fix(cli): handle piped config pull input (CLI-2425)

[7ttp]

TL;DR

Prevents supabase config pull from crashing when confirmation input is piped
the command could reach promptYesNo without providing stdinLayer

which is now fixed by:
Providing stdinLayer at the command boundary and covering the piped input path with a binary regression test.

Ref:

• closes supabase config pull crashes with Service not found: supabase/runtime/Stdin when stdin is piped into the command. #6585

[github-actions]

🤖 AI Review

The production fix is correct, but three review concerns are confirmed: missing Stdin wiring remains compile-time-permissible, the new invariant overstates which prompt paths need stdinLayer, and the E2E assertion is not Windows-portable. The proposed timeout-flakiness concern is refuted because stdin is written and closed before the command reaches the prompt. Codex completed its review and reported no findings.

Findings

Severity	Location	Category	Sources	Claim
🟡 MINOR	apps/cli/src/commands/config/pull/pull.command.ts:79	maintainability	claude	Missing stdinLayer wiring can still type-check because Stdin is permitted at the root boundary even though the root runtime does not provide it, allowing another prompt-capable command to recreate the runtime defect.
⚪ NIT	apps/cli/AGENTS.md:72	documentation	claude	The new invariant incorrectly says every production path reaching promptYesNo provides stdinLayer, although some safe paths reach the function without providing that layer and cannot enter its Stdin-reading branch.
⚪ NIT	apps/cli/src/commands/config/pull/pull.e2e.test.ts:84	test-quality	claude	The prompt assertion hardcodes a POSIX path separator and will fail on Windows despite correct CLI behavior.

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

• apps/cli/src/commands/config/pull/pull.e2e.test.ts:74 (test-flakiness): The decline assertion is flaky because a loaded runner may miss promptYesNo's 100 ms stdin window and apply the default affirmative answer.
  Refuted: The cited timeout exists, but the proposed loaded-runner race is countered by the harness queuing and closing stdin immediately after spawn. The pipe retains that complete line until the command eventually opens its lazy reader, so pre-prompt runner delay does not consume the 100 ms read window.

Stats

Claude findings: 4 · Codex findings: 0 · Confirmed: 3 · Refuted: 1 · 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.