* feat(scripts): serialize pr prepare gates and add remote testbox test gate
Concurrent scripts/pr gate runs across .worktrees queued on the shared
heavy-check lock mid-test: the queued run's children hit the 10-minute
lock timeout while its unlocked build stage piled CPU load onto the
holder's vitest shards, which then stalled past the 120s no-output
watchdog and were SIGTERMed with zero real test failures (observed
landing PRs #99935/#100026 on a loaded maintainer Mac).
- scripts/pr-gates-lock.mjs holds the shared heavy-check lock for the
whole local gate block; gate stages inherit the existing *_LOCK_HELD
child contract, so concurrent gate runs now queue as whole units
before their first command.
- OPENCLAW_PR_GATES_REMOTE=testbox runs the full-suite pnpm test gate on
a Blacksmith Testbox via scripts/crabbox-wrapper.mjs (same delegation
as check:changed). The tbx_ lease id and Actions run URL from the
crabbox --timing-json report are recorded in .local/gates.env
(REMOTE_GATES_*) and .local/prep.md. Local remains the default;
pnpm build/check stay local.
Formatting verified with the primary checkout's oxfmt (hook bypassed:
linked worktree has no hydrated node_modules).
* fix(scripts): refresh the gate stamp when lease-retry gates rerun for a rebased head
The lease-retry path reran build/check/test for the rebased prep head but
left .local/gates.env describing the pre-push head, so prep.md/prep.env
attributed stale evidence (including the new remote testbox lease id) to
the pushed commit. Extract write_gates_env_stamp as the single stamp
writer, rewrite the stamp from the retry path for all modes, and re-source
gates.env in prepare_push after the push settles.
Found by autoreview (codex/gpt-5.5); formatting verified with the primary
checkout's oxfmt (hook bypassed: linked worktree has no node_modules).
* fix(scripts): harden remote PR gate evidence
* fix(scripts): serialize complete gate setup
* fix(scripts): clear stale docs gate proof
* fix(infra): stop package-root walks at node_modules boundaries
Nested git worktrees (.worktrees/<pr>, .claude/worktrees/*) have no local
node_modules, so vitest workers run with argv1 inside the enclosing
checkout's node_modules. resolveOpenClawPackageRoot* walked up past that
boundary, resolved the enclosing checkout as the openclaw package root,
and bundled plugin discovery loaded the ancestor's stale extension
manifests, failing ~12 test shards and the scripts/pr prepare-run test
gate whenever the primary checkout sat on a different branch.
A package.json above a node_modules directory belongs to the workspace
that installed the tooling, never to the package owning the running code,
so the ancestor walk now stops there. Installed layouts
(node_modules/openclaw, .bin shims, global installs) match at or below
the boundary and are unaffected.
* fix(scripts): pin bundled plugin discovery in nested PR worktree gates
scripts/pr gates run pnpm test inside .worktrees/<pr>, which resolves
vitest tooling from the primary checkout's node_modules. PR branches that
predate the openclaw-root node_modules-boundary fix would still discover
the primary checkout's stale bundled plugin manifests, so pin
OPENCLAW_BUNDLED_PLUGINS_DIR to the worktree's own extensions before the
build/check/test gates.
One-time maintainer-authorized bootstrap merge for the release-gate verifier policy. Exact hosted CI and all supporting workflow gates passed on 66133de419.
The hardcoded `OPENCLAW_VITEST_MAX_WORKERS=4` default in gates.sh
short-circuits the host-aware scheduling introduced in c247e366.
`resolveLocalVitestScheduling` sees the explicit override and returns
maxWorkers=4, which falls below the >= 5 threshold required by
`shouldUseLargeLocalFullSuiteProfile`, so every machine—regardless of
resources—gets the DEFAULT profile (4 shard parallelism) instead of
the LARGE profile (10 shard parallelism).
Drop the hardcoded default so `test-projects.mjs` can detect actual
host resources and pick the appropriate profile automatically. When
the user explicitly sets OPENCLAW_VITEST_MAX_WORKERS, forward it as
before.
