Skip to content

fix(macos): keep unsupported memory controls visible and localize memory forms - #344

Merged
bloodf merged 23 commits into
mainfrom
roadmap/p3/ship101-macos-memory
Nov 7, 2026
Merged

bloodf merged 23 commits into
mainfrom
roadmap/p3/ship101-macos-memory

Conversation

@bloodf

@bloodf bloodf commented Oct 13, 2026 •

Copy link
Copy Markdown
Owner

Unsupported AppKit memory actions remain visible and disabled until negotiated capabilities permit them. Memory forms localize labels, consent warnings, validation errors and target descriptions while preserving wire calls, masking, confirmation and accessibility.

Integrates main through 1b58fa7, preserving the merged desktop direct-label functions and mobile forms exactly. The superseded label-map scanner and its tests are removed; the native form scanner remains.

Validation: 7 remaining scanner tests pass; localization audit, docs and diff checks pass. Prior AppKit memory fixtures pass and remain applicable to unchanged memory sources. One independent source review and parent pinpoint P2 closure are retained; the old label-map mechanism has been replaced by reviewed main. Hosted CI is queued, not passed. Physical/native E2E acceptance NOT_RUN, deferred.

…ing them

The bot memory sheet hid Reflect and the provider-controls button when the capability was not negotiated. Keep both visible and disabled; the regression asserts they stay visible. Refs #101.
@bloodf bloodf mentioned this pull request Oct 14, 2026
@bloodf bloodf added the phase:P3 Roadmap phase P3; not a blanket dependency label Oct 14, 2026
@bloodf

bloodf commented Oct 14, 2026

Copy link
Copy Markdown
Owner Author

findings 2a54889 part 1/1

Scope: independent review of PR #344 at exact head 2a54889 against cached origin/main f84e02a in the local checkout, branch roadmap/p3/ship101-macos-memory. All nine changed files read. No build, no Cargo, no gh/API, no source edits, no publication. The repository localization scanner's non-green state is accepted as a verification limit: the author's run reported no missing translation and no placeholder mismatch, and the scanner only sees literal L("…") calls, so every reachability judgment below was made by reading the production callers. Keys reached through a helper, an enum raw value or an interpolation are exactly what that scanner cannot see, and the two must findings are both of that kind.

must

  1. The Advanced sheet's negotiated-feature menu is populated from raw enum values and never localized, so the one form whose every other label this PR translated still shows English feature names in 简体中文. MemoryAdvancedViewController.swift:42 does featureMenu.addItems(withTitles: supported.map { $0.rawValue.replacingOccurrences(of: "_", with: " ") }) with no L, while the sibling action menu at :79 does wrap its items in L(...). supported is Self.features(backend:capabilities:), non-empty whenever the provider-controls button the PR made visible is enabled, so the menu is reachable in production and its items are the same English words the PR translated one screen earlier as action titles (for example get, update, clear derived, add message, delete record all have table entries added here, but the feature menu never reads them). This is the production-reachability case the literal scanner cannot flag: the source contains no L( call for these strings at all. Fix: pass the feature titles through L the way the action menu already does.

  2. The account memory-connection sheet and the entry button that opens the bot memory sheet remain English in the Chinese UI, because their labels have no table entry even though they run through L. MemoryConnectionViewController.swift:183 builds its OpenViking binding rows from the literals "Bot", "Authorization mode", "Account identity", "User identity", "Bot authorization secret" and passes each through L(label) at :184 and again for the accessibility label at :185; none of the four non-Bot keys exists in zh-Hans.lproj/Localizable.strings at this head, so both the visible row label and its accessibility label read as written. MemoryViewController.swift:63 constructs MemoryActionButton("Memory Service…"), whose initializer runs L(title) (MemoryFormViewController.swift:8), and "Memory Service…" has no entry either, so the button that opens the per-bot memory sheet is English. Same class of defect in the error surface: MemoryFormViewController.swift:124 falls back to L("Memory request failed") and :125/:126 pass rawValue.replacingOccurrences(of: "_", with: " ") to L, producing keys such as memory unavailable, unsupported memory operation, invalid memory input, memory target changed, confirmation required and invalid memory reply — none is in the table, so every failed native memory action reports an English error under the app's 简体中文 pick. These are string literals in production loadView paths, reachable from the account connection sheet, the bot memory sheet and the local editor, and the dynamic-key reachability argument does not apply: L is called, the key is simply absent. Fix: add the missing keys to the table alongside this PR's 195 entries.

Scope note for the lead: findings 1 and 2 touch lines this PR did not itself modify, so they are pre-existing gaps, not regressions. They are in scope for this slice because the PR's own new sentence at docs/architecture/memory-ui.md:62 claims the macOS memory forms resolve their shared helper labels and action titles through the Simplified Chinese table, and because finding 1 sits in a file this PR does edit (MemoryAdvancedViewController.swift) next to the line it localizes. If the lead reads the slice as strictly "translate the literals this PR touched", both can be deferred to a follow-up issue; on the PR's own documented claim they are the delta's actual gap.

should

  • MemoryEmbeddingViewController.swift:78 and BotMemoryServiceViewController.swift:260 interpolate bare English values into already-translated sentences: the vector-space confirmation renders profile.endpoint ?? "local CPU", and the deletion confirmation renders document ?? "whole bot bank". Both are argument values of an L(...) sentence whose Chinese side has no entry for the value, so under 简体中文 the user sees a translated sentence with an English fragment. MemoryConnectionViewController.swift:78 likewise passes "Hindsight" as a section label to optionRow, which does localize it but has no entry. Brand names such as Hindsight and OpenViking are reasonable as-is; local CPU and whole bot bank are ordinary English UI words and are not.

Checked, no finding

  • Head, branch, remotes and origin/main verified locally; origin is bloodf/beans. Review ran against the exact published SHA.
  • Capability absent means visible disabled, not hidden. BotMemoryServiceViewController.swift:129 and :132 set only isEnabled; both reflectButton and advancedButton are unconditionally added to the form at :66 and :83, so neither can disappear. advancedButton.isEnabled is still derived from MemoryAdvancedViewController.features(...) being non-empty, which requires both the feature and its exact verb in advanced_actions (MemoryAdvancedViewController.swift:23-35), so the tightened sentence in memory-ui.md:40 matches the code: advanced actions stay hidden by negotiation inside the advanced sheet while the entry control stays visible. MemoryUIFlows.swift:127 was updated to the same contract and matches.
  • No automatic remote Health, retain or reflect. viewDidAppear (BotMemoryServiceViewController.swift:91) calls only reloadPreferences(), which issues listConnections and getPreferences and sets health = nil at :114; only checkHealth() (:187-196) calls service.health, and only from the explicit button. retain() and queryService(reflect:) stay guarded by the negotiated capability and authorityRecovered, and the deletion path still fences before dispatch. This PR adds no transport call.
  • Reachability of the 195 new table keys. A line-anchored literal scan of macos/Sources/**/*.swift finds every new key present in production sources except "clear derived", "add message" and "delete record". Those three are real negotiated verbs reached through MemoryAdvancedViewController.knownActions (:61, :64, :66) and rendered by L(action.replacingOccurrences(of: "_", with: " ")) at :79 whenever supported is non-empty, so they are genuinely reachable dynamic keys, not dead entries. No new key is dead code.
  • Placeholder accuracy. All 195 new keys were compared position-insensitively against their values: zero mismatches, and no key equal to its own translation. The four new format keys (Bot: %@ · assigned Runner: %@…, Bot: %@. Setup runs…, Profile: %@. Model: %@, revision: %@…, Bot: %@. %@.) keep the same value slots with the Chinese side preserving order, satisfying the language doc's positional rule. The 37 whole-file "mismatches" reported earlier are pre-existing positional reorders (%@ vs %1$@), which is the intended form, not a defect.
  • Fixture assertions are meaningful, not tautological. MemoryUIControllerChecks.swift:145-150 asserts on translated titles produced by the production L path against a real zh-Hans.lproj bundle, on !isHidden && !isEnabled for both capability-gated controls, on makeFirstResponder plus a non-nil field editor for the query field, and on accessibilityTitle() == title for the translated capture checkbox. Every expected Chinese string was read back out of the table at this head. Each assertion fails if the table entries or the disabled-but-visible change regress.
  • The fixture L shim (MemoryUIControllerChecks.swift:4-7) mirrors the production contract in App/Localization.swift:53-67 (bundle lookup, value: fallback to the key, String(format:) only when arguments are supplied) and is gated on BEANS_MEMORY_UI_RESOURCES, which run-memory-ui.sh now sets. With the variable unset the whole block is skipped and the pre-existing baseline is unchanged, so the fixture cannot silently pass by translating nothing.
  • docs/architecture/memory-ui.md is updated in the same change: :40 records the visible-disabled contract for the macOS bot sheet, and the new :62 paragraph names the fixture, the optional BEANS_MEMORY_UI_IMAGE rendering, and the explicit limit that it establishes no installed-app navigation, complete keyboard traversal or assistive-technology acceptance. That limit matches the checkpoint's statement that issue [P3][Memory-Retrieval] Complete macOS memory controls and acceptance #101 stays open.
  • No duplicate keys anywhere in the table; the whole file parses with 1,239 pairs and no repeats.
  • The checkbox AXUnknown observation is handled correctly: the fixture asserts the translated accessibility title only, and no AX role or VoiceOver claim appears anywhere in the diff or the checkpoint.
  • Same-session binary hashes and the codesign verification in the author checkpoint are recorded but not re-run here; no hosted check-run URLs are claimed by this review.

Verdict: changes. Two must findings (unlocalized negotiated-feature menu, and untranslated connection-sheet / entry-button / error keys) block merge. The behavior changes in this PR are otherwise correct and the fixture additions are sound.

@bloodf

bloodf commented Oct 14, 2026

Copy link
Copy Markdown
Owner Author

review-receipt sha=2a5488958920f62e3cebc5635e08175d8b19f754 role=review round=1 model=MiniMax-M3.1-Flash-Preview session=01a13b22-66c3-736d-8097-e304a6e2efc3 verdict=changes findings=#344 (comment) findings_sha256=bd27df2c9369a1973d003f28f05c845d1aa635ab031c533acf002401cbdf0deb

@bloodf

bloodf commented Oct 14, 2026

Copy link
Copy Markdown
Owner Author

findings 423dd35 part 1/1

Scope: independent round 2 review of PR #344 at exact published head 423dd35 in the local checkout, branch roadmap/p3/ship101-macos-memory, composed on top of main 755ecf1. Verified local HEAD equals the published SHA and origin is bloodf/beans. All ten changed files against main read, plus the production callers each change reaches. No build, no Cargo, no gh/API, no source edits, no publication. The repository localization scanner's non-green state remains an accepted verification limit and was not rerun, edited or suppressed; every reachability judgment below was made by reading sources.

Prior round 1 findings are re-verified here, not assumed from the checkpoint's summary. The full-bundle build and codesign proof in the author checkpoint is explicitly bound to the earlier head and is deliberately not carried forward: this review establishes no build evidence for 423dd35, only the scoped native fixture, plutil -lint and bun run check:docs runs the author recorded for the current head. No installed-app, complete keyboard traversal, checkbox AX role or VoiceOver claim is made anywhere below.

Verification of round 1 must 1 — feature menu localization: fixed

MemoryAdvancedViewController.swift:42 now reads supported.map { L($0.rawValue.replacingOccurrences(of: "_", with: " ")) }, matching the sibling action menu at :79. Presentation is localized; selection is still indexed against supported by feature at :49, and the wire value is actionMenu.selectedItem?.representedObject, so the negotiated feature and verb are unchanged. Verified the full set of raw values features() can return (MemoryServiceAction cases bank_profile, bank_config, directives, mental_models, mental_model_history, observations, memory_edit, memory_invalidate, memory_restore, documents, tasks, sessions, resources) and confirmed all 13 display forms plus the 19 knownActions verbs have table entries. The eight basic actions (retain, recall, inspect, delete_document, operation_status, cancel_operation, reflect, clear) are unreachable through this menu: features() at :23-35 only returns members of the per-backend features arrays, none of which is in fileprivate static let basic. Leaving them untranslated is correct, not a gap.

Verification of round 1 must 2 — connection, entry and error surfaces: fixed

MemoryConnectionViewController.swift:183-185 labels now have entries (Authorization mode, Account identity, User identity, Bot authorization secret); the fixture asserts three of them through the real accessibility-label path after driving backendMenu to the OpenViking backend at index 1, which matches backends[1] == .openViking at :20. MemoryViewController.swift:63 and the MemoryFormViewController.swift:124 fallback key Memory request failed both have entries. All 23 error raw values reachable through showError across MemoryUIError, MemoryServiceDraftError and MemorySetupError were enumerated from their declarations and every space-separated form has a table entry. Zero duplicate keys; zero placeholder mismatches across the whole table.

Verification of round 1 should — confirmation fragments: fixed

MemoryEmbeddingViewController.swift:78 and BotMemoryServiceViewController.swift:260 now pass L("local CPU") and L("whole bot bank") as argument values, so the translated sentences no longer carry English fragments.

must

None.

should

None that block merge. Two notes for the record, both verified as correct and needing no change:

  • MemoryUIError.errorDescription returns rawValue unchanged for lanceCloudTransportUnavailable while every other case gets _ replaced with a space. The new table carries the raw lancedb_cloud_transport_unavailable key precisely for that path, and the three pre-existing fixture assertions at MemoryUIControllerChecks.swift:63, :85 and :96 test the untranslated code path, which runs before the language bundle is installed in the fixture. Consistent, not contradictory.
  • The new table keys Action, Priority and Reason are generic single words whose English text is also the key. Action was already translated in round 1 for the row("Action", actionMenu) label; Priority and Reason are Field labels rendered at MemoryAdvancedViewController.swift:134 and :141 and reach L through the field-definition path. No collision with an unrelated Action in the rest of the app was found, and no key was reused with a different meaning.

Checked, no finding

  • Composition is conflict-free and adds no resolution edits. The merge 7d7d56f3 has parents 2a548895 and 755ecf12; main's six incoming files (three workflows, ARCHITECTURE.md, docs/architecture/releases.md, scripts/release-pr-gates.test.ts) are byte-identical between 755ecf12 and the head, and none of them appears in the head's diff against main. merge-base of head and main is exactly 755ecf12, so nothing was replayed onto a moved base.
  • Round 1 behavior changes survived the merge intact: reflectButton and advancedButton still set only isEnabled with no isHidden assignment anywhere in BotMemoryServiceViewController.swift, and the four formatted L notes in the bot, setup, local-assets and advanced controllers are unchanged.
  • No automatic remote Health, retain or reflect. viewDidAppear calls only reloadPreferences(), which issues listConnections and getPreferences and clears health; only checkHealth() calls service.health, from the explicit button. The new fixture instantiates MemoryAdvancedViewController and MemoryConnectionViewController but never opens the connection's advanced sheet, dispatches no memory.service.* request, and its command() call only builds a request object. No new transport path was introduced.
  • MemoryServiceAction presentation is localized without touching the wire contract: select(feature:action:) at :70-73 matches against supported.firstIndex(of:) and representedObject, both of which still carry the raw enum and verb. The fixture asserts this directly at MemoryUIControllerChecks.swift:155-156 by selecting a translated item and confirming command().action == "list".
  • New fixture assertions are meaningful and would fail on regression. :154 compares the full ordered itemTitles array against ["会话", "资源", "任务"], which is order-sensitive against supported and would fail if any title fell back to English. :161 collects real accessibility labels off the constructed view hierarchy rather than asserting on source text. :163, :165 and :167 call showError with one case from each of the three error enums and compare the rendered errorLabel exactly. :169 constructs the real MemoryActionButton and checks its title. None of these can pass by re-asserting the table against itself.
  • The fixture's language bundle is still gated on BEANS_MEMORY_UI_RESOURCES, set by run-memory-ui.sh; unset, the entire block is skipped and the original baseline runs unchanged. The new work is inside that same block, so it cannot silently pass by translating nothing.
  • Placeholder and duplicate integrity across the full table: 0 duplicates, 0 mismatches, verified by parsing every pair. The 44 new keys contain no format specifiers except none at all, so no positional risk was introduced.
  • docs/architecture/memory-ui.md is updated in the same change: the new paragraph records that negotiated feature menus localize presentation without changing the selected wire feature or verb, that the fixture checks OpenViking binding labels, the memory-service entry title and local error presentation, and that the local-CPU and whole-bank fragments use the same lookup. Every clause matches the code and the fixture at this head.
  • MemoryUIFlows.swift is unchanged in round 2; its round 1 assertion at :127 still matches the visible-disabled contract.
  • The author's round 2 verification record is scoped honestly: native fixture, plutil -lint and check:docs for this head, with the full build explicitly excluded and the non-green scanner explicitly not claimed green.

Verdict: pass. Both round 1 must findings and the bounded should are fixed in production code and covered by new fixture assertions that exercise the real localization, capability and error paths. No must or should finding remains open at this head. What this review does not establish is carried forward unchanged: no build for this head, no installed-app navigation, no complete keyboard traversal, no checkbox AX role, no VoiceOver, and no live backend acceptance.

@bloodf

bloodf commented Oct 14, 2026

Copy link
Copy Markdown
Owner Author

review-receipt sha=423dd35214af0d3215b55fd923ce3773d326f6d6 role=review round=2 model=MiniMax-M3.1-Flash-Preview session=01a13b22-66c3-736d-8097-e304a6e2efc3 verdict=pass findings=#344 (comment) findings_sha256=405368f3f019e6e1239d7fb7f8ce915e80d79a7d0ff9d70a290be57aadde22c6

@bloodf

bloodf commented Oct 14, 2026

Copy link
Copy Markdown
Owner Author

findings 423dd35 integration round 1 part 1/1

Binding and scope

  • Exact local HEAD: 423dd35; branch: roadmap/p3/ship101-macos-memory; origin: bloodf/beans. Initial tracked checkout is clean.
  • Actual integration session: 01a13b5a-1475-7782-9fb6-6fa63c4d9afc; model: gpt-6.1-sol. Runtime session header and model-change record establish provenance. Session differs from author 01a13b05-9cf0-754f-96a9-69cafa50c892 and round-2 reviewer 01a13b22-66c3-736d-8097-e304a6e2efc3.
  • Frozen round-2 findings remain untouched: fleet/reviews/344-423dd35214af0d3215b55fd923ce3773d326f6d6.md. Recomputed SHA-256: 405368f3f019e6e1239d7fb7f8ce915e80d79a7d0ff9d70a290be57aadde22c6.
  • Read full author checkpoint, both prior review artifacts, ARCHITECTURE.md, memory-ui, languages and macos-app subjects; all ten current changed files and relevant production localization, binding, entry and error callers.
  • Local-only review. No API, fetch, push, ready action, merge, source edits, test edits, Cargo, full app build, hosted Tests workflow or publication. Supervisor owns publication and task recording.

Source/composition gate: APPROVE, local-source scoped

Local origin/main resolves to 755ecf1. Its merge-base with exact head is that same SHA. No network refresh was authorized: this proves composition with the locally available main, not freshness of hosted main.

Merge 7d7d56f has parents 2a54889 and 755ecf1. A fresh local git merge-tree of those parents produces tree 283788e2d7c1a9f10e889676685ace45652c1073, exactly the committed merge tree. No conflict-resolution edits exist in that composition. Main's incoming workflows, ARCHITECTURE.md, releases subject and release-pr-gates fixture have no diff against exact head. This integration session creates no commit.

Prior must 1: fixed

MemoryAdvancedViewController.swift:42 localizes negotiated feature presentation through L. The feature still comes from supported[indexOfSelectedItem]; action selection retains raw representedObject. All 13 feature display keys and known verb labels have Chinese entries. Native fixture exercises ordered Chinese OpenViking feature titles and verifies command().action remains list. Capability negotiation and transport parameters remain unchanged.

Prior must 2: fixed

MemoryConnectionViewController.swift:183-185 routes OpenViking binding labels and accessibility labels through L; all four previously missing label keys now exist. MemoryViewController.swift:63 uses the shared translated MemoryActionButton entry. MemoryFormViewController.swift:122-128 renders known local UI/draft/setup errors through translated keys without displaying arbitrary backend payloads. Error enum declarations and table entries agree, including the intentionally raw lancedb_cloud_transport_unavailable key. Native fixture exercises binding labels, service entry and one rendered error from each enum.

Prior should: fixed

MemoryEmbeddingViewController.swift:78 and BotMemoryServiceViewController.swift:260 localize local CPU and whole bot bank confirmation fragments without changing actual endpoints, document IDs or request targets.

Other checked properties

  • Reflect and provider-entry controls remain visible disabled when capability is absent; exact negotiated verbs still gate provider actions.
  • Opening bot sheet loads preferences/configuration, not remote Health, retain or reflect. Changed lines introduce no transport call.
  • Four formatted descriptions preserve interpolation slots; translated warnings retain plaintext, downstream-cost, explicit-approval and erasure limitations.
  • Whole Chinese table audit: 1,283 parsed pairs, zero duplicate keys, zero position/type-aware placeholder mismatches.
  • Fixture resources are explicitly set by runner; Chinese assertions exercise constructed production controllers, keyboard editing focus and translated accessibility titles. Fixture localization shim does not reproduce production locale selection; no locale-sensitive formatting claim is made.
  • memory-ui.md describes the source mechanisms and expressly limits fixture proof; both changed fixture files agree with the visible-disabled contract.

Local execution evidence gate: CLOSED for scoped runs only

This session compiled and executed the fixture binaries from this checkout. Successful traced native run is captured in artifact://12:13-35 and this session's background-result record:

  • Native AppKit fixture completes all five PASS summaries, including negotiated Chinese menus/wire actions, binding labels, entry/error surfaces, disabled discoverability, keyboard editing focus, accessibility titles, guarded drafts, failure recovery and light/dark scroll sizing.
  • MemoryServiceDraft reports consent, secret, capability, budget and fixture RPC checks passed.
  • MemorySetupDraft reports exact routes, scope, confirmation, expiry and failed-apply single-use checks passed.
  • MemoryUIPreviewExpiry reports real core fractional expiry and exact expiry boundary PASS.
  • Traced runner reaches its cleanup after all four executables, with no failure output. Runner uses set -eu. No numeric assertion count is emitted, so none is claimed.
  • Separate bun run check:docs completes successfully; overview and every subject fit budget, including memory-ui.md at 14.1 KiB of 24.0 KiB.
  • Separate plutil -lint macos/Resources/zh-Hans.lproj/Localizable.strings reports OK.

Initial synchronous native invocations returned only an incomplete tool response, Compiled 3 packages, without usable completion evidence. Those responses are not acceptance proof; the complete traced background run above is the evidence used here. Tool inconsistency was reported locally.

Hosted/publication gate: UNKNOWN / UNBOUND

No hosted check status or check-run URL was queried. Hosted main freshness, hosted required checks and publication are unknown. Local source approval and fixture success do not establish hosted Tests success or merge readiness. No ready or merge authorization is issued.

Earlier full bundle/codesign evidence belongs only to 2a54889, not this head. No installed-app navigation, live backend acceptance, complete keyboard traversal, checkbox accessibility role or VoiceOver proof is established. Earlier non-green whole-localization scanner remains a recorded limit; it was not rerun, suppressed or claimed green. Issue #101 installed-app/live-backend acceptance remains outside this source-scoped integration proof.

Findings and disposition

must: none in reviewed local source/composition.
should: none requiring correction in this bounded slice.

Source/composition: APPROVE against local main 755ecf1.
Local native/docs/plist evidence: CLOSED for the exercised scope.
Hosted checks/main freshness: UNKNOWN / UNBOUND.
Overall merge readiness: NOT ESTABLISHED. Stop after private artifact; supervisor publishes separately.

Changed files reviewed

  • docs/architecture/memory-ui.md
  • macos/Resources/zh-Hans.lproj/Localizable.strings
  • macos/Sources/Beans/Settings/MemoryBotSetupViewController.swift
  • macos/Sources/Beans/Settings/MemoryEmbeddingViewController.swift
  • macos/Sources/Beans/Settings/MemoryLocalAssetsViewController.swift
  • macos/Sources/Beans/Sheets/BotMemoryServiceViewController.swift
  • macos/Sources/Beans/Sheets/MemoryAdvancedViewController.swift
  • macos/Tests/MemoryUIControllerChecks.swift
  • macos/Tests/MemoryUIFlows.swift
  • macos/Tests/run-memory-ui.sh

@bloodf

bloodf commented Oct 14, 2026

Copy link
Copy Markdown
Owner Author

review-receipt sha=423dd35214af0d3215b55fd923ce3773d326f6d6 role=integration round=1 model=gpt-6.1-sol session=01a13b5a-1475-7782-9fb6-6fa63c4d9afc verdict=pass findings=#344 (comment) findings_sha256=137d392ad13a0f3e9e3d1ed103c22cede6ee3accecdcd5842145e2f5d8ea1783
Source/composition and exercised local checks only. Hosted checks UNKNOWN/UNBOUND; merge readiness NOT ESTABLISHED. No current-head full build, installed-app, AX-role or VoiceOver proof.

@bloodf bloodf added type:implementation Roadmap implementation area:memory Beans memory workstream priority:normal Urgency: normal; schedule after high-priority work. Independent of roadmap phase. labels Oct 19, 2026 — with ChatGPT Codex Connector
@bloodf bloodf mentioned this pull request Oct 19, 2026
57 tasks
@bloodf bloodf added the platform:macos Target scope: macOS. Does not imply implementation or acceptance; require target-specific evidence. label Oct 19, 2026
@bloodf bloodf changed the title fix(macos): keep unsupported memory controls visible fix(macos): keep unsupported memory controls visible and localize memory forms Oct 19, 2026
@vercel

vercel Bot commented Nov 4, 2026 •

Copy link
Copy Markdown

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

Project Deployment Actions Updated
beans Ready Ready Preview Nov 7, 2026 9:42pm UTC

Request Review

@bloodf
bloodf marked this pull request as ready for review November 7, 2026 18:43
@bloodf
bloodf merged commit a8a631e into main Nov 7, 2026
13 checks passed

This branch was successfully deployed

1 active deployment
Preview — 3ba103a7 Deployed Nov 7, 2026 by vercel[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

area:memory Beans memory workstream phase:P3 Roadmap phase P3; not a blanket dependency platform:macos Target scope: macOS. Does not imply implementation or acceptance; require target-specific evidence. priority:normal Urgency: normal; schedule after high-priority work. Independent of roadmap phase. type:implementation Roadmap implementation

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant