Files
2026-07-25 01:40:08 +00:00

20 KiB

Hermes Mobile — Delivery Plan

Operational contract: Codex must read this file before each work loop. Work exactly one coherent milestone at a time. Move completed items to the completed section in the same commit that implements them. Add a concise entry to CHANGELOG.md in the same commit. Never mark work complete without running its listed verification.

Product guardrails

  • Native Android client for upstream Hermes Agent: Kotlin + Jetpack Compose.
  • Material 3 base, deliberately customized through a reusable design system.
  • Thin client: Hermes remains the authority for auth, sessions, streaming, tool semantics, and backend behavior.
  • No WebView-first app, no cloned Hermes backend, no embedded credentials.
  • Adaptive, accessible, compact-but-readable UI: phone, tablet, light/dark, font scaling, IME/keyboard, long code/output must remain safe.
  • Keep DTO/transport, domain/UI models, repositories, view models/state, and composables separated.
  • Never commit APKs, build outputs, credentials, or generated local environment files.

Release policy

  • Start artifacts at v0.1.0-beta.1.
  • Patch: compatible fixes/polish. Minor: a coherent feature milestone. Major: stable v1.0.0+ only.
  • A release requires a pushed source commit, passing relevant checks, a versioned APK artifact, and a Telegram release post.

Now — direct gateway pivot

Complete the Android/client contract pivot before implementing backend behavior. The next release path targets a remote upstream Hermes Agent gateway/API server directly; apps/companion is legacy compatibility only.

  • R2 — Verify a clean v0.1.0-beta.1 install against a direct gateway
    • On one named real device or emulator, build the source-current versioned debug APK, confirm package metadata, uninstall prior app data, install and launch it, configure an upstream Hermes gateway directly, and record the connection result plus APK SHA-256.
    • Verification: required preflight and Android checks pass in the install environment; device/install evidence is recorded in docs/ROADMAP.md; no release or final beta criterion is claimed from source-only checks.

Externally blocked

  • R1 — Execute and document the adaptive/accessibility device matrix
    • Source safeguards and 61 cached source-current unit tests were verified on 2026-07-24, but small-phone, normal-phone, tablet, TalkBack, contrast, motion, and IME evidence still requires a permitted real device or emulator.
    • Unblock evidence is maintained in docs/BLOCKERS.md; do not mark R1 complete from source review alone.
  • B3 — Direct gateway session contract and durable chat state
    • Foundation verified 2026-07-24: repository search contract and fake implementation, capability-gated session list/create/get/delete mapping, durable selected-session/draft/pending-send storage, and duplicate-send reducer/UI protection.
    • Not complete: official upstream documentation still provides no session-search HTTP contract, and the configured local API server was not listening for a live compatibility exercise on 2026-07-24.
    • Unblock evidence is maintained in docs/BLOCKERS.md; do not mark B3 complete until documented search and real gateway behavior are verified.

V2 backlog

  • Multiple server profiles and fast switching.
  • Rich file/artifact workflows based on upstream capability.
  • Notifications for task completion.
  • Mature tablet/two-pane session + optional terminal/log pane.
  • Offline cache/search and diagnostic export with privacy controls.
  • Remove legacy companion Android routes and deprecate apps/companion install/runtime docs after direct gateway chat/session/streaming is verified.

Completed

Move finished items here with date, commit, and verification. Keep this section factual; do not claim unverified work.

  • P0 — First-run Setup Wizard — 2026-07-24, commits 5b3263e (feat: add first-run setup wizard) and 7780760 (fix: prevent setup restart bypass).

    • Added a Compose first-run flow covering welcome/privacy guidance, required Remote HTTPS and optional Local HTTPS routes, None/Bearer/username-password authentication, local-first connection testing, and completion only after a successful test.
    • Added direct password login through /auth/password-login, allowlisted session-cookie extraction, encrypted cookie/password storage, credential redaction, replacement/retention rules, and cookie-authenticated gateway probes without persisting raw credentials in UI state or connection forms.
    • Added Settings entry for editing the same profile, skip behavior for existing valid configurations, one-time legacy/current-profile setup-version migration, and accessible adaptive Compose semantics plus phone/tablet/light/dark previews.
    • Fixed a restart bypass: a profile saved by a failed first-run connection test remains in mandatory setup until setup completion is durably recorded; valid migrated and completed profiles still skip setup.
    • Repaired the setup gate so only valid profiles with recorded legacy provenance receive the one-time completion migration; untested first-run None and Bearer profiles remain mandatory after restart.
    • Follow-up regression repair on 2026-07-25: removed bearer-token-only self-migration, which could classify a failed/offline first-run Bearer profile as a historical profile. Added a focused restart regression covering that state.
    • Verification for the follow-up repair: offline focused SetupWizardReducerTest plus :app:assembleDebug passed (43 tasks, 9 executed); git diff --check passed.
    • Verification for the migration repair: source-current offline focused SetupWizardReducerTest plus :app:assembleDebug passed (43 tasks, 9 executed); git diff --check passed.
    • Verification on 2026-07-24: source-current offline :app:testDebugUnitTest :app:assembleDebug passed (43 tasks, 4 executed; all unit tests passed), including the focused restart-regression test; git diff --check passed. No device, release, or real gateway credential test is claimed.
    • Task 5 source-current re-verification on 2026-07-24: /root/.openclaw/workspace/scripts/hermes-mobile-preflight.sh passed; the documented offline :app:testDebugUnitTest :app:assembleDebug command passed (43 tasks, 1 executed); the generated Gradle XML report records 106 tests with 0 failures/errors, including 29 SetupWizardReducerTest cases; and git diff --check passed. A localhost gateway health probe returned HTTP 200, establishing reachability only. The debug APK reports package cloud.molberg.hermesmobile, version 0.1.0-beta.1 (10001), SHA-256 45e4ab4783a74eb766070b0f670a2631a9ebdc044468bbe03087788f8106f150. No ADB device/emulator or /dev/kvm was available; no device installation, real authentication, release, or end-to-end gateway verification is claimed.
    • Later Task 5 current-sandbox rerun on 2026-07-24: the exact docs/ANDROID_BETA_BUILD.md Gradle command was blocked before project execution with Could not determine a usable wildcard IP, and the recorded offline/no-daemon variant was blocked while opening Gradle's daemon TCP socket with java.net.SocketException: Operation not permitted. Direct JUnit execution of the existing source-current compiled SetupWizardReducerTest classes passed all 29 tests. The existing source-current XML reports remain 106 tests with 0 failures/errors, and the existing ignored APK was unchanged by both Gradle attempts; aapt and SHA-256 inspection reconfirmed the package/version/hash above. ADB 37.0.0 was available only by explicit path and could not start its daemon because smart-socket/netlink sockets are prohibited; /dev/kvm and /dev/bus/usb were absent. This adds fresh focused reducer execution only: no fresh Gradle compilation/test task, APK assembly, device installation, credential flow, release, or end-to-end verification is claimed.
    • Task 5 delivery verification on 2026-07-25: /root/.openclaw/workspace/scripts/hermes-mobile-preflight.sh passed; from apps/mobile/android, GRADLE_USER_HOME=/root/hermes-mobile/.gradle-user ./gradlew --offline --no-daemon -Dorg.gradle.jvmargs= -Dorg.gradle.daemon=false -Pkotlin.compiler.execution.strategy=in-process :app:testDebugUnitTest :app:assembleDebug passed (43 tasks, 1 executed); git diff --check passed. Current Gradle XML reports record 107 tests with 0 failures/errors/skips. The ignored debug APK reports package cloud.molberg.hermesmobile, version 0.1.0-beta.1 (10001), size 17,710,960 bytes, and SHA-256 714e8e9cdd3f2eed93b285f233b1f4573d5d13fc2078bb717631f9b38f8428d4. ADB reports no attached devices; no emulator/AVD or /dev/kvm is available. No device installation, release, or gateway end-to-end verification is claimed.
  • R2a — Establish beta source packaging and install documentation — 2026-07-24, commit 3b883bd (build: establish Android beta packaging); guide corrected 2026-07-25.

    • Set Android package metadata to semantic versionName 0.1.0-beta.1 with monotonic versionCode 10001, enabled generated metadata coverage, and named the debug APK hermes-mobile-v0.1.0-beta.1-debug.apk.
    • Added a direct-gateway-only build, metadata inspection, clean-install, launch, configuration, evidence, and known-limitations guide; corrected it to document mandatory clean-install Setup Wizard entry, all three auth modes, successful-test completion gating, and post-setup Settings editing. Changed the npm Android build command to run the native Gradle project without Capacitor synchronization.
    • Source verification on 2026-07-24: required recovery preflight passed; :app:testDebugUnitTest :app:assembleDebug passed with the cached offline Gradle setup (43 tasks, 5 executed); and git diff --check passed. No APK was published, uploaded, or installed, so the remaining R2 device task and all final release criteria remain unclaimed.
  • B5 — Tool, artifact, and output experience — 2026-07-24, commit feat: add gated tool output experience.

    • Added capability-gated decoding and reduction for the existing tool.started, tool.progress, tool.completed, and tool.failed direct gateway SSE events, with compact lifecycle/status cards for referenced and stream-only tools.
    • Added full-output copy plus deterministic collapsed/expanded display windows capped at 5/80 lines and 24,000 expanded characters, avoiding same-direction nested scrolling while retaining the complete output in state.
    • Persisted expanded tool IDs in durable chat state so rotation/process resume restores presentation state; artifact share/save remains unavailable because the direct gateway advertises no verified artifact handoff URI/download capability.
    • Verification: the recovery preflight passed before this run; cached Kotlin 2.0.21 compilation passed for all Android Kotlin/Compose main and unit-test sources; direct JUnit passed 57 tests including large-output bounds, capability gating, lifecycle merging, and rotation/resume restoration; git diff --check passed. A direct Gradle :app:testDebugUnitTest :app:assembleDebug retry was blocked before project execution because the sandbox prohibits Gradle's TCP control socket (java.net.SocketException: Operation not permitted).
  • B4c — Wire gateway SSE into durable chat sends — 2026-07-24, commit feat: wire durable gateway SSE sends.

    • Added a direct-gateway send coordinator that captures the selected session, draft, route lease, and capability-built SSE request before asynchronous execution; it never reselects a route or switches an active stream.
    • Publishes only the currently reserved request's intermediate and terminal reducer states; terminal completion, gateway failure, and transport failure clear only the matching durable reservation while retaining the selected session and draft.
    • Verification: /root/.openclaw/workspace/scripts/hermes-mobile-preflight.sh passed (Android :app:assembleDebug, 35 tasks, 4 executed); direct cached Kotlin/JUnit verification passed 52 tests including deterministic B4c success, gateway failure, transport failure, and stale request-ID tests; git diff --check passed.
  • B4b — Execute session SSE through the Android repository — 2026-07-24, commit feat: execute gateway session SSE.

    • Added a thin OkHttp repository that executes the capability-built session stream request and parses SSE frames into the existing decoder and lease-bound reducer.
    • Kept the route generation, request URL, session ID, endpoint, and event gate immutable for the full call; transport/protocol failure invalidates and fails only that original lease.
    • Kept gateway error events separate from transport failure and excluded reconnect, cancellation, tool/artifact handling, and UI integration.
    • Verification: focused cached Kotlin compilation and direct JUnit execution passed 15 reducer, stream-contract, and fake HTTP/SSE repository tests; Android Gradle build attempt and final checks are recorded in CHANGELOG.md.
  • B4a — Capability-gated direct streaming foundation — 2026-07-24, commit feat: add gateway streaming foundation.

    • Added capability-derived session SSE request construction using the advertised POST method and {session_id} path only.
    • Bound each stream to an immutable route generation and session ID, decoded documented core lifecycle/delta/final events, and gated structured tool events on tool_progress_events.
    • Made final assistant responses authoritative over accumulated deltas and ignored events from a different session lease.
    • Verification: focused cached Kotlin compilation plus direct JUnit execution passed 12 reducer/stream contract tests; full checks recorded in CHANGELOG.md.
  • B2 — Direct gateway connection, authentication, and local-first routing UX — 2026-07-24, commit e5cc27e (feat: add local-first gateway routing).

    • Added one logical profile with optional Local HTTPS, required Remote HTTPS, Android-keystore-protected bearer/no-auth configuration, and active Local/Remote diagnostics.
    • Added bounded redirect-disabled GET /health probes followed by bearer-authenticated GET /v1/models, deterministic Local preference/Remote fallback, and immutable request leases.
    • Added foreground, explicit reconnect, connectivity-change, and request-failure re-evaluation hooks; removed Android calls to undocumented legacy companion /api/* routes without changing apps/companion.
    • Recorded the upstream stable gateway identity-binding need in docs/BLOCKERS.md.
    • Verification: direct Kotlin/JUnit execution of ConnectionStateTest and GatewayHttpProbeTest passed 12 tests; GRADLE_USER_HOME=/root/hermes-mobile/.gradle-user ./gradlew --no-daemon -Pkotlin.compiler.execution.strategy=in-process :app:compileDebugUnitTestKotlin :app:assembleDebug passed from apps/mobile/android with temporary restored sandbox-only Gradle socket shims (37 tasks, 6 executed); git diff --check passed.
    • Unhappy paths verified deterministically: missing/cleartext/credential-bearing/path-bearing/duplicate URLs, missing bearer token, incompatible or offline Local fallback, Remote offline, unauthorized compatibility response, stored legacy HTTP rejection, and failed-route lease invalidation. Device UI walkthrough was attempted but ADB could not start because this sandbox prohibits local control sockets.
  • F0 — Audit and safely land existing work — 2026-07-24, commit da79d48 (feat: land native mobile foundation).

    • Reviewed existing Android and companion edits; preserved the native Android chat/workspace/settings work and companion chat hardening.
    • Confirmed generated build outputs are ignored and added ignores for .gradle-user/ and .dev/.
    • Reconciled docs/ROADMAP.md to the actual native Android beta state and remaining validation gaps.
    • Verification: git diff --check passed; npm run typecheck --workspace @hermes-mobile/companion passed; npm run lint --workspace @hermes-mobile/companion passed; npm run build --workspace @hermes-mobile/companion passed; npm run typecheck passed; npm run lint passed; npm run build passed.
    • Android debug build attempted with GRADLE_USER_HOME=/root/hermes-mobile/.gradle-user ./gradlew :app:assembleDebug from apps/mobile/android; blocked before compilation because network socket creation is not permitted while downloading Gradle 8.11.1.
  • F1 — Compose design system extraction — 2026-07-24, commit 9ed0969 (feat: extract compose design system).

    • Extracted color, semantic status, spacing, typography, shape, elevation, and motion tokens.
    • Added light and dark Hermes Compose themes backed by Material 3 color schemes.
    • Split shared Compose cards, rows, controls, notices, message bubbles, list scaffolds, and code surfaces out of MainActivity.kt.
    • Added Compose previews for small phone, normal phone, tablet width, light theme, and dark theme component coverage.
    • Verification: git diff --check passed; GRADLE_USER_HOME=/root/hermes-mobile/.gradle-user ./gradlew :app:assembleDebug from apps/mobile/android passed.
  • F2 — Polished fake-data chat proving ground — 2026-07-24, commit f65d4ac (feat: add fake chat proving ground).

    • Added a local fake chat inbox with session rail/list, context controls, keyboard-safe composer, and send/cancel/retry interactions.
    • Added typed user, assistant, system, error, and streaming message models; selectable markdown-friendly text, horizontally scrollable copyable code, and expandable tool results.
    • Added seeded empty, loading, reconnecting, and error states plus reducer tests for send, cancel, retry, and tool expansion.
    • Verification: git diff --check passed; GRADLE_USER_HOME=/root/hermes-mobile/.gradle-user ./gradlew :app:testDebugUnitTest :app:assembleDebug from apps/mobile/android passed.
  • B1 — Transport/domain contracts — 2026-07-24, commit feat: add transport domain contracts.

    • Added transport DTOs and domain repository contracts for server profiles, authentication, sessions, messages, stream events, tool results, and artifacts.
    • Added DTO-to-domain/UI mappers, stream reducer behavior, and a fake Hermes repository for previews and unit tests.
    • Kept the companion path as an optional adapter capability without changing upstream Hermes semantics.
    • Verification: parent preflight immediately before B1 completion passed: script PASS; Android :app:assembleDebug BUILD SUCCESSFUL, 35 tasks total and 6 executed. Parent also verified git diff --check passed.
  • B2a — Direct gateway planning and compatibility contract — 2026-07-24, commit docs: pivot mobile contract to direct gateway.

    • Added docs/DIRECT_GATEWAY_ARCHITECTURE.md with documented upstream endpoints separated from unverified assumptions.
    • Reprioritized the beta path around direct gateway auth, sessions, streaming, and later companion removal.
    • Updated Android transport/domain/fake capability contracts and connection copy so new code no longer assumes a companion server.
    • Verification: git diff --check passed; Android :app:testDebugUnitTest :app:assembleDebug passed from apps/mobile/android.

Work-loop checklist

Before coding:

  1. Read PROJECT_PLAN.md, CHANGELOG.md, docs/ROADMAP.md, git status, and the prior commit.
  2. Choose the single highest-value unchecked item or a narrowly scoped sub-item.
  3. Inspect existing implementation before rewriting it.

Before committing:

  1. Run relevant checks; record exact pass/fail scope in the changelog.
  2. Update this plan: check the completed item, move it into Completed, and leave the next executable item unchecked.
  3. Add a concise CHANGELOG.md entry under Unreleased.
  4. Commit using Conventional Commits and push origin/main.
  5. Do not claim a release without an uploaded, verified versioned APK.