Developer guide / testing

Shared tests and review

A passing suite proves specific behavior. It does not prove production readiness.

Draft · The release review is in progress.

The test boundary

Each case starts a fresh runtime and platform fixture. The harness uses loopback HTTP and WebSocket connections.

Drivers only configure libraries. They do not implement routes, middleware, analytics export, or persistence.

AIMock supplies agent and MCP responses. The harness does not load .env or use a live model key.

A new behavior starts with a failing shared case or native test. A coding agent fixes its assigned implementation.

Stable case IDs let each agent select one failure. The same case then runs against every language.

The harness also rejects an HTTP-200 stub. A response without platform effects cannot pass.

A second reviewer examines failure paths beyond the shared cases. Findings become regression tests before a fix.

Review found that a second Python stop canceled an unacknowledged batch. The shared suite did not cover that defect.

Release checks also include native tests, type checks, package builds, real framework runs, and browser integration. Open findings remain visible.

The shared suite contains 100 cases. Ruby runs the same cases through Rack and Rails.

At commit c54bfa2a1a, TypeScript passed all 100 shared cases and 2,332 Runtime tests locally.

One case uses the public frontend client with real Phoenix sockets. It runs in Node, not a browser.

Review and release checks continue. These results do not establish production readiness.

These counts describe checkpoints. The final PR must include fresh results for its exact commit.

Pull request #6967 records the test commands, review findings, and release decisions. Each result names the behavior that the test covers.

The package READMEs explain installation, configuration, and host responsibilities for your language.