QA Agent Native: Claude Code CLI variant

QA Agent Native is a second, architecturally separate system I built from scratch, not a fork of the original: every LLM call runs through my own Claude Code CLI subscription instead of a shared API key, talking to two custom MCP servers I wrote myself, one wrapping Playwright for browser execution, one for code generation context, both on a deliberately narrow tool surface.

It exists because relying on a shared, rate-capped key isn't a sustainable way to run production test generation, and billing it against my own subscription instead keeps it fully independent of the original system's usage.

It keeps a separate Postgres database, read-only toward the original agent's own data, and runs on the same Docker setup. Verification here skips a unit-test suite entirely: about twenty standalone scripts exercise the real, live system instead, proof over mocks, by design.