Skip to content

Cloud-first Keeper release pipeline for every release (R5 first) #7061

Open

Description

@antuz123

Outcome: permanent cloud-first release pipeline for every Keeper release — R5 first

Standing goal: Implement one reusable, default cloud release process for R5 and all subsequent Keeper releases, not an R5-only workaround. R5 is the first end-to-end rollout/proving run. After that, new releases should automatically use the same reviewed CI/testing/signing/distribution process, with only release-specific tests, versions and approvals varying. The release owner's personal Mac may remain offline or locked indefinitely without blocking development, native validation, release-candidate distribution, or approved publication. This does not mean no macOS anywhere (hosted iOS builds use macOS), no real-device/hardware testing, no human security review, or automatic public release.

Permanent workflow requirement: Put platform builds, exact-SHA validation, safe evidence, protected signing and approved distribution into reusable GitHub/hosted CI workflows and update AGENTS.md/release instructions so future Codex handoffs default to cloud. Do not rely on private Codex sessions or manual, release-specific copying of steps. Keep physical testing separate and genuine. No automatic merging, signing or publication without explicit owner approval.

Coordinate with hosted native validation issue #7059 / PR #7060 and R5 #7058. R4 #7045 must remain entirely untouched. Keep this infrastructure independent of R5's app-feature token budget and do not add unrelated product features.

Existing verified basis (18 Oct 2026)

  • ci: hosted native validation and exact-head evidence gate (#7059) #7060 at d865ca9 ran actual hosted iOS and Android native builds and a narrow account-free passcode-screen UI smoke successfully on GitHub Actions run 39024530330. It remains an unmerged draft; current trusted sprint gate still expects Local validation. These passes do not transfer to R5 app source.
  • R5 R5: safe key deletion and startup fixes on submitted 2.6.4 source (draft) #7058 at 625435a has meaningful focused/hosted JS tests, but no R5 native, real-Device or end-to-end recovery/deletion acceptance. R4 baseline/reconciliation remains pending.
  • Existing ios/fastlane/Fastfile provides TestFlight upload lanes, and android/fastlane/Fastfile provides a Play internal-track upload lane, but remote production signing/credentials, exact-artifact controls and distribution workflow have not been validated.
  • Audit existing tracked native signing configuration for unsafe embedded credential fields without reproducing secret values in issue/PR/logs. If potentially live, use the project's private security process for any details and plan safe replacement/rotation with the owner; never assume the values are valid/current.

Two separately scoped implementation phases, not one oversized PR

Phase A — reusable hosted native release acceptance (first rollout on R5)
After reviewed #7060 is approved and trusted workflows are actually installed on sprint, run R5's own exact-head native build and stronger disposable integration tests: key deletion/reference protections, setup offline/retry/relaunch with true native persistence, cold-start/performance, account isolation, relevant normal-page runtime and basic recovery/regression cases where feasible. Generate synthetic native Realm/MMKV/wallet state in the job. Preserve safe artifact hashes and test evidence. Do not import Mac user accounts, keystores, private signer material or historical fixtures. Do not treat a narrow startup smoke as full coverage. R4-dependent app checks must wait for R4 release reconciliation. Keep unresolved high-risk security cases blocking, not waived. Limit test scenarios and runtime cost to a useful representative set; don't rebuild R4's entire private fixture.

Phase B — reusable signed beta and approved store distribution (R5 first, then every release)
Review existing iOS/Android Fastlane release lanes and identify all signing/profile/upload/service-account prerequisites, secure credential owners, channel mappings and available CI/CD options. Implement or propose a separate protected release-only workflow (hosted macOS for iOS, hosted Linux for Android, or Apple Xcode Cloud for iOS if more appropriate). Use trusted reviewed release source only, exact commit/version/build identity, least-privileged production secrets in protected environments, explicit owner approval before any signing, beta/TestFlight/Play internal upload, external submission or public release. Do not run PR code with production secrets, do not expose or auto-copy developer key material, do not commit/upload release credentials, do not regenerate an upload key without authorization. Preserve artifacts, signatures, distribution receipt and provenance. Keep final publication approval separate and explicit. If one-time Apple/Play provisioning or key custody requires owner action, describe that as setup, not an ongoing Mac-unlock requirement.

Physical signer and phone testing — outside owner's Mac
A named trusted human tester/maintainer (or appropriate securely operated real-device facility for supported checks) can install the approved TestFlight/Play internal build on their phones and exercise actual biometrics, cameras, NFC/USB/BLE, hardware signers, signing policy/recovery scenarios. Record outcomes against the exact distributed artifact, devices/firmware and head SHA. This may require real devices and a person; it must not need the owner's Mac or be replaced by emulator evidence. Use owner-approved physical scope and preserve existing critical-defect restrictions.

Scope, delivery and noninterference

  • Complexity is likely to exceed ~20,000 changed code/test tokens if hosted feature matrices and protected dual-store release signing are combined. Split A and B into separate PRs, with a short estimate before each; if either grows beyond ~20k or two substantial independent features, warn and trim/defer. No new release paperwork/ceremony unless required for safety.
  • Do not modify R4 code, refs, private fixtures, Mac, release gate or approval process during active R4 work. Do not silently merge ci: hosted native validation and exact-head evidence gate (#7059) #7060; it includes dependency/iOS-native changes that require review against R4's eventual baseline.
  • Preserve failure evidence, strict exact-head security checks, real hardware requirements, independent code review and existing release sign-offs. Never claim a R5 build was validated using ci: hosted native validation and exact-head evidence gate (#7059) #7060's infrastructure-only app.
  • No merge, signing, release, TestFlight/Play upload or public publication is authorized by this issue. Prepare candidate PRs and report actual test/blocked status, secure one-time setup needed, ongoing Mac requirements (target zero), and the next action.

Acceptance

  1. R5 and subsequent release code can be developed, built, tested on hosted simulators/emulators and reviewed with the personal Mac off; shared workflows are the documented default.
  2. With owner authorization, candidate app can be built/signed and distributed to phones without unlocking that Mac.
  3. Physical checks remain genuine but run on designated phones/signers by authorized testers, not through a Mac session.
  4. After independent review and separate explicit owner approval for each release, R5 and future releases can be delivered through the same protected hosted workflows without the personal Mac.
  5. Until each capability is proven, label it NOT IMPLEMENTED / NOT VERIFIED and preserve current safe fallback.

Activity

  1. changed the title [-]Mac-independent R5 validation and release delivery (phased)[/-] [+]Cloud-first Keeper release pipeline for every release (R5 first)[/+] on Oct 18, 2026
  2. antuz123 commented on Oct 18, 2026

    @antuz123
    CollaboratorAuthor

    Phase A candidate: draft #7062, stacked on #7060 at 59c821ce195a5c063d99045fd97b5ac605433f62.

    Implemented reusable automatic sprint-PR workflow wiring, exact-source/base/run-attempt provenance, secret-free development workers, a separate synthetic native Realm/MMKV persistence probe and durable cloud-first release instructions. Host contracts 70 Node +40 Python PASS; all workflow actionlint and diff checks PASS. Approximate added-text scope is 19k tokens including docs; app matrix expansion and Phase B are deferred from this candidate.

    Actual new native execution: NOT RUN / NOT VERIFIED. Full Phase A acceptance remains BLOCKED: five required app scenario groups are still NOT IMPLEMENTED, including R5 deletion/reference privacy, setup/relaunch, app account isolation, runtime/recovery regression and performance. R4-dependent checks wait for baseline reconciliation; no startup/module probe substitutes for those cases. Phase B signing/distribution is NOT IMPLEMENTED / NOT VERIFIED. Physical checks NOT RUN.

    A malformed inherited emulator action reference was flagged on #7060; the new worker uses the verified immutable v2.34.0 commit. Next: review/correct #7060, obtain owner approval before installation, then review/install Phase A and execute the reconciled R5 exact head with the missing app scenarios implemented. Future sprint PRs reuse the same caller/worker automatically. Current legacy workflow/gate remain unchanged; temporary staged duplicate-build cost is documented.

    R4 code/refs/fixtures/Mac/gate/approvals remain untouched. No merge, signing, beta/store upload, submission or publication was performed or authorized.

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

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions