Skip to content

feat(browser): validate authenticated sites and platform integration #19

Closed as not planned

Description

@trmosala

Problem

Synthetic fixtures do not establish authenticated Teams/SSO, actual picker or cross-platform compatibility.

Implementation order

Browser backlog item 15 of 29. Phase: Compatibility and recovery. This is ordered planning, not a claim that Codex implements every item. Items 24–29 are optional additions.

Acceptance criteria

  • Record a reproducible manual matrix for Teams sign-in/meetings, SSO and native file pickers.
  • Use user-controlled test accounts and supported Windows/macOS builds; record actual results and limitations.
  • Add deterministic regression fixtures for discovered failures without storing real credentials.

Dependencies

Engineering and validation

Reuse existing Chromium/Electron/OS facilities where available. Preserve main-process ownership and permission enforcement, add no placeholder controls, use existing i18n patterns, and test behavior with synthetic fixtures. Document platform limitations. Baseline: browser-tabs at 47fef3e; see packages/cm-browser/README.md and password-security.md.

Activity

  1. trmosala commented on Sep 17, 2026

    @trmosala
    OwnerAuthor

    Reproducible manual matrix

    Status: preparation only; authenticated/platform acceptance remains open. No real-account, physical-picker or packaged-build result is claimed.

    Before running: record date, tester, commit/build identifier, Electron/Chromium versions, OS version/architecture, locale, and whether the profile is fresh or returning. Use a dedicated user-controlled test account/tenant and an approved disposable test meeting. The user enters credentials and completes MFA privately; do not put passwords, cookies, tokens, full callback URLs, meeting links or account identifiers in logs, screenshots, fixtures or issue comments. Keep agent access off during credential work. Use a supported branded Windows/macOS build containing the tested commit, not the running WPP session or an older installed build. Record artifact provenance; do not replace or restart the installed app for this matrix without approval.

    Run each row separately on Windows and macOS:

    Case Reproducible steps and expected observation Windows macOS
    Teams sign-in Open the official Teams web entry in a fresh private browser tab; choose web use if offered; sign in with the designated test account and complete MFA manually. Record origin sequence without query/fragment data and whether the authenticated landing page loads. Repeat from a returning test profile, then sign out and confirm account switching does not reuse unintended identity. Record unsupported-browser or tenant-policy messages rather than bypassing them. Not run Not run
    Teams meeting and permissions Join only the disposable test meeting with consenting participants. First deny camera/microphone in both app/OS prompts; verify no capture. Then explicitly allow and verify preview, audio and join/leave. Revoke permissions during the meeting, record when capture actually ends and whether document replacement is needed; do not assume immediate track termination. Close the test tab and check OS capture indicators. Screen-sharing requires separate explicit approval. Not run Not run
    Redirect SSO On an approved test app, follow its actual provider redirect. Save/select credentials only for each exact receiving origin, with agent access off; confirm the dialog names that origin. Cancel once, then freshly select and approve; submit manually. Verify no inferred cross-origin account sharing or automatic submission. A provider that uses unsupported iframe/foreign-action fields is a recorded limitation, not a reason to relax checks. Not run Not run
    Popup SSO Open the real provider popup from the test app; verify a private child tab and opener return. Select the provider-origin account explicitly. Exercise cancel/retry and callback/close; verify the parent returns to the intended test account without unexpected fill or submission. While a separate fill consent is pending, navigate or change active tab and answer the stale dialog: delivery must reject and require fresh selection. Not run Not run
    Top-document file picker In a controlled test upload destination, use synthetic non-sensitive files. Test a fresh private tab and a guarded tab separately. Open the actual OS picker: cancel gives no file; approved selection gives only the selected file to the receiving page. On guarded tabs verify receiving-origin labeling, Block policy, navigation/revocation while choosing, and no stale delivery. The page may upload immediately after selection, so use only an approved test endpoint. Not run Not run
    Direct-child picker Use controlled direct HTTP(S) same-origin and cross-origin child inputs, with explicit top/receiver host access and transfer permissions. Manually open the chooser; verify default-cancel consent names both exact origins before the real OS picker. Cancel either step; then approve a synthetic file and verify only the original child receives it. Navigation, frame replacement or revocation while pending must discard selection. Nested/opaque/directory cases remain unsupported. Not run Not run
    Native usability Repeat consent and picker cancellation/approval with keyboard only; check focus return, default Cancel/Escape, readable origin/username labels, long paths, display scaling and screen-reader announcements (Narrator/VoiceOver). Record actual locale and untranslated fallback. Do not equate programmatic dialog responses with this result. Not run Not run

    For each run, append PASS / FAIL / BLOCKED / NOT RUN, exact step, sanitized observation, and evidence reference under the recorded build/OS. A failed step stays failed even if a retry succeeds; record both. Clean up only the dedicated test data and restore test-specific permissions.

    Current blockers: no designated real-provider test account/meeting or candidate packaged-build validation session has been supplied; this workspace session is Windows, with no macOS validation environment established. No real-provider failure has been discovered in this pass, so no speculative fixture is added. For an observed failure, first reproduce with synthetic local origins/files and leave a deterministic regression without real credentials.

    Existing #17 frame and #18 exact-origin account fixtures are synthetic boundary evidence only. Optional CM_BROWSER_LIVE_SMOKE=1 exercises public unauthenticated pages in a temporary profile, not this matrix. #19 and external #34 assurance remain open.

  2. trmosala commented on Sep 19, 2026

    @trmosala
    OwnerAuthor

    Validation follow-up for #25 after publication of c7efee8 on browser-tabs: the bounded console/network implementation is complete; release/platform validation is not.

    Add diagnostics to the existing approved packaged Windows/macOS validation session: exercise separate console/network default-cancel consent, keyboard and screen-reader access, cancellation and stale-source rejection, origin/task labels and actual locale fallback. Real authenticated-site checks still require designated user-controlled test accounts; agents must not collect credentials or raw request/message payloads. Record the exact build/OS and PASS/FAIL/BLOCKED/NOT RUN using the existing matrix conventions.

    Current evidence is synthetic Windows/Electron 44.3.0/Chromium 152.0.7977.78. Packaged/macOS/Linux behavior, physical network-consent usability and reviewed translations are not established. OOPIF/worker attribution fixtures remain a coverage follow-up; current counts explicitly permit ancestor-attributed workers and preexisting in-flight completions and do not claim cache/service-worker completeness.

    No release or installed-process restart was performed. #19 remains open and #34 retains its separate security-assurance gates.

  3. trmosala commented on Sep 19, 2026

    @trmosala
    OwnerAuthor

    Follow-up from #11, published at 97ab8bd: unified permission controls and navigation read approval meet scoped implementation acceptance. Release validation remains open here. Verify packaged settings layout, physical keyboard/screen-reader access, pending-consent revocation and native dialog behavior on supported Windows/macOS builds; record unsupported or untested Linux behavior explicitly. Verify localized copy and fallback; earlier i18n failures are not claimed repaired. Use the existing build/OS evidence matrix. Current evidence is mounted Solid settings plus disposable Windows Chromium fixtures, not packaged or physical-input assurance. #34 retains separate password-manager security gates. No installed process restart or release occurred.

  4. trmosala commented on Sep 19, 2026

    @trmosala
    OwnerAuthor

    Follow-up from #12 at f617913: bounded scrolling and waits meet scoped implementation acceptance after fixing final-delivery revocation. Packaged/platform validation remains here: physical input and accessibility, complex transforms/zoom, real-site wheel interception, packaged timing, supported macOS/Windows behavior and explicit Linux limitations. Verify the seven existing English-fallback driver messages; no translation repair is claimed. Record build/OS and PASS/FAIL/BLOCKED/NOT RUN. Current evidence is synthetic router/bridge regressions and disposable Windows Electron 44.3.0 / Chromium 152.0.7977.78 fixtures, not release assurance. Full native smoke still emitted Mojo Widget, fixture ERR_EMPTY_RESPONSE and ObserverList reentrancy diagnostics. #34 retains separate password-manager security assurance. No installed process restart or release occurred.

  5. trmosala commented on Sep 21, 2026

    @trmosala
    OwnerAuthor

    Validation preparation on 2026-09-21, macOS 26.5 build 25F71, arm64, Bun 1.3.14. Source baseline 9d3cafc; discovered build fix committed and pushed as c36b886 on browser-tabs.

    FAIL then fixed: native-vault/macos-entry.m used explicit_bzero, which is undeclared in the installed Apple SDK. Replaced it with guarded memset_s, enabling its declaration before framework headers. The SDK declares availability since macOS 10.9, below our macOS 11 deployment target. Native helper compilation PASS for arm64 and x64 using scripts/build-vault-auth.ts. This does not establish runtime erasure of all credential copies.

    Verification: desktop bun test PASS, 569 passed, 4 skipped, 0 failed; desktop bun typecheck PASS; pre-push typechecks PASS, 31 tasks. Read-only review found no defects in the scoped fix. No new runtime-erasure regression fixture was added; native compilation is the direct check for this compile failure.

    Candidate packaging BLOCKED: GitHub run 35610889733 never started due to the account billing/spending restriction and produced no artifacts. Locally, lifecycle builds encountered an AE staging EEXIST collision; 22 staging tests and standalone staging passed. Direct branded unsigned production prebuild subsequently passed. Both bun x and direct electron-vite build invocations returned tool-level ChildProcess.kill errors without a compiler result. No usable candidate package is claimed.

    Teams sign-in/meetings, redirect/popup SSO, physical native pickers, accessibility and the follow-up packaged checks are NOT RUN in this pass on either platform. User confirmed a Mac and dedicated account are available, but no Windows test environment. No credentials were requested or collected, no installed app was replaced or restarted, and no release was published. Keep #19 open until candidate builds and actual sanitized Windows/macOS matrix results are available; #34 remains separate.

  6. trmosala commented on Sep 21, 2026

    @trmosala
    OwnerAuthor

    Closed at the owner's explicit request. The validation gaps and blockers recorded above remain unresolved; this closure does not claim completed authenticated-site or Windows/macOS platform validation.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions