Post-Development Verification

Automated full-stack quality verification after development. Real execution by default -- mock is the last resort. Deliverability is judged by external calls (HTTP requests, CLI invocations, browser interactions), not by internal function calls passing in isolation.

Core Philosophy: Real Execution First

Default realism level is L2: internal services run for real, only uncontrollable external dependencies (third-party APIs, paid services) may be mocked.

Downgrade signals (auto-detected from user intent):

• "快速验证" / "只测逻辑" / "mock就行" -> L0
• Pure function / utility library with no I/O -> L0
• Local environment cannot start service -> L1 (report reason)
• "生产级别" / "全面测试" / "验收测试" -> L3