Hermes parallel implementation backlog for production readiness

[ucguy4u]

Problem

Airo has multiple planned production-readiness workstreams spread across gstack context, superpowers plans, and repo docs. The current thread produced a Hermes delegation map, but the remaining work needs to be tracked on the DevelopersCoffee project board so any developer or agent can pick up a workstream independently and drive it to production.

Source artifact in repo:

• docs/operations/HERMES_URGENT_DELEGATION.md

Important repo state when created:

• Main scanned branch: codex/coins-board-cleanup
• Main worktree had uncommitted changes under app/lib/features/agent_chat/..., app/pubspec.yaml, app/pubspec.lock, and related tests.
• Developers should create separate worktrees from HEAD and avoid implementing inside the dirty main checkout.

Scope

Create and execute independent Hermes workstreams from the delegation map until the selected production slices are implemented, tested, documented, reviewed, and merged.

In Scope

• Use separate git worktree directories and branches for each parallel workstream.
• Implement the highest-priority slices listed below.
• Update source planning docs only for work actually completed.
• Add tests and run focused verification for every workstream.
• Open PRs with linked issues/cards and production risk notes.
• Split this issue into child issues if a workstream exceeds one week or needs a dedicated owner.

Out of Scope

• Implementing remote/community skill import before runtime verification exists.
• Adding GitHub secrets manually from automation.
• Shipping arbitrary connector execution directly to the LLM.
• Working in the dirty main checkout instead of isolated worktrees.

Primary Workstreams

1. Agent Skills Runtime and Calendar Connectors

Source: docs/superpowers/plans/2026-06-20-airo-agent-skills-connectors.md

Recommended branch/worktree:

• Branch: codex/hermes-agent-skills
• Worktree: ../airo-hermes-agent-skills

Deliverables:

• AgentSkill domain model and manifest parser.
• Built-in skill repository with enable/disable/search/summary behavior.
• AgentConnector interface and registry.
• Date/time connector and fake calendar connector for tests.
• Bounded skill orchestrator skeleton with max-step/tool validation.
• AgentActionTrace model and basic UI trace hooks if low risk.

Acceptance criteria:

• Valid and invalid SKILL.md parser tests pass.
• Built-in skill repository tests cover filtering/search/enable/disable.
• Connector registry tests cover dispatch and validation.
• Orchestrator tests cover no-skill, one-tool, denied permission, malformed JSON, max-step.
• Existing deterministic IntentParser and ToolRegistry behavior remains intact.
• Focused command runs: cd app && flutter test test/features/agent_chat.

2. Coins Transaction Review Loop

Source: docs/features/coins/PERSONAL_MONEY_MANAGEMENT_DELTA_TODO.md

Recommended branch/worktree:

• Branch: codex/hermes-coins-review-loop
• Worktree: ../airo-hermes-coins-review-loop

Deliverables:

• raw_financial_records and financial_sources local structures/tables.
• Source hash, parser name, parser version, confidence, and raw record reference tracking.
• Duplicate paste prevention for chat/finance ingestion.
• Review queue state for uncertain parsed transactions.
• Edit/undo support for chat-ingested transactions where current flow allows.

Acceptance criteria:

• Duplicate ingestion tests pass.
• Review queue tests pass.
• Migration/data-safety notes are documented.
• Focused command runs: cd app && flutter test test/features/coins test/features/money where those paths exist.

3. Media Hub Sprint 1 Foundation

Source: docs/features/media-hub/ENGINEERING_TICKETS.md

Recommended branch/worktree:

• Branch: codex/hermes-media-hub-foundation
• Worktree: ../airo-hermes-media-hub-foundation

Deliverables:

• MH-001 collapsible hero/player container.
• MH-002 player overlay controls.
• MH-003 settings bottom sheet and quality settings state.
• MH-004 media content card/grid/carousel.
• MH-005 unified media content/domain models.

Acceptance criteria:

• New widgets/providers have focused tests where practical.
• Player collapse, overlay, settings, cards, and unified model compile cleanly.
• cd app && flutter analyze is clean or blockers are documented.

4. Modular Platform Builds and APK Size Quick Wins

Source: docs/architecture/MODULAR_SUPER_APP_SPRINT_PLAN.md

Recommended branch/worktree:

• Branch: codex/hermes-platform-builds
• Worktree: ../airo-hermes-platform-builds

Deliverables:

• Platform build scripts verified/repaired in melos.yaml.
• Platform config for AppPlatform, AppFeature, and enabled feature checks.
• Platform entrypoints: main_tv.dart, main_mobile_streaming.dart, main_mobile_full.dart.
• TV app shell and Android TV manifest/flavor if compatible with current Gradle setup.
• APK size commands and split-per-ABI/tree-shake consistency.

Acceptance criteria:

• melos run analyze passes.
• melos run build:tv passes or exact Android setup blocker is documented.
• APK sizes are recorded after successful builds.

5. Plugin Download-On-Demand Infrastructure

Source: issue-download-on-demand.md

Recommended branch/worktree:

• Branch: codex/hermes-plugin-downloads
• Worktree: ../airo-hermes-plugin-downloads

Deliverables:

• PluginDownloadManager under packages/core_data/lib/src/plugins/.
• PluginStorageManager.
• PluginIntegrityChecker.
• Progress stream with speed/ETA, retry/backoff, cancel/pause/resume states.
• LRU storage metadata, cleanup APIs, integrity verification, rollback metadata.

Acceptance criteria:

• Unit tests pass under packages/core_data/test.
• cd packages/core_data && flutter test passes.
• melos run analyze passes.

6. Locale Onboarding and Package Docs Cleanup

Source: docs/features/LOCALE_ONBOARDING_REQUIREMENTS.md

Recommended branch/worktree:

• Branch: codex/hermes-locale-docs
• Worktree: ../airo-hermes-locale-docs

Deliverables:

• Currency selection in onboarding or settings if onboarding is unavailable.
• Persisted selected currency using existing preference/local settings pattern.
• Money screens reflect selected currency.
• Core package README TODOs and package changelog TODOs replaced with accurate content.

Acceptance criteria:

• Locale/provider tests pass.
• Currency selection widget tests pass where practical.
• Docs no longer contain template TODOs for core package READMEs/changelogs touched.

7. CI/CD and Wiki Documentation Gates

Source: docs/superpowers/plans/2026-06-20-airo-wiki-user-guide.md, docs/setup/CI_CD_CHECKLIST.md

Recommended branch/worktree:

• Branch: codex/hermes-docs-ci-gates
• Worktree: ../airo-hermes-docs-ci-gates

Deliverables:

• Script/check that detects PRs touching app/lib/features, app/lib/core/routing, or README.md without touching docs/wiki or declaring docs-not-needed.
• GitHub Actions wiring or documented local command if project label context is unavailable.
• CI checklist updated with current workflow status.

Acceptance criteria:

• New docs gate script is tested with simulated changed-file lists.
• Workflow syntax is valid.
• Manual secret steps remain documented, not automated.

Global Acceptance Criteria

• Each workstream uses its own git worktree and branch.
• Each workstream has a focused PR linked back to this issue or a child issue.
• Each PR lists tests run, risks, rollout notes, and docs updates.
• Planning docs are updated only for completed work.
• No generated artifacts, secrets, or local environment files are staged.
• Production blockers are converted into follow-up issues, not left only in PR comments.

Suggested Labels For Child Issues

• Agent skills: agent/ai-llm, type/task, priority/P0
• Coins review loop: agent/finance, type/task, priority/P0
• Media Hub: agent/mobile-ui, media-hub, sprint-1, type/task, priority/P1
• Platform builds: agent/core-architecture, agent/performance, epic/apk-optimization, type/task, priority/P0
• Plugin downloads: agent/plugin-system, epic/plugin-architecture, type/task, priority/P1
• Locale/docs: agent/mobile-ui, agent/docs, type/task, priority/P2
• CI/docs gate: agent/ci-cd, agent/docs, type/task, priority/P1

Risks

• Dirty main worktree can cause accidental conflict or untracked changes if developers do not use worktrees.
• Agent Skills touches files currently dirty in the scanned worktree, so PR conflict risk is high.
• Calendar connectors need mobile permissions and native plugin validation before production.
• Coins finance ingestion has privacy/data minimization risk; raw financial content must stay local by default.
• Platform build work can expose native Android/iOS setup blockers.
• CI/CD secret setup requires a human with GitHub repository secret access.

References

• Delegation artifact: docs/operations/HERMES_URGENT_DELEGATION.md
• Agent skills v2 plan: docs/superpowers/plans/2026-06-20-airo-agent-skills-connectors.md
• Coins delta: docs/features/coins/PERSONAL_MONEY_MANAGEMENT_DELTA_TODO.md
• Media hub tickets: docs/features/media-hub/ENGINEERING_TICKETS.md
• Modular platform plan: docs/architecture/MODULAR_SUPER_APP_SPRINT_PLAN.md
• Plugin download issue spec: issue-download-on-demand.md
• Locale onboarding: docs/features/LOCALE_ONBOARDING_REQUIREMENTS.md
• Wiki guide plan: docs/superpowers/plans/2026-06-20-airo-wiki-user-guide.md

[ucguy4u]

Closing as superseded by dedicated execution issues and merged slices.

This ticket was useful as the initial Hermes delegation umbrella, but it no longer represents one independently shippable slice. Its workstreams are now tracked directly on the board as concrete issues, for example:

• Agent Skills release gate: Productionize Agent Skills simulator journey and release gate #252
• Smart reminders hardening: Production harden smart reminders and medicine reminder framework #254
• Coins OCR production hardening: Production hardening: local PDF receipt OCR for Coins itemized split #257
• Brain AI chat revamp: Revamp Brain AI Chat to match Google AI Edge Gallery chat UX #260
• On-device model stack/runtime work: Integrate Google AI Edge, LiteRT, and Hugging Face model stack #277, Add runtime context management, compaction, and device-aware LiteRT settings #289, Add device compatibility diagnostics and crash-safe model initialization #310, Add Google AI Edge Gallery Agent Skill compatibility layer #313, feat(ai-llm): Integrate LiteRT-LM (MediaPipe LLM Inference API) for Cross-Platform On-Device LLM Support #353, feat(ai-llm): On-Device Vector Search and Local RAG using MediaPipe Text Embedder #355, feat(ai-llm): Local Voice Command Classification via MediaPipe Audio Tasks #358, feat(devex): Model Optimization Pipeline via Google AI Edge Quantizer #359
• LifeTrack breakdown: [AGENT] LT-001: LifeTrack Domain Entities — LifeTrack, Milestone, ActionItem, InputRequirement #332-[AGENT] LT-012: LifeTrack Memory Vault Backup & Searchable History #352
• Drift follow-up: Optional: upgrade Drift lockfile entries to 2.34.0 #364

Some delegation outcomes have also already landed as focused PRs instead of staying in this umbrella, including the wiki/docs gate work merged in #370 and other low-hanging issue cleanups already merged this sweep.

Keeping the umbrella open now duplicates board state and makes prioritization noisier, so I'm retiring it in favor of the concrete child/workstream tickets above.