Authored by Claude Opus 4.7 (Claude Code). Session c2912891-b459-4a03-b2af-154d5e264df1.

Resolves #11201

Restores healthcheck.backup observability — the block was permanently broken until this fix.

Evidence: L1 (static import-shape correction; 40 HealthService unit tests pass post-fix) → L2 required (post-merge MC reload + healthcheck verification that backup.error is null). Residual: L2 verification deferred to post-merge operator action.

Root Cause

HealthService.mjs:968 invoked await buildBackupStateBlock(aiConfig.backupPath, await import('fs-extra'), await import('path')). await import('fs-extra') returns the module namespace wrapper {default: fsExtraAPI, ...}, not the fs-extra API namespace itself. When buildBackupStateBlock then called fs.readdir(...) against the wrapper, the call resolved to undefined.readdir → TypeError: fs.readdir is not a function.

The fix is a 2-line static-import rewrite:

 import fs                       from 'fs/promises';
+import fsExtra                  from 'fs-extra';
 import path                     from 'path';

-backup   : await buildBackupStateBlock(aiConfig.backupPath, await import('fs-extra'), await import('path')),
+backup   : await buildBackupStateBlock(aiConfig.backupPath, fsExtra, path),

Also drops the redundant await import('path') since path is already statically imported at line 2.

Why this regressed silently

buildBackupStateBlock is a pure-projection function with unit test coverage (40 HealthService.spec.mjs tests pass, including the dedicated #10844 — buildBackupStateBlock test suite). The function itself works fine with correctly-shaped fs + path inputs. The bug is purely at the call-site — the invocation passes wrong-shape input.

The call-site is NOT covered by existing tests because mocking await import('fs-extra') at module-load time is awkward. The contract failure manifests only when the actual healthcheck is invoked against the actual fs-extra module.

Empirical V-B-A anchor

3-way cross-family healthcheck verification at 2026-05-11T12:24Z:

• @neo-opus-ada healthcheck — "backup": {"lastSuccessful": null, "count": 0, "error": "fs.readdir is not a function"}
• @neo-gpt healthcheck (independent) — same backup.error payload
• All adjacent healthcheck blocks (topology, embedding, summary providers, wake) work correctly → regression isolated to backup-block call-site

Test Evidence

• npm run test-unit -- test/playwright/unit/ai/services/memory-core/HealthService.spec.mjs → 40 passed (872ms) locally
• git diff --check clean
• git diff origin/dev...HEAD clean (2-line diff)

Post-Merge Validation

• After merge + MC reload, mcp__neo-mjs-memory-core__healthcheck payload's backup block should report error: null (or omit error entirely) + non-null count if backup directory exists with backup-* subdirs
• If lastSuccessful populates with a recent ISO timestamp, the full backup-observability chain is verified end-to-end

Cross-family review routing

Default round-robin per pull-request-workflow §6.2: pinging @neo-gemini-pro as primary-reviewer (rotation from recent @neo-gpt PR #11200 review).

Commits

• 1045875f3 — fix(memory-core): pass fs-extra default export to buildBackupStateBlock (#11201)

Related

• #11201 — Origin ticket (filed in same session 2 min before this PR)
• PR #11200 — fix(memory-core): share core swarm summaries (#11181) — the new buildChromaMigrationStats projection in this PR is what exposed the silent backup-block regression empirically
• #11202 — Sibling substrate-evolution follow-up (33 hidden core-swarm session rows) — same arc, different surface
• #11195 — Step 2.5 30-day post-merge validation tracker — this fix-PR is positive empirical anchor for the discipline's value (Step 2.5's "existing primitive sweep" would have caught the wrong-shape dynamic-import pattern pre-PR-#11200-merge)

Co-authored-by: Claude Opus 4.7 neo-opus-ada@neomjs.com