JiboExperiments/OpenJibo/docs/release-1.0.19-plan.md

Release 1.0.19 Plan

Purpose

This release starts the shift from 1.0.18 hardening to visible feature growth.

The goal is to keep compatibility work steady while shipping personality and capability slices that make OpenJibo feel less like a placeholder cloud and more like a real assistant platform.

Snapshot

• Kickoff date: 2026-05-05
• Cloud version source of truth: OpenJiboCloudBuildInfo.cs
• Active release constant: 1.0.19

Scope

1. Persona And Identity Surface

• add natural voice responses for robot identity/personality prompts
• start building reusable content hooks for question-vs-command style responses
• keep first implementation rule-based and test-backed

2. Reliability And Device Proof

• complete update/backup/restore proof path with captures and operator docs
• continue alarm/gallery/yes-no cleanup from 1.0.18 evidence where regressions are still open
• improve short-turn STT reliability and low-signal screening

3. Pegasus-To-Cloud Platform Porting

• prioritize small source-backed slices from Pegasus/JiboOS that can be shipped safely
• keep Nimbus and stock payload compatibility as the release guardrail
• avoid broad subsystem rewrites without tests and live-capture evidence

4. Holidays And Seasonal Personality

• port holiday-aware personality responses as a visible extension of the new persona slice
• start with a small, source-backed set (for example birthdays/holidays already represented in legacy data paths)
• ensure holiday responses feel characterful while still routing through stock-compatible payloads

5. Multi-Tenant Memory Storage Foundation

• define tenant boundaries across account, loop, device, and person-memory records
• add storage abstractions that can move from in-memory/local JSON to hosted SQL/Blob without reworking behavior layers
• implement memory-ready schemas and repository contracts for user facts (names, birthdays, personal dates, preferences) with strict tenant scoping
• seed person-aware state keys now so future interactions can scope to account + loop + device + person without another shape change
• keep stateful interaction flows repository-backed instead of embedding more ad hoc metadata in the websocket layer

6. Multi-Server Sync Path

• document the eventual sync boundary for stateful data that should move between servers
• treat the first pass as repository-local durability, then layer replication and conflict handling on top
• prefer explicit change records or versioned state snapshots over implicit last-writer wins when we outgrow a single node
• keep cross-server reconciliation out of the hot path until the single-server semantics are stable

First Implemented Slice In 1.0.19

The first delivered slice in this release is persona expansion:

• how old are you
• when's your birthday
• do you have a personality
• make a pizza

make a pizza is now wired to the legacy scripted-response identity (RA_JBO_MakePizza) with pizza-making animation ESML, based on the original skill manifests.

This slice is intentionally small and user-visible. It creates immediate personality gains while we keep deeper platform work in parallel.

Second Implemented Slice In 1.0.19

The second delivered slice is first tenant-scoped personal memory:

• store birthday from phrases like my birthday is April 12
• recall birthday from phrases like when is my birthday
• store preferences from phrases like my favorite music is jazz
• recall preferences from phrases like what is my favorite music

Memory keys are scoped by account/loop/device tenant context so one tenant does not leak into another.

Third Implemented Slice In 1.0.19

The third delivered slice starts memory-triggered proactivity and broadens memory parsing:

• surprise me now runs a weighted proactivity selector
• selectors use tenant-scoped memory signals (favorites and likes/dislikes) plus date triggers
• February 9 (National Pizza Day) can proactively launch the pizza animation path
• proactive pizza fact offer flow now stores pending offer state and resolves direct yes / no follow-up answers
• memory parsing now covers:
  • names (my name is ..., what is my name)
  • important dates (our anniversary is ..., when is our anniversary)
  • likes/dislikes (i like ..., i love ..., i dislike ..., i don't like ...)
  • favorite phrase variants including reverse form (pizza is my favorite food)

Fourth Implemented Slice In 1.0.19

The fourth delivered slice starts weather compatibility using Pegasus-style report-skill routing:

• weather phrases now route to report-skill instead of generic placeholder chat
• outbound NLU launch uses legacy reactive intent requestWeatherPR (source-aligned with Pegasus manifests/tests)
• weather entity hints are added for:
  • date = tomorrow on tomorrow phrasing
  • Weather = rain|snow|... on condition questions (for example will it rain tomorrow)
• websocket output now emits local skill redirect + silent completion for weather launch, matching existing local-skill launch patterns

Fifth Implemented Slice In 1.0.19

The fifth delivered slice adds provider-backed weather content while preserving Pegasus launch compatibility:

• OpenWeather provider abstraction and infrastructure wiring are added to the hosted cloud
• weather requests still launch report-skill with requestWeatherPR and legacy weather/date entities
• weather replies now include cloud-generated spoken summaries from provider data:
  • current conditions (Right now in ...)
  • tomorrow forecast shape (Tomorrow in ...) with high/low temperatures when available
• simple location extraction is supported for phrasing like what's the weather in Chicago tomorrow
• provider config supports appsettings and OPENWEATHER_API_KEY environment fallback for deployment

Personality Import Ladder

This is the practical plan for importing legacy Jibo mims into OpenJibo without pretending we already have a full Pegasus runtime.

What Is Possible Today

OpenJibo can already host a meaningful subset of legacy personality content because it has:

• a shared catalog for content-driven replies
• chitchat state-machine routing with route metadata
• outbound payload support for skillId, mim_id, mim_type, prompt_id, prompt_sub_category, and ESML
• existing examples that already behave like legacy MIMs for pizza, dance, news, weather, and generic chat

What We Need To Build

To move from hand-wired examples to broader imports, we need three small platform pieces:

1. a MIM inventory importer that can scan the legacy tree and produce a normalized catalog
2. a prompt-selection layer that can choose by skill_id, mim_id, prompt category, and condition metadata
3. a safe ESML/prompt renderer that preserves existing stock-compatible payload shapes

What Can Be Ported With Each Build

Build A: Declarative Prompt Packs

Port immediately:

• core-responses
• deflector
• the simplest emotion-responses
• any scripted-responses that are just direct prompt lists with no special state machine

Why these first:

• they are already close to the current JiboExperienceCatalog model
• they give us user-visible personality quickly
• they are the best fit for low-risk testing tomorrow

Build B: Conditioned Prompt Packs

Port after the importer and renderer are in place:

• gqa-responses
• structured emotion responses with condition gates
• prompt sets that select different replies by user state or Jibo state

Why these next:

• they are still mostly declarative
• they need a small amount of condition evaluation, but not a new conversation engine

Build C: Conversation Families

Port after Build B:

• richer scripted-responses families that depend on follow-up state
• special-date / holiday personality sets
• more nuanced chitchat branches that need context-aware routing

Why these later:

• they need state and follow-up behavior, not just prompt selection
• they are where personality feels most alive, but they are also where bugs will be easiest to introduce

Build D: Full Parity Cleanup

Port after the core ladder is stable:

• large cross-skill collections
• any MIMs that depend on Pegasus-only parser assumptions
• any files that need a dedicated runtime abstraction instead of catalog lookup

System Diagram Alignment Snapshot (2026-05-06)

Legacy architecture (system_diagram.png) has been mapped to current OpenJibo cloud services so release execution stays anchored to:

• where we were (Pegasus/Jibo cloud design intent)
• where we are (current hosted .NET modular monolith)
• where we are headed (durable memory, proactivity catalogs, parser depth, provider aggregation)

Reference:

• system-diagram-alignment.md

Greetings And Presence Planning Snapshot (2026-05-07)

Pegasus greeting and presence behavior has now been captured into a source-anchored OpenJibo implementation plan.

Reference:

• greetings-presence-plan.md

Live Validation Snapshot (2026-05-07)

User-confirmed end-to-end behavior now includes:

• Hey Jibo -> What's your cloud version? (working)
• Hey Jibo -> What's the time? (working)
• Hey Jibo -> Surprise me -> pizza fact -> $YESNO (Yes) -> fact (working)
• Hey Jibo -> Surprise me -> pizza fact -> $YESNO (No) -> decline reply (working)

This confirms the pizza-fact offer state now keeps the yes/no branch open through completion and does not require a second wake-word reset for the follow-up answer.

Personal Report Planning Snapshot (2026-05-07)

Personal report parity planning is now captured with Pegasus source anchors for weather visuals/animations, live news, commute, and calendar gap coverage.

Reference:

• personal-report-parity-plan.md

Next Queued Task (2026-05-06)

Queued next 1.0.19 implementation task (now started):

• dialog parsing expansion and ambiguity guardrails

Execution focus:

• import additional Pegasus parser phrases/entities into intent handling where safe
• reduce trigger-only captures that drop the rest of the utterance
• preserve command-vs-question personality split and local skill payload compatibility
• add focused tests for new phrase families and ambiguity boundaries
• keep listener-state observability aligned with the legacy GLSM flow while phrase guardrails are added

First completed guardrail slice under this queue:

• GLSM listener flow capture + telemetry mapping
• stale pending-listen recovery path for long-open no-context/no-audio listens

Second completed guardrail slice under this queue:

• tightened date/time ambiguity handling (what's your birthday/what's your bday no longer falls into date intent)
• expanded Pegasus-inspired memory/weather phrase coverage:
  • birthday alias parsing (my bday is ..., when is my bday)
  • shorthand preference sets (my favorite sport football)
  • weather variants (what's today's weather look like, will it be sunny tomorrow)
• listener continuation guardrail now differentiates incomplete preference fragments from complete shorthand preference sets

Third completed guardrail slice under this queue:

• expanded Pegasus userLikesThing / userDislikesThing / doesUserLikeThing / doesUserDislikeThing phrase-family coverage
  • includes additional dislike/negation variants (loathe, did not like, didn't enjoy, don't really like)
  • includes group-preference variants (we like, we love, we dislike, we can't stand)
  • includes lookup variants (do you think i like ..., do you believe i don't like ...)
• added affinity set/lookup attempt guardrails so partial captures route to affinity prompts instead of generic chat
• extended auto-finalize continuation deferral for the new Pegasus affinity stems (we like, i loathe, and related variants)
• added focused interaction + websocket tests for the new parser/guardrail behavior

Next queued implementation track after parser guardrails:

• personal report parity slices (weather visual parity, live news path, commute/calendar gap closure)

First completed slice in this personal-report parity track:

• added provider-ready news briefing path with Nimbus-compatible news payload continuity
• preserved fallback behavior when no live provider is configured
• added memory/transcript category hinting for provider requests (sports, technology, business, etc.)
• added provider-side request caching for both news and weather to reduce integration churn and repeated lookups
• added focused interaction + websocket tests for provider-backed news speech output and request-hint plumbing

Next Slices

1. MIM import foundation for personality expansion
2. Dialog parsing expansion
3. Presence-aware greetings and identity-triggered proactivity
4. Personal report parity slices
5. Holidays and seasonal personality slice beyond pizza day
6. Durable memory persistence path
7. Update/backup/restore end-to-end proof
8. STT noise-screening and short-utterance reliability pass
9. Provider-backed news expansion and deeper weather parity
10. Capture indexing and retention boundary for group testing

For slice 1, use the new import ladder above to keep the work grounded in what OpenJibo can already render today versus what needs new scaffolding. For slices 2-5, use Pegasus phrase lists, MIM IDs, and behavior patterns as the source anchor before broadening into OpenJibo-native improvements.

Definition Of Done

Release 1.0.19 is complete when:

• planned slices have focused tests and updated docs
• regression checklist passes for the existing stock-OS compatibility paths
• live runs confirm no critical regressions in alarms, gallery, yes/no, and cloud-version diagnostics
• memory/personality storage proves tenant isolation by account/loop/device boundaries and is compatible with the target hosted cloud footprint