Repository navigation
Description
Sequential Keeper delivery plan — R4 → R5 → R6 → R7
Owner decision (21 October 2026): one active release scope at a time. Parallel engineering is permitted only within the currently active release; separate repositories/PRs are work packages of that release, not independent release programs. Do not interrupt, cancel, pause or discard already-running Codex/CI/PR work, even if it belongs to R5 or Assisted Multisig. The sequential rule controls NEW work assignment and release integration, not already-running sessions. Existing later-release branches and drafts continue to their natural checkpoint and preserve results; do not initiate unrelated new work outside the active release. At each release start, inspect current PR heads, comments, checks and unfinished work, adopt completed work, and resume the remainder from its latest verified checkpoint rather than restart or duplicate it. The next release starts after the previous release is closed by the owner, with explicit acceptance of any deferred noncritical items. No automatic merges, deployments, store submissions or public releases.
Owner clarification, 22 October 2026 — release numbering and latest allocation: R5 = 2.6.5; R6 = 2.6.6; R7 = 2.6.7. Keep work within these numbered releases. The proposed M1/M2 releases are withdrawn; no intermediate maintenance release is scheduled. Hide-wallet #7075 stays in R6 as the higher-priority correction; dust/manual-spendability #7077 moves to R7, following the owner's suggestion to reduce R6 scope. This supersedes the earlier allocation of both bugs to R6. Separate task branches and PRs are work packages within their assigned release, not additional releases.
R4 — 2.6.4 delivery and prerequisites for future releases (active)
- Mobile R4: 2.6.4 mobile integration and release-readiness candidate #7045 submitted-source checkpoint is
1d8a86a7462d935cbd0e28b857934b5718c152fc: iOS620/Android627 are already submitted, signed artifacts/PGP verified. Latest posted authenticated store states are 21 October WAITING_FOR_REVIEW (Apple MANUAL) / IN_REVIEW (Play Managed publishing ON); fresh store readback is unavailable in the current cloud session. Preserve the owner-reported installed iPhone2.6.4 separately. Do not rebuild/resubmit; public release remains held under the current instruction. Required Tests/local gate pass for the recorded selected subset, not the full original matrix; current public-v2 genuine lane and CodeQL security-result failures remain explicit. - Private Relay Account Details: UI development #158 deployment and hosted-v3 acceptance are completed at
29597b27eeeb299205bf7d0555db0a90c38cc4be(accepted result); do not restart deployment. Historical private Account Details: UI and Redux-saga integration #159 proof is source-scoped; fixed submitted-app/current-backend CI is separate from native acceptance and does not pass the public295 matrix. Website Add accounts functional #19 is merged/deployed, strict8/8 HTTP acceptance complete; only installed-app/device acceptance remains. Desktop 7. Import External Wallet #100 exactef050663d33be51f25fb10310d5a0549b111e273retains approval; #102 exact06233e6a576ae1c9d0a49a34c253b3e9354b9efcretains peer-reviewed guards. Existing QA artifact11615331902 is preserved in cloud, GitHub expiry23Oct12:15:31UTC. Desktop/phone pairing, installed links and physical signers remain unrun. SigningServer BitcoinLogin (TM) with Keeper #6 source/CI is complete atdf9485e9945cdf0bdc231c70329d8905125d42ff; approval-rule settings and actual deployed-service/compatibility evidence remain separate. - ci: hosted native validation and exact-head evidence gate (#7059) #7060's hosted iOS/Android native passcode smoke is complete on its own source, not submitted R4. Keep its dependency/Pod/infrastructure adoption in the existing R5 review/integration; changing the submitted R4 inputs would create an unnecessary new build cycle. Do not import cloud signing/distribution into R4 or waive outstanding acceptance.
- Fix confirmed critical or release-blocking defects in the included flows; test their adjacent setup, restore, delete, account isolation, signing, navigation and upgrade regressions. Preserve already owner-parked noncritical C8 and recovered-fixture investigations without silently marking PASS.
- Exit: release owner closes 2.6.4 with documented tested/deferred items, backend state and deployment state, and remaining exact blockers; no false release-readiness claim.
R5 — 2.6.5: stability and cloud-first release capability
Reconciliation, 22 October: the scheduling paragraph below preserves the owner's original sequencing decision. Its “implementation is not claimed” wording is historical: separate #7067 now exists and is included in #7068; exact native and affected-phone acceptance remain pending. Keep #7066 open and preserve the original-phone report without automatically reopening the deferred R4 Delete campaign.
- 22 October owner clarification — preserve the R5 work already underway: R5: safe key deletion and startup fixes on submitted 2.6.4 source (draft) #7058 keeps its current implementation scope. Queue backup defect Protected backup still fails at local capture after 2.6.5 update #7066 as the first separate follow-up patch after that implementation reaches its checkpoint, targeting inclusion before 2.6.5 is publicly released. Use a separate branch/PR from verified R5 source; do not interrupt, expand or duplicate active work. Complete the focused backup/deletion/UI correction and its targeted acceptance before reviewing the combined release candidate. This supersedes the earlier immediate R5 scope addition; implementation is not claimed. The provisional patch estimate is 5k–10k changed code/test tokens, with the earlier combined 18k–28k advisory retained in Protected backup still fails at local capture after 2.6.5 update #7066. All-copy legacy purge and unrelated parked investigations remain excluded.
- Source reconciliation is already completed: R5: safe key deletion and startup fixes on submitted 2.6.4 source (draft) #7058 incorporates submitted R4 source
1d8a86a…; fix: protected-backup follow-up and accurate Version History #7067 implements the separate Protected backup still fails at local capture after 2.6.5 update #7066 backup/history/UI follow-up and R5 Cloud integration: app checkpoint and bounded delivery packages #7068 includes both. Preserve all running R5 work and native/paired-CI jobs; do not duplicate those implementations. R5 documents calling the baseline shipped/live/released do not establish fresh store state or owner R4 closure. Native acceptance, measured startup/PIN journeys and affected-phone confirmation retain their actual incomplete scopes. - Incorporate the relevant known defects in these same flows (e.g. Deleting keys is unintuitive and requires hiding keys first #7006, R05: Android startup and first-use UX quality — remove fixed wait and duplicate setup error #7046, R5 follow-up: Android More → Wallets exits to launcher after backup deletion #7055 if reproduced and in scope); explicitly triage Later: investigate second Android backup POST after a lost reply #7056 and other owner-deferred items without silently expanding or treating them as passed.
- Finish reusable hosted iOS/Android feature acceptance and release candidate evidence, then separately scoped secure hosted signing/beta distribution only if its one-time prerequisites, independent review and approvals fit. Split the infrastructure into bounded work packages inside R5; if full signing/distribution exceeds release complexity, retain a documented existing release route and schedule the reusable signing extension for a later release.
- Exit: stable native-tested R5 app and a repeatable verified cloud testing route; do not require the personal Mac for ordinary development/builds. Physical devices/signers and explicit signing/publication approvals remain real.
R6 — 2.6.6: hide-wallet fix and Assisted Multisig functional testnet
- Updated release placement, 22 October: include hide-wallet Hide wallet fails locally with “Something went wrong!”; investigate local persistence and backup guards #7075 in R6; explicitly defer dust/manual-spendability Mark Spendable leaves/reapplies Do Not Spend: refresh can overwrite manual choice and UI reports success too early #7077 to R7 following the owner's latest suggestion. Complete the hide correction as a focused PR within R6; preserve R5's current scope and already-running work.
- First priority — hide-wallet Hide wallet fails locally with “Something went wrong!”; investigate local persistence and backup guards #7075. Correct the reported failure to hide locally, including the demonstrated local-operation/error-reporting failure modes. A toast-only correction is insufficient. Adopt the verified completed R5 baseline and relevant existing fixes. Estimate 2k–5k changed code/test tokens, subject to affected-phone diagnosis. Require hide/unhide, persistence, accurate feedback, account/backup safeguards and native acceptance.
- Known defect deferred to R7 — Mark Spendable leaves/reapplies Do Not Spend: refresh can overwrite manual choice and UI reports success too early #7077. Retain the demonstrated refresh/manual-spendability race and its explicit deferral in R6 evidence; do not mark it fixed or passed. It can undo either Mark Spendable or Mark Do Not Spend, so it is a functional coin-control limitation. R7 must correct it before Assisted Multisig production readiness. Existing handling of critical or release-blocking findings still applies.
- Scope advisory: the hide-wallet correction adds 2k–5k provisionally to R6; the dust correction's 3k–6k is allocated to R7. These are defect estimates, not whole-release estimates. At R6 kickoff reconcile existing work, estimate the remaining Assisted Multisig and hide-wallet scope, and apply the existing ~20k / more-than-two-substantial-features advisory. If needed, revisit lower-priority work for a later numbered release; do not create intermediate maintenance releases. New R6 implementation follows R5 closure under the existing sequential rule.
- Use preview PR Add isolated Assisted Multisig preview and verification foundations #7041 as reference, not as a real wallet claim. Freeze a precise Keeper-assisted 2-of-3 policy: owner keys, assisted/server key custody and authorization, independent recovery, loss/compromise cases, service outage, timelocks, and key ownership. Do not copy Bitkey assumptions without design review.
- Implement one complete real testnet path: wallet setup, independent key/policy verification, backup/restore, receive, UTXO selection, PSBT creation, policy-constrained assisted signing, finalization/broadcast, and recovery without service dependency as promised by the policy.
- Fix all reproducible defects in that journey and neighboring signer, pairing, key metadata, fee/change-address and recovery flows that threaten correctness; include actual device/policy acceptance. Do not expand to every device or advanced policy.
- Exit: accepted hide-wallet correction with native evidence, plus a real spendable testnet wallet with repeatable end-to-end and failure-case evidence, not a simulated preview. Record Mark Spendable leaves/reapplies Do Not Spend: refresh can overwrite manual choice and UI reports success too early #7077 as explicitly deferred to R7, not passed. Assisted Multisig is not yet advertised as safe for real funds.
R7 — 2.6.7: dust/manual-spendability fix and Assisted Multisig security/production readiness
- Assigned bug fix — dust/manual-spendability Mark Spendable leaves/reapplies Do Not Spend: refresh can overwrite manual choice and UI reports success too early #7077. The owner permits moving this correction from R6 to R7 so hide-wallet can take priority in R6. Correct refresh overwriting the user's latest choice and premature success feedback in one focused PR. Adopt completed R6 work and relevant shared fixes from Hide wallet fails locally with “Something went wrong!”; investigate local persistence and backup guards #7075. Cover both Mark Spendable and Mark Do Not Spend, persistence through refresh/relaunch, accurate live badges/coin-selection eligibility, account isolation, backup-independent local persistence, and native iOS/Android acceptance.
- Scope advisory: Mark Spendable leaves/reapplies Do Not Spend: refresh can overwrite manual choice and UI reports success too early #7077 adds 3k–6k changed code/test tokens provisionally to R7, subject to reconciliation with upstream fixes. At R7 kickoff estimate the entire remaining security/production and bug-fix scope under the existing ~20k / more-than-two-substantial-features advisory. No intermediate maintenance release is created.
- Independent cryptographic/security review, actual device/signer tests, key-loss/recovery, backup restoration, assisted-service outages, authorization abuse, policy enforcement, change-address and PSBT correctness, upgrade/rollback, account isolation, data retention and operational monitoring.
- Repair any discovered defects in the same user journeys; production configuration, backend operator readiness, audited artifacts and safe staged release with explicit approval. Include only bounded hardware/policy compatibility needed for the initial supported release.
- Exit: Mark Spendable leaves/reapplies Do Not Spend: refresh can overwrite manual choice and UI reports success too early #7077 corrected and accepted, including both directions of manual spendability, plus the first Assisted Multisig release candidate suitable for real funds after security and human approval. The deferred coin-control race must be resolved before that production-readiness claim; no release until critical defects resolved.
Release discipline
- One release owns the active engineering queue; within it, Cloud/app/backend/website/Desktop/CI tasks may run concurrently if in that release's approved scope.
- A discovered bug in the same flow must be investigated, classified, and fixed before release when critical or release-blocking. Noncritical adjacent bugs may be explicitly deferred with a tracked owner decision. No unrelated feature creep.
- Before freezing each release scope, estimate changed code/test tokens and independent features. If >~20,000 tokens or >2 substantial independent features, warn and reduce scope by proposing lower-priority work for a later numbered release. Keep the R5 = 2.6.5 / R6 = 2.6.6 / R7 = 2.6.7 mapping; do not introduce intermediate maintenance releases. R4 is a pre-existing large integration: avoid broadening it; infrastructure addition is conditional on low-risk isolation.
- Existing OneKey, reproducibility, Claude reviewer, Anthropic scanner, broader cloud signing and experimental work are not separate active delivery streams. Preserve PRs; assign to a future release when selected. Do not silently merge them into R4–R7.
- Do not treat GitHub issues/this plan as authorization to launch Codex sessions, merge, deploy, sign or publish. Follow AGENTS.md and latest exact-head PR evidence.
References: mobile #7045, R5 #7058, Assisted preview #7041, hosted validation #7060, cloud-first #7061, private Relay #158/#159, website #19, Desktop #100.
Clarification — continuity of work (21 October 2026)
- This roadmap is not a stop-work order for any already-running Cloud/Local/CI activity. Do not cancel active jobs or ask contributors to stop. Let in-flight work reach a safe, documented checkpoint.
- R5, R6 and R7 scope means complete the remaining work, not reimplement features already done. Verify the actual GitHub state at handoff and reconcile work from other branches/PRs before estimating the remaining delta.
- Parallel activity already underway can finish; subsequent new assignments should be grouped under the current release. Existing parallel drafts are future-release inputs, not automatic authorization to merge or ship.
- Only unresolved work and same-flow defects should enter a release's remaining scope; mark verified completed items done, and keep owner-deferred items explicitly deferred.
R4 evidence reconciliation — 22 October 2026
Source ancestry, store submission, production deployment and installed-device acceptance are separate facts. R4 remains active pending owner closure. Current public-release hold governs further release actions; no fresh Apple/Play status is claimed without authenticated readback. Older-copy recovery/Delete/bootstrap/safety remain NOT RUN / OWNER-DEFERRED; C8/#7056 remains FAILED / PARKED and #7057 NOT ACCEPTED / PARKED. Existing native passes, failures and owner decisions keep their original SHA scopes. No deferred case is passed or automatically reopened. Remaining exact actions: fresh store status; retrieve existing signed mobile bytes into approved cloud storage; finish available cloud Desktop/paired proof and retain outstanding physical acceptance; reconcile SigningServer's admin rule and deployed-service evidence; review security/protocol limits and explicitly record R4 closure. No personal-Mac dependency, automatic merge, deployment or public release.
Completed cloud follow-ups, 22 October: private Relay #161, exact submitted1d8/deployed29597, passed43/43+3/3 with zero skips in run39832222599; public295 remains2PASS/293FAIL and CodeQL remains red. Desktop #103 full CI and all12 existing guards passed; existing QA artifact is durably retained in owner-only cloud storage with complete re-download/hash/byte comparison. R4 completion ledger; isolated R5 baseline-doc corrections #7089/#7090 preserve original R5 implementation refs. None of these results supplies missing store/installed-device evidence or R4 closure.
Desktop hosted subset final: isolated #104, unchanged QA source ef050663, verifies hashes/identity/strict signature, native launch and connected Welcome. Run39834105821 remains BLOCKED at Get Started despite true permissions and activation; QR0/no test packets, cause unresolved. Four bounded attempts are preserved. Use interactive cloud QA or a proved harness correction for QR/channel acceptance; phone/signer/PSBT remains separate. Retained QA0.2.6-qa.1 is not final production/notarized/cross-platform/PGP distribution; those artifacts and operation-specific approvals remain open. No personal-Mac fallback or automatic publication.
Publication refresh — 24 October 2026
Existing ledger#7092 is updated at 3692edc94a2d3ccae26116048e88e2c262c707dd; no duplicate PR or implementation. Completed submitted-runtime #7151/private#191 already passed20/20 caller proofs +14/14 raw-source/cleanup controls with zero skips in run40066799617. Remaining275 obligations are engineering coverage backlog, not a new blanket release condition; owner-deferred cases are not reopened. R5-derived proof/source and native#7149 continue separately.
Desktop's original Actions QA artifact11615331902 has expired/returns404. The privately retained original bytes remain available, owner-only metadata rechecked24October; no rebuild is needed to recover that test input. Get Started/QR acceptance and final production packages remain unresolved. Store access/original mobile files, existing CI/security/Sonar reachability and SigningServer rule/runtime limits remain separate. No new native build, acceptance attempt, workflow rerun, merge, deployment, signing, store action or public release was started by this publication. Existing owner closure/deferral decisions and ongoing R5 work are preserved.
Activity
[-]Sequential release roadmap: R4 → R5 → R6 → R7 (Assisted Multisig)[/-][+]Sequential release roadmap: R4 → R5 → M1 → M2 → R6 → R7 (Assisted Multisig)[/+][-]Sequential release roadmap: R4 → R5 → M1 → M2 → R6 → R7 (Assisted Multisig)[/-][+]Sequential release roadmap: R4 → R5 → R6 → R7 (Assisted Multisig)[/+]