refactor(cli): cover start e2e tests with effect lint (CLI-2414)

[7ttp]

TL;DR

brings the supabase start e2e tests under the effect lint, preserving the existing scenarios

whats introduced?

• enables effect lint for all three start e2e suites
• tests use the Effect harness with scoped project, process, stack, and proxy cleanup
• filesystem, HTTP, time, and JSON operations use Effect services and schemas
• the Docker test helper drains both output streams and retains bounded, Unicode safe diagnostics, with regression coverage in the existing helper tests

ref:

• related: CLI-2414

[github-actions]

🤖 AI Review

Both independent reviews completed. The major confirmed issue is that the lifecycle and slim-image E2E effects allocate scoped resources without applying Effect.scoped. Additional confirmed concerns cover effect lint, cleanup timeout budgets, an interruptible cleanup-registration gap, truncated parseable output, and one readability nit. Three Claude findings were refuted by concrete code or trusted conventions.

Findings

Severity	Location	Category	Sources	Claim
🟠 MAJOR	apps/cli/src/commands/start/start.lifecycle.e2e.test.ts:197	resource-lifecycle	claude+codex	All three lifecycle tests allocate scoped temporary directories and register finalizers but omit Effect.scoped, leaving an unprovided Scope requirement.
🟠 MAJOR	apps/cli/src/commands/start/start.slim-images.e2e.test.ts:286	resource-lifecycle	claude+codex	The slim-image E2E test uses a scoped project helper without scoping the enclosing effect.
🟡 MINOR	apps/cli/src/commands/start/start.lifecycle.e2e.test.ts:3	lint	claude	The newly effect-linted file imports the effectful node:net server API without the required effecttsgo/node-builtin-import handling.
🟡 MINOR	apps/cli/src/commands/start/start.lifecycle.e2e.test.ts:198	test-reliability	claude	Moving stack teardown into the test scope makes cleanup consume the test timeout, but the previous timeout budgets were retained without a cleanup allowance.
🟡 MINOR	apps/cli/src/commands/start/start.lifecycle.e2e.test.ts:309	resource-lifecycle	codex	The poisoned-image cleanup finalizer is registered only after the Docker build completes, leaving an interruption window in which the tag can remain on the runner.
⚪ NIT	apps/cli/tests/helpers/cli.ts:786	api-design	claude	runDockerEffect embeds its truncation marker directly in returned output, while callers also treat those fields as machine-readable Docker output.
⚪ NIT	apps/cli/src/commands/start/start.slim-images.e2e.test.ts:40	readability	claude	makeProject references SLIM_ENV before its declaration, relying on delayed generator execution after module initialization.

Refuted findings (kept for transparency, not posted as review comments)

• apps/cli/src/commands/start/start.lifecycle.e2e.test.ts:239 (error-handling): The proxy-address invariant is incorrectly thrown as a defect instead of returned as StartE2eSetupError.
  Refuted: After the successful TCP listen callback, server.address() must be a TCP AddressInfo: null denotes a non-listening server and a string denotes an IPC server. The checked branch is therefore an impossible invariant, which trusted/CLAUDE.md:59-62 permits treating as a defect.
• apps/cli/tests/helpers/docker-image.unit.test.ts:247 (test-organization): Placing runDockerEffect tests in docker-image.unit.test.ts violates a repository convention requiring the test basename to match its module.
  Refuted: The test uses the required .unit.test.ts suffix and is colocated in the same helper directory as cli.ts; the trusted convention does not impose the claimed basename rule.
• apps/cli/src/commands/start/start.slim-images.e2e.test.ts:27 (duplication): The two start E2E suites should extract their similar setup error, port override, and project helper into a shared helper.
  Refuted: The helpers are not verbatim because their teardown environments differ, and trusted/CLAUDE.md:171 explicitly permits some test duplication rather than unnecessary abstractions. No behavioral defect results from the local definitions.

Stats

Claude findings: 8 · Codex findings: 3 · Confirmed: 7 · Refuted: 3 · Uncertain: 0

---

Models: claude-opus-5 + gpt-5.6-sol · Trigger: auto · Workflow run

This review runs once per PR. A maintainer can request another with a /ai-review comment.