fix(desktop): localize expected errors - #4457

Reviewed the complete 50-file merge-base diff at 5625712. The earlier Plan false-success and missing-pending defects are fixed. I found no reproduced P0–P2 production defect; one nonblocking P3 regression-test gap is detailed inline.

This changes expected-error transport and presentation, not just copy: Plan preserves structured Host failures over IPC; Work Board, attachments and collaboration map stable codes to locale catalogs; unexpected errors use redacted diagnostics; WorkHub copy is separated; WebSearch settlement retains a nonempty error message. I followed producer/transport/presenter paths and verified all declared Plan error codes through the registered handlers, including an actual missing-proposal storage failure. The deferred React hook test also confirms pending stays true until the bridge settles.

A clean lockfile install, repository dependency patches, full build:test and 158 focused compiled tests passed. The independent producer/IPC/pending probes passed separately. Hosted test and package are successful on this head. No real Electron UI, native-platform acceptance or full-repository test run is claimed.

This head is not merge-ready: GitHub reports CONFLICTING/DIRTY, and merge-tree against current main 2310035 independently reports conflicts in renderer-architecture.json and workhub-surface.tsx. Resolve those and validate the resulting new head; the current review cannot certify that future merge result.

Automated review notice: This comment was posted by an automated review agent operated by hqhq1025. It is not an independent human review and does not replace one.

Re-reviewed the complete rebased change at ae0e0cf. The previous P3 regression-test gap is fixed; no new substantiated P0–P3 finding.

The new test exercises all five registered Plan IPC handlers with typed Host failures, checks the preserved error code and asserts that no mode change is emitted. Disabling the actual production conversion now makes that test fail, unlike the previous head. The unexpected-error rethrow and React pending regressions also remain covered. I checked the rebase adaptations for current WorkHub resume outcomes/clarifications, locale catalogs, fallback-title callers and the protocol compatibility acknowledgment.

On this isolated head, clean install plus repository dependency patches, full build:test, 167 focused compiled tests and 101 renderer-architecture tests all passed; the actual architecture check passed too. Real Electron visual behavior, the full Desktop suite and native-platform acceptance were not run.

The previous main-conflict caveat is now resolved: GitHub reports MERGEABLE/CLEAN, hosted test/package are successful, and a local merge-tree against current main 2310035 is clean. This COMMENT records source and test results; it is not a human review or merge authorization.

Automated review notice: This comment was posted by an automated review agent operated by hqhq1025. It is not an independent human review and does not replace one.

Thanks for the localization work. Additional independent review of ae0e0cff46611dcdf26472b95f0455c0f8d4f108 found one remaining attachment acceptance gap in the normal new-task send path. The IPC envelope behavior was independently exercised in Electron; the finding concerns a renderer preflight that runs before that boundary.

AI disclosure: Codex agents performed this review, and the coordinating agent checked the production call path and reran the preflight/presenter counterexample. This is not an independent human review.

Thanks for the updates. A fresh, multi-reviewer pass at ae0e0cff46611dcdf26472b95f0455c0f8d4f108 found one additional, optional ownership simplification below. The previously reported new-task attachment preflight issue remains unchanged. This review was AI-assisted and cross-checked by the coordinating AI reviewer; it does not represent completed human visual acceptance.

Thanks. A focused size/complexity follow-up is attached inline. This is AI-assisted source review, cross-checked by the coordinating reviewer; no new full-suite run.

Thanks for the localization work and the production-path fixes. The combined review/simplification audit found three additional P3 opportunities to reduce new contracts; they are inline. The existing future-only Session-control token comment still applies.

One further optional consolidation spans several files: reportUnexpectedError and unexpectedErrorFallback merely forward to the new core diagnostic helpers. Diagnostic callers can use those core helpers directly, removing the forwarding API and its dependency edges; keep ExpectedOperationError, which has actual WorkBoard/WorkHub consumers.

The Plan IPC envelope itself, domain-owned catalogs, attachment preflight validation, and the real pending/IPC regression coverage have current obligations and should remain.

AI-assisted review by Codex; checked against the production producers and consumers.

Thanks for the reduction. Rechecked ba95e0823, separating the rebase from the new fix commit: the future-only Session token mechanism is gone, Web Search retains its existing required-message contract, the client exception is back at its client owner, and attachment aliases/copy are consolidated. The extra fix commit removes more than it adds (+66/-148). No P0–P2 remain from this review; the one optional P3 compatibility-record cleanup is inline. Exact-head CI and Windows release checks passed.

AI-assisted delta review by Codex; this refresh checks the simplification against the earlier reviewed behavior, not a new full UI acceptance run.

Thanks for the revisions. I rechecked ba95e0823 specifically for solution size. The producer code → Electron transport → domain catalog direction is sound. Much of +1820/-632 is movement (WorkHub catalog +296 while the surface loses 256 lines), locale copy, and distinct IPC/pending/new-task regressions. I do not recommend a rewrite or a split solely to reduce this diff.

The inline P2 needs a small type import fix. An optional P3 simplification is to stop threading localized fallback text through policy.resolve and workHubNewSessionName: keep the surface→submit input, let policy return the extracted title, and apply decision.title || input.newSessionFallbackTitle once in the controller before workHub.act persists the proposal. Applying it only after returning to the surface would be too late. This removes presentation policy from routing without changing stored titles. The existing compatibility-record P3 still applies; the retired session-control tokens and other previously removed mechanisms should stay removed.

Please retain the serializable Plan envelope, attachment boundary checks, pending cleanup and real send regressions; they protect different production obligations.

AI-assisted review using OpenAI Codex/Astra; evidence checked by the coordinating agent. This is not an independent human review.

Thanks for rebasing. At 544695401, the previous P2 is fixed: PlanControlIpcResult now has a real local type import in the bridge declaration. The producer-code → IPC-envelope → domain-presenter boundaries remain coherent; the rebase does not justify replacing them with a global mapper or removing the distinct pending/preflight regressions. No new P0–P2 finding in this pass.

Ten focused tests passed through the exact-head source loader, covering expected-error presentation, Plan pending behavior, and a nine-attachment new-task rejection before Session creation. Current-head test and Windows-package checks are green.

There is now a merge conflict in apps/desktop/renderer-architecture.json against main 8b3655a17. Please refresh the architecture ledger from the resolved source tree and rerun the existing ratchet. The synthetic merge's protocol declaration is correctly pinned to epoch 133. Actual Electron Plan success/failure/pending and localized attachment-error presentation remain outside these source probes; I am not renewing UI sign-off here.

AI-assisted independent review and primary source/integration/probe verification.

Thanks for the update. Reviewed 80dc732 with an independent incremental reviewer and primary-agent reconciliation: no remaining P0–P2 findings. Error classification reuses the existing core classifier; diagnostics deduplicate the same error object. The proposal-exit channel retains its existing throwing contract: the previous preload adapter already discarded the envelope into an ordinary Error, and its only caller used the generic setting presenter. Removing that unused conversion does not regress an existing code-aware consumer.

Current-head test/package checks pass; the protocol guard passes on the synthetic merge with current main. Earlier focused source tests and the current CI provide regression evidence; no new local workspace suite or live Electron acceptance is claimed. This change does not redesign interaction, so the earlier broad manual Plan/attachment acceptance request is withdrawn. AI-assisted review.