wa-whatsapp-agent

Characterize a WhatsApp AI agent before writing any code. Use when a student has finished wa-setup and is ready to define what the bot does, or says 'wa-characterize', 'אפיין סוכן', 'בוא נתכנן את הבוט', 'מה הבוט עושה', 'הגדר את הבוט'. This skill asks the hard questions (who does it answer? what's its knowledge? which tools?) and produces a spec.json that wa-build reads to generate the bot.

Deploy a WhatsApp AI agent to Render.com - automated via Render CLI + API. Use after wa-build (bot's tone is confirmed locally) or when student says 'העלה סוכן', 'wa-deploy', 'deploy agent', 'תעלה לפרודקשן', 'תעלה את הסוכן', 'הסוכן מוכן מה עכשיו'. Minimizes browser clicks: student provides a Render API key once, Claude Code runs CLI + REST API calls to create the web service, attach disk, optionally provision Postgres (for Outlook), deploy, and wire the Green API webhook. Student only intervenes for one-time GitHub↔Render OAuth connection and any payment approval.

Maintain, debug, and update a deployed WhatsApp AI agent. Use when the bot is live and needs changes, or the student says 'wa-maintain', 'תתקן את הסוכן', 'הסוכן לא עובד', 'תשנה את הסוכן', 'תעדכן את הבוט', 'שנה prompt', 'הסוכן תקוע', 'הסוכן לא עונה', 'תוסיף כלי', 'תוסיף פיצ'ר'. Routes to the right remedy: prompt tune, scope change, tool add/remove, Google/Microsoft token refresh, debug checklist for outages.

Set up Green API for WhatsApp agent connection. Use when a student needs to connect WhatsApp, register for Green API, or says 'חבר WhatsApp', 'הקם Green API', 'wa-setup', 'אני רוצה להתחיל לבנות סוכן WhatsApp', 'בוא נתחיל עם WhatsApp'. Also trigger when someone mentions needing a WhatsApp connection for their bot or agent. This skill handles instance type selection (free for personal assistant, paid for customer service), phone number setup (eSIM recommended), QR scanning, incoming webhook configuration, and end-to-end verification.

Build the WhatsApp agent code from a characterization spec. Use after wa-characterize when the student has an approved spec.json, or says 'wa-build', 'בנה את הסוכן', 'תבנה את הקוד', 'יאללה בוא נבנה'. This skill enforces the opinionated architecture (FastAPI + direct LLM SDK + SQLite + explicit tool registry) and guides Claude Code to generate a clean, deploy-ready codebase. No framework magic - the student can read every line.

Wire a tool (Google Calendar, Gmail, WhatsApp groups, Human handoff, Outlook) into a deployed WhatsApp agent. Use after wa-deploy when the student is ready to give the bot capabilities, or says 'wa-connect', 'חבר כלי', 'חבר יומן', 'חבר מייל', 'חבר קבוצות', 'תוסיף כלי לסוכן'. Each invocation wires exactly ONE tool and redeploys. Run multiple times to add multiple tools. Handles auth, credential storage, tool implementation, redeploy, and live verification.

Give the deployed WhatsApp bot durable memory that survives Render restarts. Use when the student says 'הבוט שוכח שיחות', 'הבוט לא זוכר אותי', 'wa-persistence', 'תוסיף זיכרון לבוט', 'reminders נמחקים', or any variant of 'memory doesn't persist'. Also called automatically at end of wa-deploy if the student is on Render Free without a disk. Four sub-flows: (A) Render Disk, (B) Supabase Pooler — most common, (C) Render Postgres, (D) Skip with warning.