app.personalos-mobile

PersonalOS Mobile

Premium Expo Router shell for the PersonalOS golden journey on iOS and Android. It composes @lssm-tech/module.personal-experience with the goals, productivity, household, and health module view-model seams.

The mounted shell now loads the authenticated subject-bound experience and workspace projections from the canonical PersonalOS API. Guided native forms build schema-validated, domain-separated commands for the complete golden journey instead of presenting inert affordances. Server denial and revision conflicts fail closed; accepted commands refresh the authoritative projection. The shell consumes the integrity-bound presentation member of that projection for exact area, component, and <area>.manage boundaries. Missing or malformed presentation metadata, plus resolutions older than five minutes, are discarded so safe display-only state remains usable, while every governed action fails closed until a current bound presentation permits it. Presentation preferences change only density, detail, guidance, control explanation, media feedback, pace, and narrative display; they cannot grant authority and can be reset or restored without clearing form drafts. Household invitation, activation, delegation, and revocation use the shared runtime draft constructor before Expo Crypto performs the native domain-separated SHA-256 binding. Mobile does not redefine their risk, authority, offline eligibility, or excluded-action policy. The embedding native host registers its AuthOS login/deep-link exchange through registerPersonalOsNativeAuthBootstrap; the returned opaque session enters the device-only vault before projection loading. A clean standalone install without that host seam stays signed out and fails closed—no development token or public environment secret is accepted. For isolated simulator qualification, Metro resolves an ephemeral AuthOS bootstrap only when PERSONAL_OS_MOBILE_E2E_AUTH=enabled; the bootstrap exchanges a run-specific nonce with the loopback API host. Normal production exports resolve a no-op module instead, and the export command scans both Hermes bundles for the route, header, token, and E2E error sentinels before succeeding.

The native runtime is pinned to the Expo 57-compatible React Native 0.86 line. SecureStore identifiers use only Expo-supported letters, numbers, periods, hyphens, and underscores, so both opaque sessions and SQLCipher keys survive the real native boundary instead of failing only after installation.

Product surfaces

• Today: capacity-aware plan, check-in, and missed-day recovery.
• Goals: personal mission, observable goals, minimized outcome measurements, and online-authorized record/correct/remove controls that preserve history.
• Health: accessible, non-clinical category history with minimized status, observation/quarantine counts, and latest source timestamp plus native HealthKit/Health Connect host ports; raw measurements and provider payloads never enter the view contract. A separate native summary renders the optional exact-subject wellbeing-goal window with localized safe metric, lifecycle, direction, observation count, and uncertainty-aware outcome plus explicit non-causal/non-medical copy.
• Household: membership plus explicit, expiring, revocable support, including one-goal proposal-only grants and immediate revoke, block, and categorical report controls.
• Privacy: minimized purpose/downstream-consumer/expiry/status consent dashboard, fixed one-year PersonalOS planning grant, active-grant revocation, coach explanation/undo, and online-only rights operations. Consent controls use the shared lifecycle schema and never enter the offline outbox.
• Review: progress and wellbeing reflection without medical or causal claims.

The shared Personal Calm design pack drives comfortable density, calm motion, platform-native navigation, progressive disclosure, 44-point touch targets, and light/dark readiness. Copy is typed in English, French, and Spanish; layout direction is derived independently so RTL locales can be added without restructuring screens.

Trust boundaries

• Session tokens use device-only expo-secure-store custody.
• The last authenticated subject marker uses the same device-only custody. An offline process relaunch can load only that subject's exact encrypted experience/workspace pair; foreign subjects, mismatched revisions, missing rows, and malformed projections fail closed.
• Local projections use SQLCipher. A 256-bit random key is generated by expo-crypto, stored in device-only secure storage, and applied before schema access.
• Projection payload policy rejects tokens, credentials, secrets, and raw provider payload keys.
• Only commands in the shared exact-action registry can enter the encrypted offline outbox. Caller-provided risk or undo metadata is never authoritative.
• Présentation allowed, conditional, approval, blocked, and hidden boundaries are enforced before canonical command construction and again at the authenticated session boundary. Preferences never widen those boundaries.
• Conditional and approval-required actions remain disabled until an exact condition or approval artifact and submission route exist; they cannot fall through to direct governed-command dispatch or offline queuing.
• Missing or invalid presentation metadata leaves generic display state usable but blocks all governed actions. Sensitive and high-risk actions always require current online authority, regardless of presentation preferences.
• Form drafts live above pending, denial, and conflict screen transitions. Hidden or blocked actions suppress their controls without deleting drafts, and reduced-motion mode removes modal slide animation. Draft storage is bound to the exact subject and actor refs and clears synchronously on sign-out or identity change, preventing a person's or delegate's draft from appearing in another authority context.
• The mounted session also observes the server-issued opaque tenant/consent binding. Exact subject, actor, tenant, and consent changes discard retiring Personalization buffers before the next scope can collect; transient checking, denial, offline, and error states preserve the current scope. Buffer attachment is unavailable without granted consent.
• The outbox validates the complete PersonalOS command envelope, authenticated subject, expected revision, canonical payload SHA-256 digest, and registry-owned undo command before its SQLCipher write.
• Canonical command binding uses the @lssm-tech/lib.personal-os-runtime/digest-port lane (runtime domain map and canonical JSON) with an expo-crypto SHA-256 adapter, so every shared workspace domain is mapped exactly as on the server, including private progress_celebration and high-risk guardrails domains. Both remain online-only and are not exposed as offline or silently authoritative actions.
• Replay runs the owner @lssm-tech/module.personal-experience/offline-replay state machine (25 commands per pass, 5 attempts, backoff, policy re-check). It reconstructs and validates the exact workspace command from the trusted registry, authenticated subject, revision, envelope, idempotency key, and domain-separated payload digest, then submits it to the canonical /v1/domain-commands authority boundary. Server denial and revision conflict remain blocked for recovery; transient failures use bounded retry and preserve order.
• Concurrent reconnect/session refresh triggers share one in-flight load and one replay pass, preventing duplicate transport attempts before server idempotency is needed.
• Version-one outbox rows are retained but marked legacy_envelope_missing because they cannot be proven safe to replay.
• Consent, sharing, native health connections, delegation, authority, export, and deletion revalidate online.
• HealthKit and Health Connect are real host ports, not mocked production adapters. Unavailable or denied native authorization fails closed.

Commandes

bun run lint:check
bun run typecheck
bun run test
bun run build
bun run ios
bun run android

Maestro definitions live in tests-mobile/flows. On 2026-08-13, bounded flows passed on an iPhone 16 Pro simulator running iOS 18.5: clean-install AuthOS denial; an ephemeral loopback AuthOS exchange followed by authenticated navigation across all six tabs and every canonical golden-journey action entry; the authenticated missed-day recovery form; and a source-bound native mutation run that durably created the household, mission, and first active goal through the production-composition API. These runs qualify native startup, navigation, guided-form reachability, and that exact revision-three mutation foundation. An additional iPhone 16 Pro simulator sequence established an authoritative workspace, disconnected the app from the preserved authority server, relaunched the app from an encrypted subject-bound projection pair, queued one reversible check-in, restored connectivity to that same server, and committed exactly one replay at expected revision 2. The server advanced to revision 3 with one minimized check-in and zero pending low-risk commands. These runs do not prove the remaining mutation journey, broader denial/recovery, Android, assistive technology, a real device, production AuthOS, provider accounts, store signing, or a canary and must not be counted as live_verified evidence.

The earlier navigation and recovery replay remains a historical artifact, but later mobile source changes invalidate its current source binding. The current matrix therefore credits only the newer source-bound mutation run until those broader flows are rerun on the latest source commit.

ContractSpec Connect was unavailable to this lane. The release capsule records the resulting review gap; it is not treated as approval evidence. The remaining native mutation and broader denial/recovery journeys, Android simulator, VoiceOver, TalkBack, real-device, provider-account, store-signing, deployed-host, and canary evidence remain pending and are not inferred from local tests or native bundle exports.

Maintainers can replay a current-source iOS simulator flow with the manually dispatched PersonalOS Native Qualification GitHub Actions workflow. The job creates an isolated simulator, builds the current Expo development client, starts the real PersonalOS E2E API and Metro hosts with a run-specific masked nonce, runs the selected Maestro flow, validates the authoritative workspace, and retains JUnit plus minimized test-state diagnostics for 14 days. A workflow run is qualification evidence only after its exact commit, result, artifact digest, simulator/runtime, and observed time are recorded in the PersonalOS source-bound qualification ledger.