Openclaw Test Performance

Codex adaptation of the pinned OpenClaw source. Repository-specific commands apply only to a verified OpenClaw checkout; portable debugger commands apply to their selected process.

Measure the real suite shape

Resolve the selected checkout and scoped agent/plugin/channel/Gateway test instructions. Read the packaged source for grouped reports, extension batches, import-duration flags, profiling and package-boundary commands. Verify these entrypoints exist in the current checkout before running them.

Record a like-for-like baseline: command/config, exact commit, worker budget, machine/load, wall time, test/import time and peak RSS. Start with the smallest relevant file or plugin batch; collect broader ranking only for an explicit performance question that needs it. Distinguish harness noise and shared-worker module retention from a test body hotspot or app-object leak.

Repair the owning metadata/import/lifecycle/fixture cause, preserve the contract coverage and rerun the same workload. Do not delete expensive proof merely for speed. For growing memory read $openclaw-test-heap-leaks; for packaging-sensitive plugin changes read $openclaw-release-plugin-testing. Use $openclaw-testing and existing $openclaw-crabbox for runner choice.

Preserve dependency-install ownership and separate Vitest cache paths across concurrent checkouts. Missing dependencies require the applicable install authorization; upstream pnpm-install examples are not a grant. Do not benchmark a shared live Gateway under source-test authority.

Report before/after seconds, CPU and peak RSS, coverage changes, exact commands/commits and residual noise. A gain measured under different machine/load/worker conditions is not a proven optimization. Commit/push only when the task authorizes them.

Selective installation

This workflow requires these installed companion skills: $openclaw-testing. The installer does not resolve them automatically; select these names together or verify they are already installed.