kandev/.agents/skills/e2e/references/resource-safety.md

E2E and Frontend Test Resource Safety

Keep local browser runs bounded. The managed and raw E2E entry points enforce one Playwright worker per shard. The guard recognizes --workers, --workers=N, -j N, -j=N, and compact -jN forms, so a worker override cannot bypass the limit by changing CLI spelling. The managed runner allows at most three local shards, and lowers that limit when available memory is below 12 GiB or 6 GiB. Its available-memory value is the minimum of host MemAvailable and the remaining cgroup v1 or v2 memory limit when the process runs inside a container. This prevents host memory reporting from authorizing more shards than the container can hold.

Use pnpm e2e:run, guarded pnpm e2e:raw, or make test-e2e for ordinary verification. Reserve direct Playwright commands, all-worker overrides, and overlapping full suites for deliberate pressure or guard reproduction. Keep those experiments to one worker or shard by default and record the reason.

Start with the default single shard; use two or three only when the host has capacity for separate Go backends, SPA processes, Chromium instances, and mock agents. If a command is rejected, reduce its shard or worker count. Do not silence the guard during ordinary verification.

Do not run a local full Vitest suite with VITEST_MAX_WORKERS=100%, another all-worker value, or several overlapping full-suite processes. Local Vitest configuration clamps unsafe worker overrides to its 20-percent budget. Set KANDEV_E2E_ALLOW_UNSAFE_PARALLELISM=1 or KANDEV_ALLOW_UNSAFE_TEST_PARALLELISM=1 only for a deliberate, monitored resource experiment, and record the resource limit and result.