feat(test-harness): per-worker shadow-repo + git support

[nick-inkeep]

Summary

• Flips the dev-plugin test-isolation guard so Playwright workers spawn with a real per-worker shadow repo at <tmpdir>/.git/open-knowledge/ (FR1) instead of silently disabling git, and adds a withShadow: true opt-in to the Tier 1 integration harness (FR2) — together unlocking automated test coverage for timeline / save-version / attribution / reconciliation surfaces that were previously uncoverable.
• Seals seven projectDir couplings in hocuspocus-plugin.ts via a single module-level projectRoot binding (FR4 / D12), preventing /api/save-version and adjacent surfaces from writing real ok/v<N> tags + commits into the developer's actual OK repo during local test runs (the worst-case audit finding).
• Closes a latent FR1 race (commit b9f1fb28, post-QA) where agent-write/save-version/rollback requests arriving within ~100-200 ms of bun run dev startup silently lost their L2 shadow commit. Now: api-extension awaits the dev plugin's shadowReadyPromise before reading shadowRef.current. End-to-end Playwright validation in _qa-shadow-default-on.e2e.ts proves the full path now works without artificial delays.
• Internal test-harness change only — zero customer-visible behavior, no CI runner config change, production dev path byte-for-byte identical.

Spec: specs/2026-04-22-per-worker-shadow-repo-test-harness/SPEC.md (15 LOCKED decisions, 8 functional requirements).

Changeset: bumps @inkeep/open-knowledge-server + @inkeep/open-knowledge-app only (the two packages with actual source changes; core, cli, desktop are unchanged in this PR per git diff main...HEAD --stat).

Key decisions

D2 — Tier-appropriate defaults	Playwright tier defaults shadow ON (full-app topology matches production); Tier 1 integration tier defaults shadow OFF, opt-in via { withShadow: true }. The 37/38 shadow-orthogonal Tier 1 tests stay untouched; only the 1/38 that actually asserts shadow state pays the ~50-200 ms init cost.
D12 — Single projectRoot binding	One module-scope const projectRoot = isTestIsolated ? CONTENT_DIR : PROJECT_ROOT threaded through all 7 PROJECT_ROOT consumers in hocuspocus-plugin.ts. Audit found the previous per-site decomposition missed 3 sites (L245, L250, L275); the single-binding pattern is the precedent so any future addition is obviously wrong without it.
D13 — Broadened fail-fast	Under isTestIsolated: true, handleDevShadowInitError now exits non-zero on ANY shadow-init throw (not just ProjectGitInitError). Production path unchanged — retains existing degraded-warn for non-ProjectGitInitError. Closes a silent-degradation hole that would otherwise let tests pass without producing the shadow refs they assert on.
D14 — Acceptance bar	All four FR8 acceptance tests (T1, T2, T3, T6) ship with this spec as Must-tier. Greenfield posture rejects Could-tagged acceptance. T6 (write/read round-trip via /api/history) is the regression guard for the L245 leak D12 just sealed.
D15 — SPEC 2026-04-21 §Q3 corrigendum	The prior shadow-repo spec closed Q3 with "No harness change needed." That closure is invalidated by this PR; per CLAUDE.md post-ship corrigendum protocol, an italic/bracketed/dated breadcrumb pointing here was added to that spec's §Q3 row inline.
Post-QA: shadowReadyPromise gate	ApiExtensionOptions gains an optional shadowReadyPromise?: Promise<void> field. The dev plugin captures runDevShadowInit's promise (previously void-discarded) and passes it through. flushDocToGit (covers all 4 agent-write handlers) + handleSaveVersion + handleRollback await it before reading shadowRef.current. Non-breaking; Tier 1 / CLI callers omit the field and behavior is unchanged.

What changed (FR coverage)

FR	What it does	Where
FR1	Playwright workers default-on shadow under OK_TEST_CONTENT_DIR	packages/app/src/server/hocuspocus-plugin.ts (drop if (!isTestIsolated) guard; flip gitEnabled to true unconditional; capture shadowReadyPromise and pass through)
FR2	createTestServer({ withShadow: true }) opt-in	packages/app/tests/integration/test-harness.ts (new withShadow?: boolean option, shadowDir?: string exposure)
FR3	Drop redundant gitEnabled: false from migrated tests	symlink-alias.test.ts:46, provider-pool-reconnect.test.ts:74
FR4	Single projectRoot binding sealing 7 PROJECT_ROOT couplings	packages/app/src/server/hocuspocus-plugin.ts:121 (the binding) + 7 threaded sites
FR5	Fail-fast on ALL shadow-init errors under isolation (D13)	packages/app/src/server/dev-shadow-init.ts (broadened handleDevShadowInitError) + dev-shadow-init.test.ts (3 new tests)
FR6	Migrate persistence-fan-out.test.ts (app integration tier) to harness opt-in	packages/app/tests/integration/persistence-fan-out.test.ts (delete hand-fork; clearContributors→swapContributors; -130 net lines of duplicated boot code)
FR7	Auto-wire swapContributors() lifecycle	packages/app/tests/integration/test-harness.ts (setup + cleanup; serial-execution constraint documented as STOP rule)
FR8	T1-T3 + T6 acceptance tests (all Must-tier)	packages/app/tests/integration/shadow-harness-{agent-write,save-version,external-write,history-roundtrip}.test.ts (4 new files)
FR1 (post-QA)	Race gate so Playwright agent writes deterministically produce shadow refs	packages/server/src/api-extension.ts (shadowReadyPromise option + 3 await sites) + packages/app/tests/stress/_qa-shadow-default-on.e2e.ts (regression guard)

Plus:

• Server package re-exports — swapContributors + destroyShadowRepo added to packages/server/src/index.ts (US-001 substrate).
• requireShadowDir helper at test-harness.ts:506 for typed-boundary error messages when callers forget withShadow: true (with explicit unit test coverage in test-harness-helpers.test.ts after review-cloud pass 1).

Verification

QA verdict: GO — full plan at tmp/ship/qa-progress.json (gitignored, generated locally).

Story	Tested	Gaps	Fidelity
US-001: server exports	1/1 ✅	—	shell
US-002: projectRoot binding (FR4)	2/2 ✅	—	shell
US-003: dev-plugin shadow + fail-fast (FR1, FR5)	2/2 ✅	—	shell
US-004: harness withShadow + lifecycle	5/5 ✅	—	shell
US-005: migrated persistence-fan-out	1/1 ✅	—	shell
US-007: T1-T3+T6 acceptance	4/4 ✅	—	shell
US-008: corrigendum	1/1 ✅	—	shell
FR1 e2e (post-QA, this commit)	1/1 ✅	—	browser (Playwright + simpleGit assertion)
(integration baseline)	1/1 ✅	—	shell
(perf baseline)	1/1 ✅	—	shell

Highlights:

• ✅ FR1 end-to-end via real Playwright worker — _qa-shadow-default-on.e2e.ts passes in ~4s. Asserts shadow exists at <contentDir>/.git/open-knowledge/, agent write produces refs/wip/main/agent-<id> with wip: subject and ok-actor: body within ~5s, no artificial delays. Without the post-QA fix, this test fails deterministically (verified during /debug).
• ✅ FR4 invariant proof — full bun run check; pre-suite vs post-suite tag count: 0→0; HEAD: unchanged; git status --porcelain: empty; .open-knowledge/cache/ not created in dev's repo.
• ✅ All 4 FR8 acceptance tests pass — shadow-harness-{agent-write,save-version,external-write,history-roundtrip}.test.ts green (8 pass / 0 fail combined with the migrated persistence-fan-out.test.ts).
• ✅ D13 broadened fail-fast — dev-shadow-init.test.ts 11/11 including new tests for plain Error + non-Error thrown values under isTestIsolated: true.
• ✅ Cross-worker isolation — verified via docs-open.e2e.ts --workers=4 (19/20 pass; 1 unrelated pre-existing flake on F10 source-editor warm-swap, confirmed reproducible on pre-fix git stash state).

Pre-merge checklist for human reviewer:

• Production dev-server smoke (manual): cd to your actual OK repo (not this worktree), run bun run dev, perform a few agent writes, trigger Save Version. Confirm shadow at PROJECT_ROOT/.git/open-knowledge/, ok/v<N> tag in PROJECT_ROOT/.git/, behavior identical to pre-PR.
• Read-through of AGENTS.md updates: 7b5af64c documents the new withShadow option, dev-mode shadow default-on, and the PROJECT_ROOT STOP rule. Verify the section reads clearly enough that the next test author can write a shadow-asserting test in <10 min from docs alone.

Out of scope (per SPEC §16)

• packages/server/src/persistence-fan-out.test.ts (server-tier sibling — deliberately independent of the app integration harness)
• CRDT observer code, shadow-lock / server-lock primitives, production CLI behavior (no behavior changes)
• PR Editor asset + embed surface (streaming uploads) #270's "unify dev-plugin + createServer" refactor (D1 — separate future spec; merge-order protocol documented in §13)
• T4 (rollback E2E) + T5 (TimelinePanel UI render) — downstream feature tests (NG6); harness now supports them
• Production CLI race in bootServer/standalone.ts (analogous to the dev plugin race fixed by b9f1fb28) — future spec; bounded customer impact (production users navigate UI before writing) and shadowReadyPromise is now a one-line addition at bootServer when needed.

Future considerations

Surfaced opportunities (from Phase 4 docs subprocess):

• AGENTS.md / CLAUDE.md duplication — the file has two near-identical halves with drift on precedent counts (25 vs 27), V0-14 status, and a few label names. This PR's docs additions correctly mirror into both halves (verified byte-identical). The underlying duplication predates this PR and warrants a separate docs-housekeeping pass.
• Child claude -p subprocess MCP visibility — nested subprocesses (Phase 4 docs, Phase 6 qa-plan) do not inherit parent-session MCP server config (no mcp__open-knowledge__* tools available in the child). Used the documented escape hatch ("Open Knowledge MCP unavailable") per CLAUDE.md.

Deferred (in-scope follow-up, not this PR):

• Per-server-instance contributor-tracker state — long-term fix for the pendingContributors module-level state. Today's STOP rule (no test.concurrent() in withShadow:true files) is the convention enforcing serial execution.
• Expose persistence.flushPendingGitCommit() on ServerInstance — would enable T2/T6 acceptance tests to drain L2 deterministically without polling. Out of spec scope (touches packages/server/src/standalone.ts per SPEC §16 EXCLUDE) but a single-line addition next time ServerInstance's shape opens.
• handleDevShadowInitError category-extraction (cloud reviewer Consider, declined) — speculative refactor for hypothetical future error types; the 3-branch if/else is the right shape for 3 cases.
• Production CLI race fix at bootServer — same shape as the dev plugin fix in b9f1fb28; separate spec when prioritized.

Review-cloud iteration history

Pass 1 (commit 120ccc13):

• ✅ Trimmed changeset to 2 actually-changed packages (reviewer flagged 1 of 3 unchanged)
• ✅ Added test-harness-helpers.test.ts with 3 tests covering requireShadowDir error path
• ✅ Added point-in-time disclaimer to evidence/projectdir-couplings.md
• ❌ Declined extract-error-categorization Consider (per reviewer's own "no change needed" rationale)

Pass 2 (cloud review of 120ccc13): ✅ APPROVE with 0 issues.

Pass 3 fix (commit b9f1fb28, post-QA validation):

• ✅ Discovered FR1 race via /debug while writing end-to-end Playwright validation test
• ✅ Fixed via shadowReadyPromise gate in api-extension (3 sites: flushDocToGit + handleSaveVersion + handleRollback)
• ✅ Permanent regression-guard test in _qa-shadow-default-on.e2e.ts

🤖 Generated with Claude Code

Co-Authored-By: Claude Opus 4.7 (1M context) noreply@anthropic.com

[changeset-bot]

🦋 Changeset detected

Latest commit: 4c94ed7

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 5 packages

Name	Type
@inkeep/open-knowledge-server	Patch
@inkeep/open-knowledge-app	Patch
@inkeep/open-knowledge	Patch
@inkeep/open-knowledge-desktop	Patch
@inkeep/open-knowledge-core	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[inkeep-internal-ci]

PR Review Summary

(1) Total Issues | Risk: Low

🟡 Minor (1) 🟡

Inline Comments:

• 🟡 Minor: .changeset/per-worker-shadow-repo-test-harness.md:7 Changeset includes @inkeep/open-knowledge-core: patch but that package has no changes

💭 Consider (2) 💭

💭 1) test-harness.ts Add explicit test for requireShadowDir error path

Issue: The requireShadowDir() helper throws a descriptive error when shadowDir is undefined, but there's no explicit test for this path.

Why: While the error message is good and the helper is simple, an explicit test would document the expected behavior and catch regressions if the message changes.

Fix: Consider adding a small test case in the harness tests:

test('requireShadowDir throws descriptive error when shadowDir undefined', () => {
  const server = { shadowDir: undefined } as TestServer;
  expect(() => requireShadowDir(server)).toThrow(/withShadow: true/);
});

Refs:

• test-harness.ts:506 — helper implementation

💭 2) dev-shadow-init.ts Consider extracting error categorization for future extensibility

Issue: The handleDevShadowInitError function uses an if/else if/else chain that's correct and clear, but as more error types need special handling, this could grow.

Why: The current structure is appropriate for 2-3 branches but the pattern could be extended to a category-based approach if needed in the future.

Fix: No change needed now — the current implementation is clean and the if/else structure correctly expresses the mutual exclusion. Just noting this as a potential evolution path if more error types emerge.

Refs:

• dev-shadow-init.ts:35-54 — error handling logic

🧹 While You're Here (1) 🧹

🧹 1) evidence/projectdir-couplings.md Line numbers may have drifted

Issue: The evidence file documents specific line numbers for projectRoot consumers, but these may drift as the codebase evolves.

Why: Stale line numbers in documentation can mislead future readers investigating the STOP rule.

Fix: Consider adding a note that line numbers are point-in-time snapshots from the PR, or updating them if they've already drifted since the evidence was captured.

Refs:

• evidence/projectdir-couplings.md

---

💡 APPROVE WITH SUGGESTIONS

Summary: This is a well-structured, carefully designed internal test infrastructure change. The single projectRoot binding pattern (D12) correctly seals the PROJECT_ROOT leak surface, the fail-fast behavior under test isolation (D13) is properly implemented with clear mutual exclusion in the error handling, and the withShadow harness opt-in is thoughtfully designed with tier-appropriate defaults. The acceptance tests (T1-T3, T6) provide strong coverage of the new capabilities.

The main actionable item is the changeset incorrectly including @inkeep/open-knowledge-core: patch when that package has no changes — removing it will avoid a spurious version bump.

The test patterns are solid: try/finally lifecycle, AggregateError for multi-phase cleanup failures, and the STOP rule documentation for test.concurrent() with withShadow: true is well-placed. Nice work on the comprehensive spec and evidence artifacts.

Discarded (3)

Location	Issue	Reason Discarded
test-harness.ts	L2 drain via destroy() is implicit	Working correctly — destroy() is documented to flush L2 and the pattern is consistent across all acceptance tests
AGENTS.md	Two halves have drift in precedent counts	Pre-existing issue outside this PR's scope — the PR correctly mirrors new content in both halves
server-observers.ts	No changes needed	Standards reviewer found no issues meeting confidence threshold

Reviewers (3)

Reviewer	Returned	Main Findings	Consider	While You're Here	Inline Comments	Pending Recs	Discarded
pr-review-tests	5	0	2	0	0	0	3
pr-review-standards	0	0	0	0	0	0	0
pr-review-devops	3	0	0	1	1	0	1
Total	8	0	2	1	1	0	4

Note: Standards reviewer found no issues meeting confidence threshold — this is a clean implementation.

[nick-inkeep]

Review pass 1 — assessment of review-body suggestions

Inline thread on .changeset/... resolved separately (see thread). Top-level summary for the 3 review-body items:

💭 Consider 1 — Add explicit test for requireShadowDir error path → ACCEPTED

Implemented in commit 120ccc13 as packages/app/tests/integration/test-harness-helpers.test.ts (3 tests):

1. withShadow: true mention in error message (matches your suggested regex)
2. [test-harness] namespace prefix in error message
3. Happy-path return of the populated shadowDir

The contract is now locked in for the 9 call sites that depend on the descriptive error.

💭 Consider 2 — Extract error categorization for future extensibility → DECLINED (per your own resolution)

Your own rationale: "No change needed now — the current implementation is clean and the if/else structure correctly expresses the mutual exclusion." Agreed. Speculative refactor for hypothetical future error types is over-engineering. The 3-branch if/else if/else is the right shape for 3 cases. Greenfield posture says no deferred tech debt; this isn't tech debt, it's an evolution path that may never be needed.

🧹 While You're Here — Line numbers may have drifted in evidence/projectdir-couplings.md → PARTIALLY ACCEPTED

Verified via sed -n '<line>p' that the evidence file's claimed lines (L123, L136, L175, L207, L213, L220, L225) all match the current hocuspocus-plugin.ts byte-for-byte — there is no drift in this PR's HEAD.

However, your underlying concern (drift over future code evolution) is valid. Added a one-line disclaimer in commit 120ccc13:

Line numbers are point-in-time snapshots from this PR's HEAD ... the projectRoot binding identity (and the AGENTS.md STOP rule that names it) is the durable contract, not these line numbers. To re-verify the enumeration after future changes, run grep -nE 'projectRoot|PROJECT_ROOT' packages/app/src/server/hocuspocus-plugin.ts and reconcile.

This makes the durable contract (projectRoot binding identity + AGENTS.md STOP rule) explicit and gives future maintainers a verification one-liner.

---

bun run check green: 216 integration pass / 0 fail (3 new tests added).

[inkeep-internal-ci]

PR Review Summary

(0) Total Issues | Risk: Low

This PR adds per-worker shadow-repo support for Playwright workers and a withShadow: true opt-in for the Tier 1 integration test harness. The implementation is thorough, well-documented, and follows established codebase conventions.

Key Observations

D12 Single projectRoot Binding ✅
The single module-level binding pattern at hocuspocus-plugin.ts:121 correctly seals all 7 PROJECT_ROOT coupling sites. The STOP rule in AGENTS.md clearly documents this constraint with references to the evidence file.

D13 Broadened Fail-Fast ✅
The handleDevShadowInitError if/else if/else chain correctly implements mutual exclusion:

• ProjectGitInitError → exit(1) (production AND isolation)
• Any error + isTestIsolated: true → exit(1)
• Non-ProjectGitInitError + production → degraded warn (continue)

Test coverage is comprehensive with 7 test cases covering all branches.

Test Harness withShadow Implementation ✅

• Proper lifecycle management with per-phase try/catch
• swapContributors() correctly wired in setup and cleanup
• destroyShadowRepo() called before rmSync() (correct ordering)
• Error aggregation via AggregateError preserves debugging context
• requireShadowDir() helper provides actionable error messages

Acceptance Tests (FR8) ✅
All four acceptance tests (T1, T2, T3, T6) are well-structured:

• Use plain test() (not test.concurrent()) per the STOP rule
• Use pollUntil for event-driven waits (not fixed sleeps)
• Cover agent write, save-version, external write, and history round-trip paths

🕐 Pending Recommendations (1)

• 🟡 .changeset:7 Changeset included unchanged @inkeep/open-knowledge-core package — addressed in commit 120ccc13

---

✅ APPROVE

Summary: Clean implementation of per-worker shadow-repo test infrastructure. The single prior review finding (changeset including unchanged packages) has been addressed. The code follows established patterns (DI for testability, bracket-prefixed logging, STOP rule documentation), has comprehensive test coverage, and the AGENTS.md documentation accurately reflects the implementation. Ship it! 🚀

Discarded (0)

No findings discarded.

Reviewers (5)

Reviewer	Returned	Main Findings	Consider	While You're Here	Inline Comments	Pending Recs	Discarded
pr-review-standards	0	0	0	0	0	0	0
pr-review-tests	5	0	0	0	0	0	5
pr-review-errors	0	0	0	0	0	0	0
pr-review-devops	0	0	0	0	0	0	0
pr-review-consistency	0	0	0	0	0	0	0
Total	5	0	0	0	0	0	5

Note: pr-review-tests returned 5 observations (all MINOR) that were discarded as they were explicit "no fix needed" observations documenting good patterns rather than issues.

[inkeep-internal-ci]

PR Review Summary

(0) Total Issues | Risk: Low

This is a re-review following commit b9f1fb28 which was pushed after the prior approval. The new commit addresses a race condition where shadow operations (flushDocToGit, handleSaveVersion, handleRollback) could execute before runDevShadowInit completed.

Review of New Commit b9f1fb28

Race Condition Fix ✅

The fix introduces shadowReadyPromise — the promise returned by runDevShadowInit() — and gates all shadow-dependent operations on it:

• hocuspocus-plugin.ts: Captures the init promise and passes it to createApiExtension
• api-extension.ts: Gates flushDocToGit and related operations via Promise.resolve(shadowReadyPromise).then(...)

Design Validation ✅

The "promise never rejects" claim is accurate:

• runDevShadowInit wraps initShadowRepo in a try/catch
• All errors route through handleDevShadowInitError which either exits (test isolation) or logs + continues (production degraded mode)
• The outer .catch() in api-extension.ts is defense-in-depth for unexpected edge cases

Test Coverage ✅

_qa-shadow-default-on.e2e.ts provides regression coverage:

• Validates shadow repo exists immediately after boot
• Drives an agent write with no artificial pre-write delay (the prior race window)
• Asserts the WIP ref appears, proving the shadowReadyPromise gate works

Prior Review Items

All feedback from the prior review has been addressed:

• ✅ Changeset trimmed to only changed packages (@inkeep/open-knowledge-server, @inkeep/open-knowledge-app)
• ✅ requireShadowDir error-path tests added in test-harness-helpers.test.ts

---

✅ APPROVE

Summary: The new commit cleanly fixes a legitimate race condition with sound async gating. The "promise never rejects" design is verified accurate, and the E2E test provides adequate regression coverage. All prior feedback addressed. Ship it! 🚀

Reviewers (3)

Reviewer	Returned	Main Findings	Consider	While You're Here	Inline Comments	Pending Recs	Discarded
pr-review-standards	0	0	0	0	0	0	0
pr-review-errors	0	0	0	0	0	0	0
pr-review-tests	0	0	0	0	0	0	0
Total	0	0	0	0	0	0	0

Note: Focused review on new commit b9f1fb28 since prior approval already covered the main PR content.

[inkeep-internal-ci]

PR Review Summary

(0) Total Issues | Risk: Low

This is a re-review following the pull_request:synchronize event. The latest commit (4c94ed71) was already reviewed and APPROVED in the prior review cycle (2026-04-23T11:05:26Z). All prior feedback has been addressed.

Review Coverage

Reviewer	Findings
pr-review-standards	✅ 0 issues — code quality, async gating pattern, and AGENTS.md STOP rule verified
pr-review-tests	✅ 0 blocking — E2E coverage is appropriate for the race condition fix
pr-review-errors	✅ 0 issues — fail-fast (D13) and degraded-warn paths correctly implemented
pr-review-devops	✅ 0 issues — changeset accurate, AGENTS.md updates complete

Key Validations

1. shadowReadyPromise Race Condition Fix ✅

• The promise gating is correctly applied to all 3 shadow-dependent paths (flushDocToGit, handleSaveVersion, handleRollback)
• The "promise never rejects" claim is accurate — handleDevShadowInitError routes all errors through exit(1) or log+return, never throw
• E2E regression test (_qa-shadow-default-on.e2e.ts) proves the fix end-to-end with no artificial delays

2. Single projectRoot Binding (D12) ✅

• All 9 consumer sites correctly thread through projectRoot instead of raw PROJECT_ROOT
• AGENTS.md STOP rule clearly documents the constraint with evidence file reference

3. Fail-Fast Under Test Isolation (D13) ✅

• The 3-branch if/else if/else chain correctly implements mutual exclusion
• 11 unit tests cover all branches including edge cases (non-Error thrown values)

4. Test Harness Cleanup Lifecycle ✅

• Per-phase try/catch with AggregateError preserves debugging context
• Correct ordering: destroyShadowRepo() before rmSync()
• swapContributors() drain in both setup and cleanup

💭 Consider (1) 💭

💭 1) api-extension.ts / _qa-shadow-default-on.e2e.ts Tier 1 test for shadowReadyPromise gate

Issue: The race condition fix is tested only via E2E (120s timeout), not Tier 1 integration.

Why: The race was discovered in the dev plugin's async init path (hocuspocus-plugin.ts), which isn't exercised by createTestServer (synchronous init). A Tier 1 test would need to inject a delayed shadowReadyPromise via test-only API surface that doesn't exist.

Resolution: The E2E test is appropriate coverage for this specific path. The test proves the full dev plugin → api-extension → shadow commit pipeline works with real Playwright isolation. Future work could expose a test hook, but this is not blocking.

---

✅ APPROVE

Summary: Excellent internal test infrastructure work. The single projectRoot binding seals a real leak vector (PROJECT_ROOT couplings that could write to the developer's checkout during local test runs), the shadowReadyPromise fix closes a legitimate race condition discovered during QA, and the test harness improvements enable shadow-repo coverage that was previously impossible. All prior feedback addressed. Ready to ship! 🚀

Reviewers (4)

Reviewer	Returned	Main Findings	Consider	Inline Comments	Discarded
pr-review-standards	0	0	0	0	0
pr-review-tests	6	0	1	0	5
pr-review-errors	0	0	0	0	0
pr-review-devops	0	0	0	0	0
Total	6	0	1	0	5

Note: Test reviewer returned 6 observations (1 MAJOR downgraded to Consider after validation, 5 INFO/validated-good patterns). All other reviewers found no issues meeting confidence threshold.

[github-actions]

This pull request has been automatically marked as stale because it has not had recent activity.
It will be closed in 7 days if no further activity occurs.

If this PR is still relevant:

• Rebase it on the latest main branch
• Add a comment explaining its current status
• Request a review if it's ready

Thank you for your contributions!