Closes #104471

What Problem This Solves

Canonical pull-request CI could wait more than six minutes on one compact Node test job because the serial tooling config was packed into a broad whole-config runner.

Why This Change Was Made

The compact PR planner now balances whole-config groups deterministically and stripes the 329-file serial tooling config into three near-equal groups. The Docker and isolated tooling configs remain explicit whole-config groups with coverage invariants.

This changes only canonical pull-request admission: main pushes, manual dispatches, release validation, tooling-isolated coverage, and tooling-Docker coverage retain their existing topology.

The compact plan grows from 18 to 20 Node jobs. That two-job increase removes the dominant serial critical path while remaining trivial against the live 10,000 runner-registration bucket.

The wider shard coverage also exposed and fixed three latent test couplings: a stale direct-import assertion, a child config probe inheriting the parent shard filter, and a deadline assertion depending on a live millisecond clock.

User Impact

Maintainers and contributors get materially faster pull-request feedback without dropping test coverage or changing release validation behavior.

Evidence

• Baseline CI: https://github.com/openclaw/openclaw/actions/runs/29152646175
  • compact small whole jobs: 224s, 403s, 77s
  • compact critical path: 403s
• First balancing pass: https://github.com/openclaw/openclaw/actions/runs/29153077047
  • compact small whole jobs: 189s, 382s, 160s
  • only 21s better at the critical path; rejected as insufficient on its own
• Final exact-head CI: https://github.com/openclaw/openclaw/actions/runs/29154172407
  • SHA: e60eae386a1f2e3bed27522995cf3c5c67e4b9b0
  • 58/58 jobs completed successfully
  • compact tooling-bearing jobs: 156s, 211s, 188s
  • compact small whole jobs: 188s, 129s, 168s
  • slowest compact job: 300s, down 103s / 25.6% from baseline
  • former 403s small-whole bottleneck: 188s, down 215s / 53.3%
• Blacksmith Testbox tbx_01kx8jyk8e89h303dqtv905pbr
  • 69/69 focused planner and workflow-guard tests passed
  • 34/34 exposed regression tests passed
• node scripts/check-workflows.mjs
• node scripts/docs-list.js
• targeted oxfmt --check
• git diff --check
• fresh branch autoreview: no accepted/actionable findings; explicitly verified Docker and isolated tooling coverage