[LinuxDevil]

Fixes audit finding OPS-4 (Vercel competitor audit, 2026-10-19).

What was wrong

The built node server only had GET /health -> ok, so orchestrators could not tell "process is up" from "able to serve" (e.g. during a SIGTERM drain).

What changed

• createDeployedServer answers GET /ready: 200 {"status":"ready"} while serving, 503 {"status":"draining"} once shutdown() has begun. No auth, no detail in the body, cache-control: no-store.
• /health is unchanged (liveness); the generated Dockerfile HEALTHCHECK stays on /health.
• No store check: AgentStore has no cheap ping, so it is skipped. The agent is loaded before the server listens, so "starting" is never observable.
• Cloudflare Worker and lousho dev untouched (node-server/docker only; documented).
• docs/deployment.md documents health vs ready; llms regenerated.

Tests

src/deploy/nodeServerReady.test.ts: 200 ready and /health still ok; no auth needed with LOUSHO_API_TOKEN set (while /chat still 401); 503 draining on a non-idle connection once shutdown starts. tsc, typecheck:tests, lint, test:coverage (6076 passed) and fallow all pass.

Changelog: The built node server and Docker image add GET /ready (200 ready, 503 draining during shutdown), separate from the /health liveness route.

🤖 Generated with Claude Code