diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml index b1dae37..c6966d3 100644 --- a/.github/workflows/ci.yml +++ b/.github/workflows/ci.yml @@ -28,6 +28,12 @@ jobs: - name: Install Frontend Dependencies run: npm ci + # V-96: `npm run build` is a Vite transpile, NOT a type check. A type + # error could reach main even though the local gate `npx tsc --noEmit` + # would have caught it. CI now enforces the same contract. + - name: Verify TypeScript (no emit) + run: npx tsc --noEmit + - name: Verify TypeScript & Frontend Build run: npm run build diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml index efb439e..e06dc23 100644 --- a/.github/workflows/release.yml +++ b/.github/workflows/release.yml @@ -328,6 +328,10 @@ jobs: printf '%s %s\n' "$hash" "$file" >> SHA256SUMS.txt done <<< "$files" cat SHA256SUMS.txt + # V-96: verify the manifest BEFORE uploading. A truncated or + # corrupt SHA256SUMS.txt must fail the release instead of shipping + # to users who run `sha256sum -c` after download. + sha256sum -c SHA256SUMS.txt - name: Create or Update GitHub Release & Upload Assets env: @@ -370,18 +374,12 @@ jobs: [ -f "$file" ] || continue filename=$(basename "$file") echo "Uploading $filename..." - asset_id=$(gh api "repos/${{ github.repository }}/releases/tags/$TAG/assets" \ - --jq ".[] | select(.name==\"$filename\") | .id" 2>/dev/null || true) - if [ -n "$asset_id" ]; then - gh api -X DELETE "repos/${{ github.repository }}/releases/assets/$asset_id" >/dev/null - sleep 1 - fi # No `|| true`: a failed upload must fail the release. gh release upload "$TAG" "$file" --clobber done echo "Verifying the published asset contract..." - names=$(gh api "repos/${{ github.repository }}/releases/tags/$TAG/assets" --jq '.[].name') + names=$(gh release view "$TAG" --json assets --jq '.assets[].name') for required in \ "VaultZero-Windows-x64-Setup.exe" \ "VaultZero-Windows-x64-Portable.exe" \ diff --git a/checklistprogress.md b/checklistprogress.md index 25bb74d..858d7d6 100644 --- a/checklistprogress.md +++ b/checklistprogress.md @@ -207,3 +207,167 @@ Active branch: `arena/317fbb3c-vaultzero` (never rename). PR #2 + #3 MERGED, #1 - [x] Overwritten GitHub release `v2.5.2` with fresh assets via `gh release upload v2.5.2 ... --clobber`. - [x] `node scripts/check-live-release.mjs` -> ALL 5 links resolve (`exit 0`). - [x] All 5 live download URLs verified returning HTTP 200 with matching Content-Length. + +## PHASE F / V-90 — Step 0 + Batch 1 (2026-11-08) — NEW PR campaign + +- [x] Step 0: `npx skills` v1.7.6; 18/19 installs OK -> 38 skill dirs in gitignored `.agents/skills/`; 2 failures reported honestly +- [x] Applied mobile guidelines: 375px, 8pt, 44x44, thumb zone, mobile reflow, no blur-slop, TDD + verification-before-completion +- [x] Re-baselined: GitHub main == branch at `c6ea9b2` (5 external commits); `git pull --ff-only`; CI 42405097576 success +- [x] Re-observed live release: 10 assets, canonical 5/5 present -> `check-live-release.mjs` EXIT 0 (STALE notice remains) +- [x] Batch 1 RED: B-01/B-02/B-03 guards PASS, B-04 + B-05 FAIL (stale native flag keeps Unlock screen) +- [x] Corrected a wrong TEST assumption in B-02 (`addInitScript` wiped the container on reload) instead of touching app code +- [x] Batch 1 GREEN fix: native `.pmv` authoritative + stale marker cleared; browser ignores native marker; LockScreen probe sets both directions with manual-toggle guard +- [x] Gates: tsc 0 · verify 0 (38) · node20 38/38 · audit 0 · Playwright **152/152 (9.0m)** · live check 0 +- [ ] Batch 2 (resume with `continue`): Android 36px status-bar clearance, overflow-x lock, All items / touch-scrollable sort pills +- [ ] Batches 3-7: P2P real LAN IP + camera fallback, updater resilience, OCR/TOTP/logo, DOM audit, CI/release + final loop +- [ ] Final: Bengali report + SEPARATE Antigravity prompt (local-only work + merge) + English master prompt LAST +- [x] NEW **PR #4** OPEN (`arena/317fbb3c-vaultzero -> main`) — created after batch 1 because main/branch were identical before +- [x] CI `42413851086` watched to exit 0; `gh pr checks 4` = pass/pass (8m15s + 2m38s); no merge/close +- [x] Batch 1 CLOSED — resume with `continue` for Batch 2 + +## PHASE F / V-91 — Batch 2: status bar, overflow lock, All items + sort pills (2026-11-08) + +- [x] Re-baselined: HEAD `52866f1`, clean, PR #4 pass/pass, no external advance +- [x] Measured 360x800: EN top=10px, overscroll `auto`, h2 `white-space:normal`, sort row `touch-action:auto` +- [x] RED: B-06 padding 0 FAIL · B-07 `auto` FAIL · B-09 `normal` FAIL · B-08/B-10 guards PASS +- [x] GREEN fix 1: `html/body/#root` width/margin/padding + `overscroll-behavior-x:none` +- [x] GREEN fix 2: mobile `.mobile-safe-top` = `max(var(--safe-top), 36px) !important` (vault + lock); EN top 10 -> 46 +- [x] GREEN fix 3: `All items` h2 nowrap/shrink-0/truncate at 70vw; sort row `no-scrollbar touch-pan-x overscroll-x-contain` +- [x] M-03 strengthened (floor >=36 at inset 0/24, exact 64 at inset 64) — no existing guard weakened +- [x] Re-captured 360x800 screenshot; targeted 26/26 (B-06..B-10 + M-01..M-19) +- [x] Gates: tsc 0 · verify 0 (38) · node20 38/38 · audit 0 · Playwright **157/157 (10.1m)** · live check 0 +- [!] Trailing footer note partly behind nav at scrollTop=0 (pre-existing, scroll-reachable, M-04); batch 6 will assert its box +- [ ] Batch 3 (resume with `continue`): P2P real LAN IP, strict RFC-1918 IPv4, camera/QR-photo fallback, footer clipping +- [ ] Batches 4-7 + final Antigravity prompt (local-only work + merge) + English master prompt LAST +- [x] `b99fe78` pushed; push `42417004735` + PR `42417008987` both watched to exit 0; `gh pr checks 4` pass/pass +- [x] Batch 2 CLOSED — resume with `continue` for Batch 3 + +## PHASE F / V-92 — Batch 3: P2P real LAN IP, strict peers, camera + footer (2026-11-08) + +- [x] Re-baselined: HEAD `a0fc6ec`, PR #4 `pr_open` pass/pass, no remote advance +- [x] Found UNCOMMITTED V-92 work in the working tree; preserved, reviewed, executed, improved (nothing discarded) +- [x] Verified strict RFC-1918 validator, Rust `lan_ipv4` UDP-route trick, `generate_handler!` registration, `core:default` ACL +- [x] Verified `sync_from_peer` really does a WebSocket key-exchange handshake before "Connected" (no fake connection) +- [x] Verified manifest CAMERA/INTERNET + 44px QR-photo fallback with `capture="environment"` + footer inside 360px +- [x] B-16 RED (false GREEN first: StrictMode double effect → hardened with probe counter + allow flag); true RED 2==2 +- [x] B-16 GREEN fix: reusable `useCallback probeLanIp`, StrictMode-safe aliveRef, Discover re-probes only when no IP +- [x] Harness gained reusable-probe + re-probe + positive-control assertions +- [x] Full-suite regressions resolved honestly: M-19 updated to reachability invariant (NOT weakened), B-16 test race fixed with `expect.poll` +- [x] Gates: tsc 0 · verify 0 (39 harness, 23 rustfmt) · node20 39/39 · audit 0 · Playwright **163/163 (15.7m)** · live check 0 +- [ ] Batch 4 (resume with `continue`): Check for Updates CSP, resilient GitHub fetch, strict SemVer +- [ ] Batches 5-7 + final Antigravity prompt (local-only work + merge) + English master prompt LAST +- [!] `.vztransfer` snapshot-file transfer NOT implemented (reported, not faked); real-device camera/LAN/onPermissionRequest needs Antigravity +- [x] `09d5747` pushed; push `42435123882` + PR `42435126914` both watched to exit 0 (Windows Rust compile/clippy green); `gh pr checks 4` pass/pass +- [x] Batch 3 CLOSED — resume with `continue` for Batch 4 + +## PHASE F / V-93 — Batch 4: resilient update check, CSP, strict SemVer (2026-11-08) + +- [x] Re-baselined: HEAD `8945e62`, PR #4 `pr_open` pass/pass, no remote advance +- [x] Measured CORS for real: api.github.com = `ACAO:*`; github.com 302 has NO CORS header (cannot read Location from renderer) +- [x] Reproduced the screenshot's false "Offline" in real Chromium (`ERR_CERT_AUTHORITY_INVALID` -> offline) +- [x] RED: 11 harness FAIL (CSP tokens, 8s timeout, API version header, no fallback, false offline) +- [x] GREEN fix 1: connect-src += `*.githubusercontent.com`, `ipc:`, `http://ipc.localhost` +- [x] GREEN fix 2: 10s timeout, pinned API version, `/releases?per_page=1` fallback, no-cors reachability probe, `error` != `offline` +- [x] GREEN fix 3: new bilingual `error` render in UpdateModal (previously invisible); up_to_date/Update Now unchanged (already strict SemVer) +- [x] Review after fix: all 3 endpoints attempted in order; sandbox cert-fails all -> offline is honest; B-17/B-18/B-20 prove online behaviour +- [x] Gates: tsc 0 · verify 0 (**40 harness**) · node20 40/40 · audit 0 · Playwright **167/167 (13.0m)** · live check 0 +- [!] Renderer cannot read the 302 Location (CORS); a native Rust HTTP command to do so is NOT implemented — reported, not faked +- [ ] Batch 5 (resume with `continue`): offline OCR (eng+ben), strict RFC 6238 TOTP, official APK/Windows logo parity +- [ ] Batches 6-7 + final Antigravity prompt (local-only work + merge) + English master prompt LAST +- [x] `0e616c7` pushed; push `42458168857` + PR `42458173011` both watched to exit 0; `gh pr checks 4` pass/pass +- [x] Batch 4 CLOSED — resume with `continue` for Batch 5 + +## PHASE F / V-94 — Batch 5: offline OCR, strict RFC 6238, icon parity (2026-11-08) + +- [x] Re-baselined: HEAD `a3fa6cb`, PR #4 `pr_open` pass/pass, worktree clean, no remote advance +- [x] Inspected exact files: OCR already `worker.recognize()` -> `ret.data.text` (no metadata result), TOTP already 16+ Base32 + RFC 6238 SHA-1/256/512 +- [x] Measured real parity: all 15 Android mipmaps decode to the same emblem (mean-RGB delta <= 9 vs 512px master); Windows `icon.ico`/NSIS referenced +- [x] RED: `?secret=abc` accepted -> bogus 2FA key (5 harness FAIL + B-23 browser FAIL); `#FFFFFF` adaptive background (2 FAIL); "vision model" caption (1 FAIL) +- [x] B-21/B-22 GREEN at RED time: real `eng` + dual `eng+ben` OCR text, no filename/size/dimensions, monotonic progress 10 -> 100 +- [x] GREEN fix 1: `parseOtpAuthUri()` validates the inner Base32 secret and returns `cleanKey`; `isValidTotpSecret` + `resolveTotpConfig` now fail closed +- [x] GREEN fix 2: `ic_launcher_background.xml` `#FFFFFF` -> `#13182D` (master corner 19,24,45) +- [x] GREEN fix 3: OCR caption -> "on-device offline OCR (Tesseract)" +- [x] New contracts: 41st harness `totp-uri-strict`, 42nd harness `icon-parity` (pure-Node zlib PNG decoder + fingerprint), +9 `offline-ocr` guards, B-21/B-22/B-23 browser tests +- [x] Review after fix: guard was reading XML comments -> TEST fixed to parse the actual `` value, assertion NOT weakened; Chromium path corrected to `/tmp/chromium/chrome` +- [x] Gates: tsc 0 · verify 0 (**42 harness**) · node20 42/42 · audit 0 · Playwright **170/170 (10.2m)** · live check 0 +- [!] Real-device launcher masking, installer artwork and Bengali handwriting accuracy need Antigravity verification; no release published, no merge +- [x] `249f116` pushed; push `42503258511` + PR `42503261967` both watched to exit 0 (4/4 jobs success); `gh pr checks 4` pass/pass +- [x] Batch 5 CLOSED — resume with `continue` for Batch 6 +- [ ] Batch 6 (resume with `continue`): full verification + explicit 360x800 / 1440x900 bounding-box DOM audit +- [ ] Batch 7 + final regression loop + Bengali Engineering Report + Antigravity prompt (local work AND merge) + English master prompt LAST + +## PHASE F / V-95 — Batch 6: explicit 360x800 / 1440x900 bounding-box DOM audit (2026-11-08) + +- [x] Re-baselined: HEAD `da78cf8`, PR #4 `pr_open` pass/pass, worktree clean, no remote advance +- [x] Measured real boxes at BOTH viewports: Lock/Vault/Sync `scrollWidth===clientWidth`; header y=36 (phone) / y=33 (desktop) +- [x] Measured all three Sync footer actions after Connect: all fully in bounds at 360x800 and 1440x900 +- [x] Verified `192.1` rejected and a fresh vault created at BOTH viewports (B-27/B-28 GREEN at RED time) +- [x] FOUND DEFECT: shared Modal footer `py-3.5` (14px) ignored `--safe-bottom`; real 24/48px inset left `Receive & Merge` at bottom=785 of 800 +- [x] RED: 3 harness FAIL + B-26 browser FAIL (`Expected > 14, Received 14`); B-24/B-25/B-27/B-28 GREEN +- [x] GREEN fix: `ui.tsx` footer += `modal-footer` class; `index.css` `.modal-footer { padding-bottom: calc(14px + var(--safe-bottom)) !important }` +- [x] No artificial floor deliberately: a floor would steal body height and break M-16 at 390x844 +- [x] Review after fix: inset 0 -> 14px / bottom 785 (unchanged); inset 24 -> 38px / 761; inset 48 -> 62px / 737; body 428/428 and 411/411 unchanged (M-16/M-19 safe) +- [x] New contracts: 43rd harness `modal-safe-area.test.mjs` (positive control + guards), `vault-bounding-audit.spec.ts` B-24..B-28 +- [x] Gates: tsc 0 · verify 0 (**43 harness**) · node20 43/43 · audit 0 · Playwright **175/175 (10.9m)** · live check 0 +- [!] Whether a real WebView reports the bottom inset correctly needs physical-device/Antigravity verification; no unmeasured floor was faked +- [x] `df72f73` pushed; push `42507639824` + PR `42507643399` both watched to exit 0 (4/4 jobs success); `gh pr checks 4` pass/pass +- [x] Batch 6 CLOSED — resume with `continue` for Batch 7 +- [ ] Batch 7 (resume with `continue`): CI/CD + 5 canonical release assets + final Batch 1→7 regression loop +- [ ] Final deliverables: comprehensive Bengali Engineering Report + Antigravity prompt (local work AND merge) + English master prompt LAST + +## PHASE F / V-95b — Batch 6 CI flake RCA + correction (2026-11-08) + +- [!] `f21cf4e` docs commit: PR run `42508535834` green (175/175) BUT push run `42508531423` Frontend job FAILED (174/1) — not ignored +- [x] Signed job log `128816346519` root-caused it: B-28 `toHaveCount(0)` 20s timeout; same commit green in PR run + 4/4 local repeats -> TEST flake, not app regression +- [x] Hardened B-28: one test per viewport (fresh context), container-absent precondition, ALL creation/validation errors rejected, `expect.poll` on the real vault key (60s), dialog unmount (30s) +- [x] 8/8 repeat runs GREEN at 8.5-11.5s each; full audit spec 6/6 +- [x] Re-ran all gates: tsc 0 · verify 0 (43 harness) · node20 43/43 · audit 0 · live 0 · Playwright **176/176 (10.8m)** +- [x] `5bc4b42` pushed; push `42510871635` + PR `42510875667` BOTH watched to exit 0 (4/4 jobs success); `gh pr checks 4` pass/pass +- [x] Batch 6 FULLY CLOSED (including CI flake correction) — resume with `continue` for Batch 7 + +## PHASE F / V-96 — Batch 7: CI/CD + 5 canonical assets + final Batch 1→7 loop (2026-11-08) + +- [x] Re-baselined: HEAD `fb692d7`, PR #4 `pr_open` pass/pass, worktree clean +- [x] Inspected both workflows: CI had NO `npx tsc --noEmit` (only Vite build); release built `SHA256SUMS.txt` but never ran `sha256sum -c` +- [x] Guards confirmed already correct: exact 5 canonical names in README / checker / publish verify, `--latest` no `--prerelease`, `--clobber`, no `|| true`, Rust fmt/clippy/test, Playwright, audit, Node 20 +- [x] RED: 44th harness `ci-release-gates.test.mjs` failed EXACTLY 2 assertions; 4 positive controls prove both detectors work +- [x] GREEN fix 1: `ci.yml` += `Verify TypeScript (no emit)` step (`npx tsc --noEmit`) after `npm ci` +- [x] GREEN fix 2: `release.yml` += `sha256sum -c SHA256SUMS.txt` before upload +- [x] Review: exact shell snippet replayed on 6 dummy assets -> all `OK`, exit 0; YAML indentation structurally verified; no dependency installed +- [x] FINAL LOOP Batch 1→7: tsc 0 · verify 0 (**44 harness**) · node20 44/44 · audit 0 · live 5/5 · batch suites in order **50 passed (3.3m)** · FULL Playwright **176/176 (11.7m)** — zero regressions +- [!] No release/tag/merge: needs 4 signing secrets + keytool + Windows runner + signed-log gate; live `v2.5.2` still STALE; real-device checks remain for Antigravity +- [x] `6a564a6` pushed; push `42546855522` + PR `42546859915` both watched to exit 0 (4/4 jobs success); CI step `Verify TypeScript (no emit) -> success` confirmed; `gh pr checks 4` pass/pass +- [x] Docs RED→GREEN: 5 new `release-runbook` assertions (4 FAIL -> all pass); runbook intro fixed to 10 assets / 5-of-5 resolving / still STALE; section 8 added with local verification + `gh pr merge 4 --merge` +- [x] Re-ran after docs edit: tsc 0 · verify 0 (44 harness) · node20 44/44 +- [x] Batches 1-7 CLOSED · final Batch 1→7 loop zero regressions · PR #4 OPEN pass/pass · NO merge/tag/publish +- [x] Final deliverables delivered in chat: comprehensive Bengali Engineering Report + SEPARATE Bengali Antigravity prompt (local work AND merge) + English master prompt LAST + +## PHASE G / V-97 — Release Keystore Secrets, ManifestMerger2 Duplicate Attribute Fix & Signed Release Pipeline (2026-11-09) + +- [x] Step 1 Keystore: Generated 2048-bit RSA PKCS12 release keystore valid for 10,000 days at `C:\Users\Tonmoy\.vaultzero-keys\vaultzero-release.jks` with alias `vaultzero` and distinguished name `CN=VaultZero, OU=Security, O=VaultZero, L=Dhaka, ST=Dhaka, C=BD` via OpenJDK 17 `keytool`. +- [x] Step 1 Secrets: Configured all 4 production repository secrets on `tonmoycoder/VaultZero` via `gh secret set`: `ANDROID_KEYSTORE_BASE64`, `ANDROID_KEYSTORE_PASSWORD`, `ANDROID_KEY_ALIAS`, `ANDROID_KEY_PASSWORD`. Confirmed active via `gh secret list`. +- [x] Step 3 Local Telemetry: Measured Windows cold start latency on `Desktop\VaultZero.exe`: **146 ms** (WaitForInputIdle: True, PID 2272), well within the < 2000 ms SLA. +- [x] Playwright 176/176 Verified: All 176 tests passed on local (including 3 cold-start suites verified in 5.1s, 7.2s, 9.5s). +- [x] RCA of Android Release Build Failure: In `release.yml` run `42668919907`, `Build Windows Installer & Portable` succeeded in 9m12s (NSIS, Portable, MSI built), but Android build failed with `ManifestMerger2$MergeFailureException: Error parsing ... AndroidManifest.xml`. Root-caused from signed log chunk 1579: Tauri CLI template already provides `android:usesCleartextTraffic="${usesCleartextTraffic}"`. The previous check in `scripts/patch-android-manifest.mjs` only looked for `="true"`, causing a duplicate attribute `android:usesCleartextTraffic="${usesCleartextTraffic}" android:usesCleartextTraffic="true"` which broke the XML parser. +- [x] RED -> GREEN Fix: Updated `scripts/patch-android-manifest.mjs` to replace any existing `android:usesCleartextTraffic="[^"]*"` with `"true"`, or append cleanly if missing. +- [x] Guard Added: `tests/harness/android-runtime.test.mjs` extended to assert that Tauri template `${usesCleartextTraffic}` is replaced with `"true"` and contains exactly 1 attribute with no duplicate XML attributes. +- [x] Gates Verified: `npx tsc --noEmit` exit 0, `npm run verify` exit 0 (44/44 harness pass, contrast AA 100%, 23 rustfmt, bundle budget ok), `cargo test -p vault-core` 34/34 pass, `cargo clippy` 0 warnings. +- [x] RCA of Publish Job 129355385531: Windows installer & portable succeeded (5m47s) and Android signed build succeeded (5m30s) verifying `"Restored the long-lived release signing keystore from GitHub Secrets."` in the signed log. The publish job failed trying to query non-existent `/releases/tags/$TAG/assets` and passing the 404 JSON to `gh api -X DELETE`. Fixed `.github/workflows/release.yml` to rely on native `gh release upload --clobber` and `gh release view "$TAG" --json assets` for asset verification. + +## PHASE G / V-98 — Release Workflow Green Exit, 5 Canonical Assets & Timezone Parity in Live Checker (2026-11-09) + +- [x] Release Run 42679507033 Complete & Green: + - `Build Signed Android APK`: SUCCESS in 5m22s (ID 129369930228). + - Signed log assertion confirmed: `"Restored the long-lived release signing keystore from GitHub Secrets."` (timestamp: 2026-11-09T19:07:51.2348541Z). + - `Build Windows Installer & Portable`: SUCCESS in 8m4s (ID 129369930548). + - `Publish to GitHub Releases`: SUCCESS in 20s (ID 129374565199). +- [x] All 5 Canonical Assets Published & Verified: + - `VaultZero-Windows-x64-Setup.exe` (12,865,560 B) + - `VaultZero-Windows-x64-Portable.exe` (21,520,896 B) + - `VaultZero-Windows-x64.msi` (14,131,200 B) + - `VaultZero-Android.apk` (19,691,780 B) + - `SHA256SUMS.txt` (691 B) +- [x] Live Checker Parity: Fixed date string comparison in `scripts/check-live-release.mjs` to compare ISO 8601 UTC dates (`.toISOString().slice(0, 10)`), resolving UTC/local offset mismatch. `node scripts/check-live-release.mjs` exits with code 0 and ZERO `STALE` notices (`RESULT: all 5 links resolve`). + + + diff --git a/checkpoint.md b/checkpoint.md index ffbd5a7..485bcc5 100644 --- a/checkpoint.md +++ b/checkpoint.md @@ -617,3 +617,633 @@ remaining action is the human/Antigravity publication in - Executed `gh release upload v2.5.2 ... --clobber` with authenticated `GH_TOKEN`. - Verified via `scripts/check-live-release.mjs`: ALL 5 links resolve (`exit 0`). - Verified all 5 live download links returned HTTP 200 with exact matching byte counts. + +## V-90 — NEW PR CAMPAIGN, Step 0 + Batch 1 (2026-11-08) + +**New user instruction.** Put all work in a NEW pull request; work in 7 +micro-batches, one fully verified per turn; anything impossible here goes to a +SEPARATE Antigravity prompt that must be told to do it locally AND merge. +User asked for the best coding/reasoning effort. Branch stays +`arena/317fbb3c-vaultzero` (active); no other branch was created. + +**Step 0 — skills (honest).** `npx -y skills` v1.7.6 works. 19 install commands +attempted, 18 succeeded -> 38 skill directories in gitignored `.agents/skills/` +(`frontend-design`, `webapp-testing`, `skill-creator`, `test-driven-development`, +`systematic-debugging`, `verification-before-completion`, `using-superpowers`, +`brainstorming`, `mobile-app-ui-design`, `mobile-android-design`, +`antislop-layoutmobile`, `design-mobile-apps`, `mobile-native`, +`emil-design-eng`, `momentic-mobile-test`, `ui-ux-pro-max`, etc.). +`git status` stayed clean. TWO failed: `AIBiz-Automatyzacje/workspace-template-mobile` +(a sibling `freshness-audit/SKILL.md` has a YAML parse error, installer skipped +the package) and `leonxlnx/taste-skill --skill imagegen-frontend-mobile` (the +repo exists but that skill is not published). Applied guidelines: 375px +baseline, 8pt spacing, 44x44 tap targets, thumb zone, no overused blur, +"mobile is a different layout, not a smaller desktop", TDD + verification +before completion. Nothing fabricated. + +**Re-baseline — the workspace advanced AGAIN.** `gh pr create` failed with +"No commits between main and arena/317fbb3c-vaultzero": GitHub `main` and this +branch were IDENTICAL at `c6ea9b2`. Five external commits had landed since +`ff5ba5a`: `fcf48b1` (custom-protocol default), `ebe805e` (resilient Entry +deserialization + async native vault detection + unlock toggle), +`9890618` (workspace deep-clean + fresh production assets), `d139472` (release +sync doc), `c6ea9b2` (ignore root release artifacts). Fast-forwarded local +HEAD with `git pull --ff-only`. CI `42405097576` (c6ea9b2) = success. +LIVE RELEASE CHANGED: `v2.5.2` now has 10 assets including canonical +`VaultZero-Windows-x64-Portable.exe` and `VaultZero-Windows-x64.msi`, so +`check-live-release.mjs` is now EXIT 0 (5/5 links resolve). It STILL prints +`STALE` because publishedAt is 2026-11-02 while source is 2026-11-08. +PR state unchanged: #3 MERGED, #2 MERGED, #1 OPEN (never touched). + +**Batch 1 — "Failed to create vault" + first-run flow.** The `ebe805e` external +fix already repaired the reported root cause (native `upsert_entry` sent an +Entry missing Rust-required fields; `crates/vault-core/src/models.rs` gained +`#[serde(default)]`/aliases; `vault_file_path` already does +`create_dir_all(app_data_dir())`). It was NOT redone. RED tests were written +against the exact screenshots instead, exposing ONE remaining real defect: +`checkVaultConfiguredAsync()` only ever called `setIsSetup(false)`, and when +`has_vault_file` returned false it still returned the stale +`vaultzero_has_native_vault` localStorage marker -> a fresh install stayed on +"Welcome back, Vault Owner." after the .pmv file disappeared. In browser mode +the same marker also made a browser launch look configured. + +- NEW `tests/vault-first-run.spec.ts` (B-01..B-05): B-01 reproduces PC + `aA2525@` + EMPTY optional duress + NO keyfile -> vault created, no + "Failed to create vault"; B-02 real vault -> reload -> "Welcome back"; + B-03 stubbed native `has_vault_file=true`, no marker -> "Welcome back"; + B-04 stale marker + missing .pmv -> must fall back to SETUP and clear the + marker; B-05 browser launch must ignore a leftover native-only marker. +- First run: 1 wrong TEST assumption (B-02 used `addInitScript`, which wiped + the container again on reload) -> fixed the TEST, not the app. Clean RED: + 3 guards PASS, B-04/B-05 FAIL. +- Minimal fix: `src/lib/storage.ts` — `NATIVE_VAULT_FLAG` constant, + `isVaultConfigured()` counts the native marker ONLY in native runtime, and + `checkVaultConfiguredAsync()` treats the `.pmv` file as authoritative + (sets the marker when present, REMOVES it when absent; browser mode checks + only `VAULT_STORAGE_KEY`). `src/components/LockScreen.tsx` — the async probe + now sets `isSetup(!configured)` in BOTH directions, guarded by a + `manualModeRef` so an explicit user toggle always wins. +- GREEN: 5/5 first-run tests. + +**Gates (batch 1).** `tsc` 0; `npm run verify` 0 (38 harness, rustfmt, bundle +budget); node20 38/38; `npm audit --audit-level=high` 0; +`npx playwright test` **152 passed (9.0m)**; `check-live-release.mjs` EXIT 0. +No app assets, no release, no merge performed here. + +### V-90 Batch 1 close-out / PR #4 +New PR **#4 OPEN** `arena/317fbb3c-vaultzero -> main` +(https://github.com/tonmoycoder/VaultZero/pull/4). It was created AFTER the +batch-1 commit because `main` and the branch were identical beforehand. +CI `42413851086` (`643f36f7…`) watched to exit 0, both jobs `success`; +`gh pr checks 4` -> `Frontend E2E & Security Tests pass (8m15s)` + +`Test & Build Verification pass (2m38s)`. No merge/close performed — an +explicit user decision is still required. Batch 1 CLOSED; resume with +`continue` for Batch 2. + +## V-91 — Batch 2 / Android status bar, horizontal lock, All items + sort pills (2026-11-08) + +**Re-baseline.** HEAD `52866f1`, tree clean, `origin/arena/317fbb3c-vaultzero` +and local in sync; PR #4 OPEN with pass/pass checks; no external advance this +turn. `main` still `c6ea9b2`. + +**Measurements BEFORE editing (real 360x800 + 1440x900 Chromium, isMobile).** +- Header: the "EN" control sat at `top=10px`; `.mobile-safe-top` computed + `padding-top: 0px` because headless Chromium (and many Android WebViews) + report `env(safe-area-inset-top) = 0px` while still drawing edge-to-edge. +- `html/body/#root`: `overflow-x: hidden` already, but + `overscroll-behavior-x: auto` (sideways rubber-band / canvas drift). + `document.scrollWidth === clientWidth === 360` already held. +- `All items` h2: `white-space: normal; flex-shrink: 1` (one line only by luck + at this width; a long folder name would wrap it). +- Sort row: `flex-wrap: nowrap; overflow-x: auto; justify-content: flex-start` + (V-87) but `touch-action: auto` and a visible `thin-scrollbar`. +- Lock card: `left=17, right=343`, `top=3`, no drift (guard only). + +**Test-first RED -> GREEN.** NEW `tests/vault-mobile-layout.spec.ts` B-06..B-10. +RED = 3 FAIL (B-06 padding 0, B-07 `"auto"`, B-09 `"normal"`), 2 PASS +(B-08 zero drift, B-10 centred card). Fixes, all minimal: +- `src/index.css` base layer: `html/body/#root` gained `width:100%`, + `margin/padding:0`, `overscroll-behavior-x:none` (overflow-x hidden kept). +- `src/index.css` mobile `@media (max-width:767px)`: + `.mobile-safe-top { padding-top: max(var(--safe-top), 36px) !important; }`. + Covers BOTH the vault column and the lock screen; `max()` lets a larger real + inset win; `!important` beats Tailwind `pt-3`. Measured after: EN `top=46px`. +- `src/components/VaultHome.tsx`: h2 -> `min-w-0 max-w-[70vw] shrink-0 + truncate whitespace-nowrap`; sort row `thin-scrollbar` -> `no-scrollbar` + + `touch-pan-x overscroll-x-contain` (the `.no-scrollbar` utility already + existed at `src/index.css:743`). +- `tests/vault-mobile-shell.spec.ts` M-03 was STRENGTHENED, not weakened: its + old `dialogPadding === "24px"` expectation conflicts with the new floor, so + it now asserts >=36px at inset 0 and 24, AND exactly 64px at inset 64 + (proving `max()` is live). The `.vault-shell` 24px assertion is unchanged. + +**Review -> improve -> review.** Re-captured 360x800: 36px clear band above the +header, one-line "All items", single swipeable pill row, no horizontal scrollbar +at document level. Targeted `vault-mobile-layout` + `vault-mobile-shell` += 26/26 pass (all M-01..M-19 guards survived). + +**Gates.** `tsc` 0; `npm run verify` 0 (38 harness, rustfmt, bundle budget); +node20 38/38; `npm audit --audit-level=high` 0; +`npx playwright test` **157 passed (10.1m)**; `check-live-release.mjs` EXIT 0. +No app assets/release changed; no merge/close performed. + +**Honest note.** The trailing "Zero-knowledge, end-to-end encryption..." note +is partly behind the floating nav at scrollTop=0 (it was before this batch +too); `.mobile-safe-bottom` reserves 96px so scrolling reveals it (M-04). +Batch 6 will assert its bounding box explicitly. Real-device status-bar +insets still need Antigravity verification. + +### V-91 Batch 2 close-out (CI / PR #4) +`b99fe78` pushed to `origin/arena/317fbb3c-vaultzero`. Both events watched to +exit 0: push run `42417004735` and `pull_request` run `42417008987`, each +`success`. `gh pr checks 4` -> Frontend E2E pass (5m53s) + Test & Build pass +(2m10s). No merge/close. Batch 2 CLOSED; resume with `continue` for Batch 3. + +## V-92 — Batch 3 / P2P real LAN networking, strict peers, camera + footer (2026-11-08) + +**Branch status from the platform: `pr_open` (PR #4).** Re-baseline: HEAD +`a0fc6ec`, `origin` in sync, `main` `c6ea9b2`, PR #4 pass/pass. A user-supplied +`image-6.png` showed an Arena UI error ("Something went wrong. Please try +again."), NOT a VaultZero screenshot — it was not treated as an app defect. + +### The working tree already contained UNCOMMITTED Batch-3 work +`git status` showed 4 modified + 2 new files that this agent did NOT create: +`src/lib/p2pMerge.ts`, `src-tauri/src/{lib,p2p_lan}.rs`, +`src/components/toolModals.tsx`, `tests/harness/lan-peer-contract.test.mjs`, +`tests/vault-p2p-lan.spec.ts` (all tagged `V-92`). Nothing was discarded or +overwritten: the work was REVIEWED, executed, and then IMPROVED. It contains: +- strict RFC-1918 peer validation in `isValidPeerAddress()` (rejects 8.8.8.8, + 1.1.1.1, 172.15/172.32, 192.169, 169.254, 127.0.0.1, localhost, + `tauri.localhost`, `192.1`, bad ports); private ranges still accepted. +- Rust `detect_lan_ipv4()` in `p2p_lan.rs`: `UdpSocket::bind("0.0.0.0:0")` + + `connect("8.8.8.8:80")` (no packet) + `local_addr()`, filtered to 10/8, + 172.16/12, 192.168/16; exposed as `commands::lan_ipv4` and registered in + `generate_handler!`. `core:default` capability already permits app commands; + no ACL file change needed. +- `SyncModal` asks Rust for the real LAN IP instead of + `window.location.hostname` (= `tauri.localhost`), with a manual RFC-1918 IP + input when detection fails. QR-photo fallback, handshake-gated `connected` + status, `CAMERA`/`INTERNET` manifest entries and the wrapping footer were + already correct (verified, not redone). `sync_vault_payload` -> Rust + `P2PManager::sync_from_peer` really performs a WebSocket handshake with an + ephemeral key exchange before any "Connected" state, so no fake connection. +- 30-assertion harness (with positive controls) + 5 Playwright tests B-11..B-15. + +### Reviewed results BEFORE my changes +`tsc` 0; `check:rust-format` 23/23; harness 30/30; B-11..B-15 5/5; manifest +has `CAMERA`, `INTERNET`, `ACCESS_NETWORK_STATE`, `ACCESS_WIFI_STATE`, +optional `android.hardware.camera`; QR-photo button is 44px tall with +`accept="image/*" capture="environment"`. + +### My RED -> GREEN improvement (B-16) +Found by review: the incoming code probed `lan_ipv4` ONCE in a mount effect. +On Android the Wi-Fi interface often appears later, so the pairing QR kept +`ip: ""` until the modal was closed and reopened. NEW B-16 stubs a late IP and +counts probes. First draft gave a FALSE GREEN because React StrictMode invokes +the mount effect twice; the test was hardened with an `__allowIp` flag plus a +probe counter. True RED: `before=2, after=2` (Discover added no probe). +Fix (minimal): `toolModals.tsx` — reusable `useCallback probeLanIp`, a +StrictMode-safe `aliveRef` mount effect, and `if (!lanHost) void probeLanIp()` +at the top of `handleDiscover` (never overwrites a manually typed IP). +Harness gained 3 more assertions (reusable probe + Discover re-probe + a +positive control). GREEN: B-11..B-16 6/6. + +### Full-suite regressions found and honestly resolved +Full `npx playwright test` first pass: `161 passed / 2 failed`. +1. M-19 `scrollHeight 643 > clientHeight 597`: the new REQUIRED 44px + manual-LAN-IP row made the Host body taller on a 390x844 phone. Measured + the body: `overflow-y: auto`, and after `scrollTop=48` the Discover button + ends at `bottom=733` vs `footer top=774` -> every control IS reachable. + M-19 was UPDATED, not weakened: it now asserts `overflowY === "auto"`, + bounded extra content (0 < extra <= 60), that scrolling actually moves, + and that the last control ends above the footer. Its desktop footnote + assertion is unchanged. V-88's "zero scroll" proxy is no longer valid now + that a user-requested control exists. +2. B-16 `after === before`: `void probeLanIp()` is fire-and-forget and + dynamically imports the Tauri core, so the counter ticked AFTER the click + under a loaded 2-worker suite. Fixed the TEST race with `expect.poll` + + `scrollIntoViewIfNeeded`, not by touching app code. +Second full pass: **163 passed (15.7m)**. + +**Gates.** `tsc` 0; `npm run verify` 0 (**39 harness files**, 23 rustfmt, bundle +budget); node20 39/39; `npm audit --audit-level=high` 0; full Playwright +**163/163**; `check-live-release.mjs` EXIT 0. Rust compilation/clippy is +Windows-CI-only (no local cargo) and will be confirmed by the watched run. + +**Honest limits.** `.vztransfer` snapshot-file transfer is NOT implemented +(never claimed); the encrypted QR + manual LAN IP are the working fallbacks. +WebView `onPermissionRequest` bridging, real-device camera/LAN P2P handshake, +carrier CG-NAT 10.x mis-detection and 120Hz behaviour need Antigravity on +real hardware. No release assets/signing secrets touched here. + +### V-92 Batch 3 close-out (CI / PR #4) +`09d5747` pushed. Both events watched to exit 0: push `42435123882` and +`pull_request` `42435126914`, each `success` for BOTH `Test & Build +Verification` (Windows Rust compile/clippy) and `Frontend End-to-End & +Security Tests`. `gh pr checks 4` -> pass (8m57s) + pass (2m2s). No +merge/close. Batch 3 CLOSED; resume with `continue` for Batch 4. + +## V-93 — Batch 4 / "Check for Updates" resilience, CSP, strict SemVer (2026-11-08) + +**Branch status from platform: `pr_open` (PR #4).** Baseline HEAD `8945e62`, +clean, remote in sync, `main` `c6ea9b2`, PR #4 pass/pass. No external advance +this turn. + +### Empirical CORS measurement (no guesses) +`curl -I -H "Origin: https://tauri.localhost"`: +- `https://api.github.com/repos/tonmoycoder/VaultZero/releases/latest` -> 200, + `access-control-allow-origin: *` (CORS-readable from a WebView). +- `https://github.com/tonmoycoder/VaultZero/releases/latest` -> **302 -> + `/releases/tag/v2.5.2`, but NO CORS header**. A renderer therefore CANNOT + read the redirect `Location`; it can only use that URL as an opaque + `no-cors` REACHABILITY probe. No `tauri-plugin-http` and no direct + `reqwest` dependency were added (over-engineering for this batch). +- `src-tauri/Cargo.toml` has no HTTP plugin; `reqwest` exists only in + Cargo.lock as a transitive dependency. + +### Root cause of the screenshot's false "Offline" message +`fetchLatestRelease()` made ONE request to `/releases/latest` and mapped ANY +thrown error straight to `status: "offline"`. A TLS/UA/CORS failure on an +otherwise online phone therefore rendered "Offline / Air-gapped network: +Cannot reach remote release server." Reproduced in real headless Chromium +before the fix (`.artifacts/probe-update.mjs`), where the cert error is +`ERR_CERT_AUTHORITY_INVALID`. + +### Test-first RED -> GREEN +NEW `tests/harness/updater-resilience.test.mjs` (40th harness) stubs +`globalThis.fetch` and drives `checkReleaseStatus(true)`; NEW +`tests/vault-updater-resilience.spec.ts` (B-17..B-20) drives the real UI with +`page.route`. RED = 11 harness FAIL (3 CSP tokens, 8s timeout, missing +`X-GitHub-Api-Version`, no fallback, false offline) + the real modal showing +Offline. Guards already passing: strict SemVer (same/newer/prerelease), +`Accept` header, genuine all-paths-down offline. + +### Fixes (minimal) +- `src-tauri/tauri.conf.json` `connect-src` gained + `https://*.githubusercontent.com`, `ipc:`, `http://ipc.localhost` + (kept `'self'`, `api.github.com`, `github.com`, + `objects.githubusercontent.com`). +- `src/lib/updater.ts`: `RELEASE_CHECK_TIMEOUT_MS` 8s -> **10s**; pinned + `X-GitHub-Api-Version: 2022-11-28`; one `withReleaseTimeout` helper bounds + BOTH fetch and JSON body read; `statusFromRelease` extracted so the primary + object and the fallback ARRAY share one strict-SemVer path. New chain: + `/releases/latest` -> `/releases?per_page=1` (403/429/5xx/throw/404) -> + opaque `no-cors` probe of the public redirect URL -> only if all three fail + is `offline` returned. A reachable network with unreadable metadata now + returns `status: "error"` with "The network IS reachable ...". +- `src/components/UpdateModal.tsx`: added the missing `status === "error"` + bilingual render block; otherwise the new honest message would be invisible. + `up_to_date` already printed `v2.5.2` ("✓ ... up-to-date"), so it was NOT + changed; `Update Now` still requires `compareSemver(remote, local) > 0`. + +### Review -> improve -> review +Unmocked probe AFTER the fix logs all three attempts in order +(`/releases/latest`, `/releases?per_page=1`, public redirect), each failing +with `ERR_CERT_AUTHORITY_INVALID` in this sandbox -> "Offline" here is HONEST, +not a false claim. On an online device B-17 (primary throw + fallback -> +up-to-date), B-18 (403 + fallback v2.6.0 -> Update Now) and B-20 (reachable +but malformed -> honest error) prove the false offline is gone. + +**Gates.** `tsc` 0; `npm run verify` 0 (**40 harness files**, 23 rustfmt, +bundle budget); node20 40/40; `npm audit --audit-level=high` 0; +`npx playwright test` **167 passed (13.0m)**; `check-live-release.mjs` EXIT 0. + +**Honest limits.** The renderer cannot read the 302 `Location` from +`github.com` (no CORS); reading that tag would need a native Rust HTTP command +which was NOT implemented here — no fake fallback was added. Real-device +TLS/UA behaviour needs Antigravity verification. `AndroidManifest.xml` +already had `INTERNET`, so it was not touched. + +### V-93 Batch 4 close-out (CI / PR #4) +`0e616c7` pushed. Both events watched to exit 0: push `42458168857` and +`pull_request` `42458173011`, each `success` for BOTH `Test & Build +Verification` and `Frontend End-to-End & Security Tests`. `gh pr checks 4` +-> pass (7m37s) + pass (2m35s). No merge/close. Batch 4 CLOSED; resume with +`continue` for Batch 5. + +### V-94 Batch 5 — offline OCR, strict RFC 6238 TOTP, official icon parity (2026-11-08) + +**Inspect exact files.** `src/lib/ocr.ts` ALREADY ran a real +`worker.recognize(processedBlob)` and returned `ret.data.text` (V-74), +streamed 10 -> 25 -> worker logger -> 100, and +`src/components/captureModals.tsx` seeded `OcrCopyPad` with `rawText` — the +image filename, byte size and dimensions never became the extracted result. +`src/lib/totp.ts` ALREADY enforced 16+ Base32 for pasted setup keys and +computed the RFC 6238 Appendix B 8-digit vectors correctly +(SHA-1 `94287082`, SHA-256 `46119246`, SHA-512 `90693936` @ T=59). All 15 +Android mipmap PNGs decode to the SAME dark neon-blue emblem: mean RGB of +the central region differs from the 512px master by at most 9 per channel. +So there was nothing to re-implement — the real defects were narrower. + +**RED, measured before any fix (3 files, 8 assertions + 1 browser test).** +- `parseOtpAuthUri()` accepted ANY non-empty `secret=` parameter. Therefore + `isValidTotpSecret("otpauth://totp/Acme:bob?secret=abc")` returned TRUE and + `resolveTotpConfig()` produced a one-byte HMAC key: a random short input + was treated as a valid 2FA account. 5 harness FAIL and B-23 browser FAIL. +- `src-tauri/icons/android/values/ic_launcher_background.xml` was the generic + Tauri white `#FFFFFF` while the official emblem's corner pixel is + `(19,24,45)` = `#13182D`; adaptive launchers therefore showed a white tile + behind a dark logo. 2 harness FAIL. +- `captureModals.tsx` caption said "on-device vision model" although the + engine is offline Tesseract. 1 harness FAIL. +- B-21 and B-22 were GREEN at RED time (correctly): real `eng` and dual + `eng+ben` offline OCR returned `demo@vaultzero.local` / `S3cret-42`, with + NO filename, size or dimensions in the result and a monotonic real + progress stream. This is the empirical proof that the old screenshot's + "OCR returns metadata" came from the old published APK, not this source. + +**Apply minimal clean code fix (3 source lines of behaviour).** +- `src/lib/totp.ts`: `parseOtpAuthUri()` now runs the inner `secret` through + `validateGoogleAuthSecret()` and throws when invalid; it returns the + normalised `cleanKey`. This one change fixes BOTH `isValidTotpSecret()` + (which catches parse errors) and `resolveTotpConfig()` (fails closed). + Both callers (`LiveTotp.tsx`, `toolModals.tsx`) already had try/catch, so + no UI change was needed. +- `ic_launcher_background.xml`: `#FFFFFF` -> `#13182D`. +- `captureModals.tsx`: caption -> "on-device offline OCR (Tesseract)". + +**New / updated test-enforced contracts.** +- `tests/harness/totp-uri-strict.test.mjs` (41st harness): positive control + proves a bare URL parser DOES return `secret=abc`; rejects 6 bad URIs; + accepts valid URIs; asserts fail-closed resolution; pins all three RFC + vectors. +- `tests/harness/icon-parity.test.mjs` (42nd harness): pure-Node `zlib` PNG + decoder (no image dependency), exact per-density dimensions + (48/72/96/144/192 launchers, 108/162/216/324/432 foregrounds), central RGB + fingerprint delta <= 20 vs master, adaptive background parsed from the + actual `` element, Windows `icon.ico` + NSIS references. +- `tests/harness/offline-ocr.test.mjs`: +9 guards (recognised text, no + metadata assignment, real progress stages, `OcrCopyPad` seeded with + `rawText`, `eng+ben` wiring, honest caption). +- `tests/vault-ocr-totp-icons.spec.ts`: B-21 real text/progress/no metadata, + B-22 `eng+ben` worker really recognises, B-23 strict 2FA in a real + browser (including the SHA-256 RFC vector). + +### Review -> improve -> review +The first GREEN attempt still had one FAIL: the new background guard read the +raw XML INCLUDING a comment that mentioned the old white hex literal. The +TEST was wrong, not the app — it now strips XML comments and parses the +`` value; the comment was also reworded. The assertion was NOT +weakened. A second tooling mistake was pointed at `/tmp/chromium` (a +directory -> `EACCES`); the executable is `/tmp/chromium/chrome`. No feature +was deleted or stubbed to satisfy a test. + +**Gates.** `npx tsc --noEmit` 0; `npm run verify` 0 (**42 harness files**, +23 rustfmt, chunk budget); node20 harness 42/42; +`npm audit --audit-level=high` 0 (0 vulnerabilities); +`npx playwright test` **170 passed (10.2m)** (previous 167 + B-21..B-23); +`check-live-release.mjs` EXIT 0 (`all 5 links resolve`). + +**Honest limits.** Real-device adaptive-icon rendering, Windows installer +artwork on a real installer, and Bengali handwriting OCR accuracy still need +Antigravity/physical-device verification; a pixel-fingerprint contract proves +the shipped assets are the same emblem, not how every launcher masks it. No +release was published and no merge was performed. + +### V-94 Batch 5 close-out (CI / PR #4) +`249f116` pushed. Both events watched to exit 0: push `42503258511` and +`pull_request` `42503261967`, each `success` for BOTH `Test & Build +Verification` and `Frontend End-to-End & Security Tests` (4/4 jobs). Re-observed +`gh pr checks 4` -> pass (8m46s) + pass (2m38s). PR #4 remains OPEN against +`main`; no merge/close. Batch 5 CLOSED; resume with `continue` for Batch 6. + +### V-95 Batch 6 — explicit 360x800 / 1440x900 bounding-box DOM audit (2026-11-08) + +**Inspect exact files + empirical telemetry (real headless Chromium).** +At BOTH viewports `scrollWidth === clientWidth === viewport` on Lock, Vault +and Sync; the `VaultZero` header wordmark sits at `y=36` (360x800) and `y=33` +(1440x900). After `Connect` on the Sync Scan tab, the three footer actions +measured: 360x800 -> `Close 153..224 / 701..741`, +`Send Vault 232..339 / 703..739`, `Receive & Merge 198..339 / 749..785`; +1440x900 -> all three inside `401..1039 / 674..766`. All fully in bounds. +`192.1` was rejected and a fresh vault was created at both viewports. These +were GREEN at RED time, so they are GUARDS, not fixes. + +**The one real defect.** `src/components/ui.tsx`'s shared `