WABA Agent Operator

Use this skill to

• Run regression and smoke checks for waba-agent.
• Diagnose CLI command mismatches and runtime errors.
• Patch onboarding for Meta + AI + staging verification.
• Validate safety rules: no shell/eval action flow; approval for high-risk actions.
• Configure or verify local AI defaults with Ollama.

Standard workflow

1. Confirm repo state before edits.
   git status --porcelain=v1 -b
2. Run tests using npm.cmd on PowerShell systems.
   npm.cmd run test:ai
npm.cmd run test:chat
npm.cmd run test:gateway
npm.cmd run test:resale
npm.cmd test
3. Run command-surface smoke checks.
   node bin/waba.js --no-banner --help
node bin/waba.js --no-banner chat --help
node bin/waba.js --no-banner gateway --help
node bin/waba.js --no-banner resale --help
4. Verify gateway root headers to avoid stale UI.
   Expect:

• Cache-Control: no-store, no-cache, must-revalidate, proxy-revalidate
• Pragma: no-cache
• Expires: 0

5. Verify high-risk gating behavior.

• allowHighRisk=false must block message.send_text, template.send, and schedule send tools.
• allowHighRisk=true may proceed but must still fail safely when auth/config is missing.

6. For multi-client checks, prefer explicit --client where supported.

Onboarding requirements

Capture in wizard whenever possible:

• Meta auth: token, phone-id, business-id
• AI setup: provider, key, model, base URL
• Staging verification profile: test recipient, template name/language/params

Persist:

• Global AI config in ~/.waba/config.json
• Client verification profile in ~/.waba/context/<client>/client.json under verification.staging

Local AI defaults

If hosted keys are absent, runtime should use local Ollama:

• Base URL: http://127.0.0.1:11434/v1
• Primary model: deepseek-coder-v2:16b
• Fallback model: qwen2.5:7b

Quick local prep:
ollama pull deepseek-coder-v2:16b
ollama pull qwen2.5:7b
ollama serve

Non-negotiable safety rules

• Keep approval gating for high-risk actions.
• Do not introduce eval, shell execution, or dynamic code execution in action flow.
• Do not remove existing user-facing commands unless explicitly requested.

Output contract for takeover and validation tasks

Return:

1. Exact issues found
2. Code changes made
3. Commands run
4. Demo checklist with pass/fail and explicit blockers