Openclaw Test Heap Leaks

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.

Separate test retention from runtime leakage

Read $openclaw-test-performance for current workload/profiling routes, then the packaged source for heap collection, snapshot comparison and dedicated runtime validation. Match the real entrypoint, configuration, worker budget and PID before classifying growth.

RSS is triage evidence, not proof of a leak. A sampling .heapprofile is not a .heapsnapshot. Compare equivalent snapshots from the same PID, inspect positive allocation families, and use dominators/retainers to distinguish transformed module graphs from retained app objects, timers, handles, buffers and caches.

The bundled read-only snapshot helper is references/upstream/scripts/heapsnapshot-delta.mjs. Invoke it with Node and two task-owned snapshots, or its documented --lane-dir selection. It does not need to be installed into the OpenClaw repo. Heap captures can contain private data: keep them local/scoped and out of public changes.

For a runtime closure fix, use the selected checkout's dedicated producer described in the source, including its pre-fix/synthetic-leak sensitivity controls. Do not retrofit an unrelated harness merely to obtain a PASS. Fix process lifetime or cleanup at the evidenced owner, then repeat the same producer/workload and normal test proof.

Report same-PID/source identities, retained families, retainer support, overhead and leak classification as proven, likely or inconclusive. A lower peak RSS alone does not establish the runtime objects were released.

Selective installation

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