exiao/product-factory/verify-deploy

Verify deploy

This is an observational check. Determine the target services, environment, and endpoints from the user's request, repository documentation, or configured provider data. Never invent a host, substitute a similarly named service, or probe a dead/legacy endpoint. Confirm the current CLI/API shape with --help or provider documentation before using it. Credentials come from the environment or connector and must never appear in output.

For every explicitly monitored endpoint, record status, latency, and a bounded body summary when safe. Distinguish frontend and API paths, redirects that are expected for auth, cold-start latency, cross-origin resource status 0, and wrapper/tool failures. Retry a slow 200 once or twice before flagging a transient spike, while retaining the original measurement. For post-release verification, establish that the intended commit or deploy ID reached the target environment; health alone may describe the previous release. Read relevant existing provider runs and logs. Use documented non-destructive smoke checks when in scope; verification alone does not authorize repairs or state-changing tests.

When the release distributes packages or builds, derive its artifact set and intended channels from the authoritative release tag or manifest before probing live state. Verify the checksum or digest of downloaded evidence when the source publishes one. Reconcile the planned artifacts and versions against live registry or store readback; distinguish missing items from those intentionally absent on a track or platform. If a non-destructive published-install smoke check is in scope, use isolated or ephemeral state.

Lead the report with actionable failures, then include the endpoint table, recent deploy signal, and any cron/workflow status. State the exact evidence, timestamp, auth state, and untested surfaces. A health check does not authorize a deploy, rollback, message, or code fix. If the user also asks for a fix, continue through the authorized repair and reverify; do not stop at diagnosis or ask for the same authorization again.

Read references/probe-and-report.md for endpoint semantics, workflow/cron interpretation, and a compact report shape.