Repository navigation
feat(harness): evidence-bound completion, coverage CI, drift-safe verification - #41
Merged
Merged
Conversation
Implement the merchant-onboarding reference feature end to end: pure domain validation (value objects, child/root aggregates, stable failure codes and paths), raw form input, use-case and typed repository boundaries, a deterministic fake adapter, one feature Cubit with one-shot effects, four design-system step pages with dynamic owner rows, review masking, an isolated GoRouter route, a localized Home entry, and mirrored domain/cubit/widget/DI tests. Adds the merchant feature boundary to the architecture lint configuration and makes the validated aggregate, raw input, reference snapshot, and bank schedules immutable. Co-authored-by: CommandCodeBot <noreply@commandcode.ai>
Add the 94 merchant-onboarding keys (labels, actions, errors, and declarations) to the en, id, ar, en-XA, and ar-XB ARB catalogs with generated output, keeping the demo fully localized. Co-authored-by: CommandCodeBot <noreply@commandcode.ai>
Move the plan to completed with the full acceptance record: three work packages, controlled verification evidence (profile full, attempt 1), the three reviewer-fix rounds (typed-id conditionals, immutable snapshots and aggregates, complete focus routing, bank schedule defensive copy), runtime evidence limitations, and the backend integration follow-up. Co-authored-by: CommandCodeBot <noreply@commandcode.ai>
Sync the accepted backend artifact (b10a5bf9) into the mobile-owned snapshot with its full SHA-256 digest and revision, adding the authenticated reference-data and idempotent application submission paths. Co-authored-by: CommandCodeBot <noreply@commandcode.ai>
Add the authenticated remote datasource (core host, standard envelope, one Idempotency-Key per invocation), Freezed reference/submit DTOs, pure mappers with lossless integer basis-point conversion and leading-zero account numbers, and a repository adapter mapping every documented backend error code into the feature failure taxonomy. Production DI now registers the real datasource and repository; the deterministic fake moves to test support and the merchant step-widget pairs are recorded as reviewed-acceptable duplication. Co-authored-by: CommandCodeBot <noreply@commandcode.ai>
Extend the failure taxonomy with server validation, registration conflict, stale terms, and idempotency outcomes; route recognized server field failures through the existing Cubit step/focus behavior; and add localized repository-level copy (unauthorized, already submitted, conflict, stale reference/terms, idempotency, retryable) across all ARB catalogs. Co-authored-by: CommandCodeBot <noreply@commandcode.ai>
Move the plan to completed with the acceptance record: contract sync from the accepted backend commit b10a5bf9, remote adapter and DTOs, failure taxonomy and localization, production DI wiring, reviewed duplication allowlist, and controlled full verification evidence (profile full, attempt 1). The live E2E plan remains queued. Co-authored-by: CommandCodeBot <noreply@commandcode.ai>
…h repository guides - Eliminate standalone data mappers in favor of model-owned methods - Extract backend error codes and failure mapping into data/error/ - Refactor MerchantOnboardingRepositoryImpl to use toEitherWithFallback - Remove pure pass-through LoadMerchantReferenceDataUseCase - Standardize remote DTO naming to Model - Strip explanatory comments across the feature
…presentation cubit - Decompose monolithic reference data into discrete entities under domain/entity/ - Decompose monolithic value objects into discrete single-purpose files under domain/value/ - Streamline MerchantOnboardingCubit adhering to forms-use-cubit pattern - Encapsulate wizard step routing and field path mappings in MerchantOnboardingStep extensions - Extract OwnerMoveDirection to presentation/models/ - Update widgets, DI, routing, and unit/widget/integration tests
- Record reference entity refactoring plan - Record domain types decomposition plan - Record BLoC conversion plan - Record clean Cubit refactoring plan
Remove effect streams from account cubits and manage snackbars and dialogs through BlocListener. Convert AccountPage, CompleteProfilePage, and RequestAccountDeletionPage to StatelessWidget.
…e and fix wizard navigation Remove effect streams from MerchantOnboardingCubit in favor of BlocListener on submission status and local failures. Fix Android system back to navigate wizard steps backwards rather than immediately exiting, handle empty validation failures safely without throwing StateError, and update unit and widget tests.
Add headerName and headers([String? key]) helper to IdempotencyKeyUtils to avoid repeating boilerplate header maps across remote datasources. Refactor account, profile, auth, and merchant onboarding remote datasources to use the helper, and add dedicated unit tests.
…d review presentation
Local emulator proof, not CI. Split plumbing vs login pilot because oracle IDs must exist at task begin. Park evidence-push so plumbing is the single active V2.
mobilekit runtime evidence runs maestro-flow oracles. Raw Maestro CLI and MCP stay iteration only. Local emulator proof. Not CI or Cloud.
dart:io File == is identity, so summary.md was tagged transient-local-log and handoff ignored passing Maestro runs.
Cold start hits onboarding then complete-profile before Home. Labels are not EditTexts; text selectors are regex. Never hideKeyboard on this emulator.
YAML runs no longer need --task. They attach flutter logs for the Maestro process. Drop the runtime logs command. Add register.yaml and close the login and register pilots.
Merchant onboarding, password-reset request, logout, and change password each get a maestro-flow oracle. Document the allowed command subset so agents stop using point taps, swipe, and greedy regex. Close the iterate CLI plan.
…edundant Value Objects Remove 9 redundant ID Value Object classes that performed external catalog lookups or wrapped client-side UUIDs. Store plain String identifiers directly in aggregates (BusinessProfile, OwnershipStructure, SettlementAccount, OwnerRow) and validate catalog constraints against MerchantReferenceDataEntity. Use UuidV4Utils.generate() for owner row IDs.
Keep field vs operation failure types. Stop parking them on the domain root next to aggregates and value objects.
One state stream. Builder paints. Listener handles snackbar and navigation. Drop the second cubit effect stream from the guide.
Kick reference load from BlocProvider. Submit failure is an effect instead of a build-time snackbar. Add SignInPageOfc as the BlocListener-shaped login reference.
Remove effect streams from auth cubits and use BlocListener with listenWhen for one-shot UI actions (snackbars, navigation). Convert pages to StatelessWidget and clean up prototype sign_in_page_ofc.
- Align scaffold page imports with design system barrels (adaptive, button, loading) - Adopt AppStateMessagePanel for scaffolded success and error states - Support incremental slice scaffolding for existing features - Prune empty directory generation to only create populated directories - Fix AppDotWave required color parameter and lowerCamelCase route list identifier - Update next-steps guidance to reference feature_modules_registrar.dart - Upgrade generated unit tests to use bloc_test - Add test coverage verifying incremental slicing, barrel imports, and directory pruning
Implement mobilekit scaffold data subcommand to generate Freezed DTOs, core endpoint constants, and remote datasource methods from local OpenAPI spec with targeted build_runner execution.
… data Recursively collect transitive sub-schemas and enums for Freezed DTO generation, support allOf schema references, and remove deprecated build_runner flag.
… in data scaffolding
CLI reference lagged the scaffold data and scaffold all commands already on this branch.
Coordinate swipes missed fields after layout shifts. Scroll to the field label, then tap below it.
flutter logs is logcat only, so Maestro evidence missed dart:developer lines. Forward the VM service port and subscribe to the Logging stream.
event intake would have pushed 84d10d3, which is no longer the branch head. No push and no draft PR were made.
A green Maestro exit could hide a missing VM log or a 4xx. evidence.json stores the signal ids only.
Core actionable duplication now fails verification. Small-helper and presentation reports stay advisory. Golden tests are a real oracle, and CI runtime reads those targets from the registry. Invented or local-only operating-evidence revisions are rejected. Engineering guides now live in subject folders.
Flutter 3.47.5 cannot format freezed 3 output, and freezed 4 needs analyzer 14 while custom_lint still pins analyzer 8. Flutter 3.44.9 makes IconData final, which phosphor_flutter 2.1.0 extends. The 3.41.9 clone failed, so the pin stays 3.41.4. Took the newest packages that resolve there, including secure storage 11 and go_router 17.5.0. Dio transform timeouts and permanently denied notification permission are now handled.
Current stable is 3.47.5 (Dart 3.13.4). phosphor_flutter 2.1.0 extends IconData, which is final on this SDK, so icons move to phosphor_icons 3.0.1. go_router stays on 17.5.0: 18.0.1 looks up material_ui's MaterialApp, and this SDK still ships flutter/material.dart, so 18 would drop platform transitions. freezed stays on 3.2.3 and the language floor stays 3.10.3 because freezed 4 needs analyzer 14 and custom_lint 0.8.1 still pins analyzer 8. The compact settings golden matches the 3.47 border raster. The merchant phone tap is bounded above Next so the input lands in the phone field.
Core actionable groups exit 1 inside verify --profile full and ci. small-helpers and presentation stay reports. The registry has no metric-assertion kind, so the oracle and runtime docs stop requiring one.
CI formats the CLI and lint packages at language 3.10. Those files had been formatted at 3.13 through a stale nested package config. Gradle 8.13 is below Flutter's 8.14 minimum, and Kotlin 2.1.0 is below the 2.2.20 minimum.
The emulator runner executes its script with /bin/sh. mapfile and process substitution are bash-only, so the Android lane exited before any integration test started.
android-emulator-runner runs each script line as its own sh -c, so a multi-line while loop never reached done.
fikrilal
marked this pull request as ready for review
September 28, 2026 12:58
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Evidence-bound harness completion:
handoff checkgates local acceptance oncurrent passing oracle evidence; coverage moves into
CI Required; CI Runtimeexecutes both registered integration targets on an Android 35 emulator;
verification re-checks candidate identity after the lane; maintenance detects
dirty-file content mutation with independent dep/codegen outcomes; three
focused regressions registered as oracles; evidence-collection workflow
documented. See
docs/exec-plans/completed/2026-09-07_harness-evidence-integrity-v4.md.Commits in this PR:
84d10d3feat(harness): evidence-bound completion, coverage CI, drift-safe verificationefe1c8edocs(exec-plans): retire committed v2 plan, begin push baselineRisk Class
lowmediumhighRisk notes:
risk_classifier.dart; new oracle kind consumers, CI aggregate change,verification-controller behavior change.
Acceptance Criteria
handoff check; drafts disclose it.CI Required; CI Runtime runs both integration targets.verified.harness/oracles.yaml.Checks Run
Hosted checks:
CI RiskpassedCI FullpassedCI CoveragepassedCI Runtimepassed or was intentionally skipped by classificationCI GovernancepassedCI Requiredaggregate passedEvidence
emulator runs above; hosted runtime pending on this PR
pending hosted run
Links/artifacts:
ee8a208b…(v4/commit baseline, full green x3),2084afb1…(clean push baseline, full green)handoff check: passed read-only for both baselinesRuntime evidence details (if applicable):
ee8a208b…auth.integration,startup.integrationpassed locally.tmp/harness-*-integration.log, ignored)Safety And Rollback
84d10d3via follow-up PR through normal review; never force-push.records through the documented workflow (human review + exact-revision hosted CI).
Reviewer Focus
packages/mobile_core_kit_cli/lib/src/handoff/completion_evidence.dart(newreader: latest-wins, tamper checks, manual attestation limits)
packages/mobile_core_kit_cli/lib/src/handoff/handoff_service.dart(check vspublication split; draft disclosure)
packages/mobile_core_kit_cli/lib/src/task/task_controller.dart(post-lanefingerprint re-check)
.github/workflows/required.yml(coverage job + runtime integration executionharness/oracles.yaml(3 new regression oracles)