Turn an idea into an agent build specification

Design, but do not deploy, an AI agent for [USE CASE]. Users are [AUDIENCE], inputs are [DATA], and the desired outcome is [OUTCOME].

Specify: trigger; step-by-step workflow; tools needed and why; minimum permissions; data retention; human approval points; source-of-truth rules; output schema; retry and stop limits; cost/latency budget; audit log; ten acceptance tests; and a manual/no-code fallback. Separate deterministic code from tasks needing AI. Identify any step that is unsafe to automate. End with the smallest viable pilot and what evidence would justify expansion.