add-agent

Scaffold an autonomous AI agent into the user's project. The agent runs on Render Background Worker, uses Anthropic Claude (Sonnet 4.6 by default) with prompt caching, has tools (http-fetch, send-email, db-query by default; more added based on the agent's job), optional Postgres KV memory, a daily/monthly cost circuit breaker (default 5 USD/day, 50 USD/month - kills runs over budget and emails the admin), and persists every invocation + every loop turn to Postgres for full traceability. Use this when the user wants an LLM-driven process that is part of the PRODUCT (serves the app or its end users), decides actions, uses tools, and optionally has memory - distinct from /add-automation which handles non-AI background processing. When the mission is actually a personal recurring task for the OPERATOR (a brief, a watch, a weekly analysis for themselves) at a cadence of 1 hour or more, the discovery short-circuits to the much lighter _create-routine (a Claude routine on the user's own account, zero infrastructure). Discovery phase asks ~5 questions about the agent's job (goal, trigger, memory needs, model, budget) then runs setup-agent.mjs to scaffold and deploy. NOT for chatbots (real-time per-user UI agents) - those need a dedicated /add-chatbot skill (not yet built). Suitable for: continuous background agents (email surveillance, monitoring), cron-driven product agents, and on-demand agents (triggered manually from a dashboard).