template”
  • “pat_” / “kbai_pk_”, “phone_number_id”
  • “api.keebai.com”, “Keebai webhooks”
Contents (SKILL.md + bundles): 
integrate-keebai-whatsapp/
├── SKILL.md                          # entry point
├── references/                       # detailed docs, lazy-loaded
│   ├── getting-started.md
│   ├── messages-reference.md
│   ├── media-reference.md
│   ├── templates-reference.md
│   ├── interactive-reference.md
│   ├── webhooks-reference.md
│   ├── errors-reference.md
│   ├── meta-direct-mode.md
│   └── public-api-reference.md
├── scripts/                          # 25 Node.js executables
│   ├── send-text.mjs
│   ├── send-image.mjs
│   ├── send-template.mjs
│   ├── send-bulk.mjs
│   ├── upload-media.mjs
│   ├── mark-read.mjs
│   └── …
└── assets/                           # JSON examples
    ├── webhook-message-received.example.json
    ├── webhook-message-status.example.json
    └── template-create.example.json

​
Setup

Once installed, export the credentials once (the skill scripts read them automatically): 
# Recommended: ~/.zshenv (loaded by every zsh shell, including non-interactive ones)
cat >> ~/.zshenv <<'EOF'

# Keebai credentials for the integrate-keebai-whatsapp skill
export KEEBAI_API_KEY=kbai_pk_<hex64>
export KEEBAI_PHONE_NUMBER_ID=<numeric>
EOF

# Install the script dependencies (once)
cd ~/.agents/skills/integrate-keebai-whatsapp && npm install
Use ~/.zshenv, not ~/.zshrc. The Bash tool in Claude Code (and other agents) runs commands in non-interactive mode, and ~/.zshrc only loads in interactive shells. ~/.zshenv is read by every zsh shell.

​
Usage with Claude Code

After setup, in any session:
User: I need to send “Hello from Keebai” to +15551234567
Claude (with the skill loaded) runs directly: 
node ~/.claude/skills/integrate-keebai-whatsapp/scripts/send-text.mjs \
  --to +15551234567 --text "Hello from Keebai"
And returns: 
{
  "messagingProduct": "whatsapp",
  "messages": [{ "id": "wamid.HBgL..." }],
  "raw": { "message_id": "wamid.HBgL...", "status": "sent", "sent_at": "..." }
}
Without the user having to say which script, which flags, which endpoint, or which credentials.

​
CLI scripts (no agent)

The scripts also work as standalone CLI utilities, no agent required: 
# Send text
node ~/.claude/skills/integrate-keebai-whatsapp/scripts/send-text.mjs \
  --to +15551234567 --text "hello"

# Send a template with variables
node ~/.claude/skills/integrate-keebai-whatsapp/scripts/send-template.mjs \
  --to +15551234567 --name welcome_v2 --language en_US \
  --variables '{"name":"Alex"}'

# Upload media
node ~/.claude/skills/integrate-keebai-whatsapp/scripts/upload-media.mjs \
  --file ./logo.png --type image/png

# Mark as read + typing
node ~/.claude/skills/integrate-keebai-whatsapp/scripts/mark-read.mjs \
  --message-id wamid.HBgL... --typing text

# List approved templates
node ~/.claude/skills/integrate-keebai-whatsapp/scripts/list-templates.mjs
Every script accepts --help with flag documentation.

​
Agent compatibility

The SKILL.md format is supported by:
  • Claude Code (Anthropic CLI)
  • Cursor
  • Codex (OpenAI)
  • Gemini CLI (Google)
  • Aider
  • Continue
  • Cline
  • Roo Code
  • Warp
  • Goose
  • OpenHands
  • Windsurf
  • And 50+ other agents via the open registry
Full list in the skills npm package.

​
Upcoming skills

Roadmap, no committed dates:
  • automate-keebai-whatsapp — AI assistants, knowledge base, scheduling, workflows.
  • observe-keebai-whatsapp — delivery debugging, webhook inspection, health checks.
Need one before we ship it? Write to support@keebai.com or open an issue in the repo.

​
Resources

GitHub repo

Skill source, issues, and PRs.

TypeScript SDK

The client the skills are built on top of.

Agent Skills format

Spec for the open format.

REST API

Raw endpoints for cases without an SDK or agent.