webhookWorkflow times out on '- stable' local e2e lanes: run stalls in 'running' after all hooks received

Stable-lane e2e jobs (E2E Local Dev/Prod Tests (<app> - stable)) are failing frequently across apps with e2e > webhookWorkflow timing out at 120s. Every main push run today (2026-08-03/04) failed at least one such lane on it — apps rotate (astro, sveltekit, nextjs-webpack, express, nest, nitro), astro most often. Examples: 30866461896, 30863551393, 30829099358. On PR #3314 (no package-code changes, based on main @ 89ede82fa) astro - stable failed on it 5 rerun attempts in a row — deterministic there, while other apps' lanes pass/fail intermittently.

Note on naming: - stable in these job names is the framework toolchain label from scripts/create-test-matrix.mjs (runLabel: canary ? 'canary' : 'stable'), not the stable release channel — these jobs run the repo's current packages (job logs show Workflow CLI v5.0.0-beta.39). So this is a hang in current main code.

Failure shape

The run gets all the way through hook delivery and the built-in response steps, then stalls at running forever (from the vitest diagnostics on attempt 4, run wrun_01KZ58CGGNW6HP56G9B1K7T0SR):

+0.0s  run_created
+0.0s  run_started
+0.1s  hook_created  x3
+1.0s  hook_received [..W]
+1.1s  hook_received [..X]
+1.1s  step_started/created (__builtin_response_text) [..1Z]
+1.2s  hook_received [..Y]
+1.2s  step_completed (__builtin_response_text) [..1Z]
+1.3s  step_started/created (__builtin_response_text) [..20]
+1.4s  step_completed (__builtin_response_text) [..20]
(nothing further; status stays 'running' until the 120s test timeout)

All three webhooks (default / static / manual respondWith) receive their payloads and both response-body steps complete, but the workflow never progresses to completion. webhookWorkflow and its workflow source were last touched well before today, so the trigger is likely a recent runtime change (the VM-retention work #3046/#3047 landed in this window, though failures appear in at least one run predating them — worth a bisect with WORKFLOW_RETAINED_VM=0 as a first probe).

🤖 Generated with Claude Code

The failure is exercised by e2e > webhookWorkflow, with matrix labels in scripts/create-test-matrix.mjs; start by reproducing the astro - stable lane and inspecting the listed vitest diagnostics. Use WORKFLOW_RETAINED_VM=0 as the first comparison or bisect probe; done means webhookWorkflow reaches completion and the affected stable lanes no longer time out.