pre-push-gate-debugging

Read the gate's own logs first: the failure block prints a reproduce command for the failing lane and a log directory with per-lane output and a progress.log timeline. Re-run only that lane for fast iteration instead of the full gate.

Use the reproduce command printed in the failure block, typically `bun <script> <lane>` such as the api-e2e lane. Only re-run the full gate once the single lane passes, since whole-gate runs can take five minutes or more.

Piping push output buffers everything, so the lefthook banner never appears and the running gate looks like a network stall. Re-run the push unpiped or in the background and check the first lines for the hook banner.

Reused dev stacks carry emulator state across runs, so durable orchestration instance IDs taken by a prior run cause 409s or silent dedupes later. Fix it by killing the reused stack so the lane starts fresh ephemeral volumes.

A bare worktree has no node_modules, so tooling lanes fail with missing type definitions, and a broken worktree .venv can shadow the main checkout's good one. Symlink node_modules from the main checkout and delete the stray venv.

Use `VERIFY_SKIP=<lane>` when a lane is red due to another task's uncommitted files, after proving your diff is clean. `git push --no-verify` skips every lane and is only appropriate when the user has explicitly waived the gate.