Skip to content

test: qualify native review isolation and failure cleanup - #517

Merged
BunsDev merged 3 commits into
mainfrom
cr8-native-boundary
Nov 12, 2026
Merged

BunsDev merged 3 commits into
mainfrom
cr8-native-boundary

Conversation

@BunsDev

@BunsDev BunsDev commented Nov 12, 2026 •

Copy link
Copy Markdown
Contributor

Connects Code Review’s real Electron/main path to the installed Claude CLI and existing scripted provider. One explicit review reads hostile PR context before forbidden tool/outside-read proposals; the test verifies confinement, one start and observed process exit. The shared fixture retains its direct-run defaults.

A fresh independent review found that callback/shutdown errors could discard observations or skip cleanup. One coalesced correction preserves provider data on callback rejection, copies inputs before shutdown, closes resources independently and records primary and cleanup errors separately. Unknown process-inspection failures now fail qualification. The supplied-token coverage limitation remains documented.

Validation on macOS 27.2.1 arm64, Electron 44.4.5 and Claude 2.1.322:

  • Fresh build and full units: 2,534 passed, 36 skipped, zero failures/cancellations. Complete post-correction logs are retained.
  • Existing direct native controls: six passed, zero skips. Focused checks: four ordinary passes and two intentional expected qualification failures; independent verifiers check each exact fault, retained snapshot/provider observations and provider closure. Each of the three native fixture reviews starts once; all provider wire data and credentials are synthetic.
  • Default opt-out: all six new cases explicitly skipped. The preceding full E2E run passed 520 desktop cases with 12 skips, eight site cases and four web cases. Application/build inputs and the other default E2E sources are unchanged; the corrected native paths were rerun. Final-head hosted qualification is now complete; protected-queue qualification remains required.

Final-head hosted CI passed all twelve observed source checks. The macOS job passed 2,534 unit tests with 36 skips, 523 desktop E2E tests with 14 explicit skips, eight site tests and four web tests; no framework retry markers were observed. The six opt-in native cases remain explicitly skipped in ordinary CI and are supported by the separate local native receipts above. Source iPhone CI passed, without implying physical-device acceptance. All four CI artifacts were preserved with matching API SHA-256 digests and verified ZIP integrity.

The dated receipt retains source hashes, first failures, runtime/viewport and qualification limits. Full #327 adversarial/race acceptance, live-provider/live-write and human gates remain open. Product behavior, feature defaults and release state are unchanged.

Refs #327.

Protected delivery is complete: signed fd80cd49155b449bd3ae7fa3bbd39146d80223ce, tree 33330d75a7d9cffaee7a5adf2bf52bba7614d645, merged on parent c695072 through queue run 43294368140. The actual required macOS job passed 2,576 unit tests with 36 skips, 525 desktop E2E tests with 14 explicit skips, eight site tests and four web tests. Queue iPhone was skipped. Raw logs and both queue artifacts are retained with matching API SHA-256 digests and verified ZIPs; the full merged tree matches an independent merge projection, all four non-status authored files match reviewed source exactly, and the status file includes main integration. The ordinary queue still skips the six opt-in native cases; their local installed-CLI evidence stays separately pinned to source 085b4d5. This delivers the test/receipt slice without accepting the full #327 adversarial criterion, live-provider/live-write, human or release gates.

Drive one explicit Electron review through the installed Claude CLI and a scripted provider, checking hostile context, denied tools and outside reads, scoped environment and one-start process closure. Preserve direct fixture defaults and keep app observations separate from unobserved child diagnostics.

Record local native qualification and its first failures; full E2E, independent review and protected delivery remain pending.
Record actual app runtime and viewport in the opt-in fixture, plus the completed default E2E results and explicit skip limits. Keep review and protected delivery pending.
Capture provider observations even when the app callback rejects, retain inputs before shutdown, and close owned resources independently while preserving primary and cleanup errors. Reject unknown process-inspection failures instead of treating them as child absence.

Callback, close and process-lookup regressions failed first and now pass their controls. Six focused checks include two intentional expected failures; full build/units and six existing direct native controls pass. Full CR-8 and protected delivery remain open.
@vercel

vercel Bot commented Nov 12, 2026

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
modex Ready Ready Preview Nov 12, 2026 6:51pm UTC
modex-app Ready Ready Preview Nov 12, 2026 6:51pm UTC

Request Review

@github-code-quality

Copy link
Copy Markdown

Code coverage overview

Languages: TypeScript

TypeScript / code-coverage-agent

main d6bda9b cr8-native-boundary 085b4d5 +/-
66% 66% 0%

ℹ️ Calculations represent line coverage. Results may change as calculations are updated.

@BunsDev
BunsDev added this pull request to the merge queue Nov 12, 2026
Merged via the queue into main with commit fd80cd4 Nov 12, 2026
13 checks passed
@BunsDev
BunsDev deleted the cr8-native-boundary branch November 12, 2026 20:40

This branch was successfully deployed

1 active deployment
Preview – modex — 085b4d50 Deployed Nov 12, 2026 by vercel[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant