ltcai 8.8.0 → 9.0.0

package/docs/CHANGELOG.md

@@ -6,6 +6,118 @@ existed at that release.
 
 ## [Unreleased]
 
+No unreleased changes yet.
+
+## [9.0.0] - 2026-07-08
+
+### Added
+- Added Brain Brief suggested questions that turn current memory, recall proof,
+  graph concepts, and conversation history into clickable first-screen prompts.
+- Suggested Brain questions now send immediately from the first screen instead
+  of only filling the composer.
+- Added one-click follow-up prompts under the latest Brain answer for turning a
+  reply into a checklist, evidence review, or prioritized next steps.
+- Added a Brain chat to Review Center handoff so users can save an answer as a
+  reviewable task draft and manage it alongside automation suggestions.
+- Added direct Brain-to-Agent delegation and successful agent-run synthesis into
+  durable Brain memory/graph context.
+- Surfaced recent agent-synthesis memories in Brain overview and memory rings
+  so delegated work is visibly reflected on the home screen.
+- Improved agent-run Brain synthesis quality by splitting successful results
+  into key facts, decisions, and follow-up memories with structured metadata.
+- Agent follow-ups now enter Review Center as task drafts so delegated work
+  produces actionable approval candidates instead of passive memory only.
+- Approving an Agent follow-up review item now promotes it into a manual
+  workflow draft with trigger, agent, and output nodes.
+- Added large-feature foundations for KG/Retrieval scale diagnostics,
+  background ingestion scheduling, offline multimodal image captions,
+  proactive contradiction detection, and ingestion bridge marketplace templates.
+- Added proactive Brain action cards that turn Brain Brief evidence into
+  one-click ask, Agent delegation, Review Center draft, or graph navigation
+  actions on the Brain home screen.
+- Added a visible proactive Brain action trail so one-click suggestions show
+  their running/completed/failed state after the user acts on them.
+
+### Changed
+- Continued app-factory decomposition by extracting user profile/API-key helper
+  wiring into `latticeai.runtime.user_key_runtime`, keeping the legacy
+  `server_app` callable surface while making keyring/plaintext fallback policy
+  independently testable.
+- Split additional runtime and static-data seams out of app, chat, MCP, model,
+  and Knowledge Graph modules while preserving re-export compatibility for
+  existing imports.
+- Routed Computer Use direct `/cu/*` actions through the shared ToolRegistry
+  policy gate and audit lifecycle, preserving route paths while blocking
+  non-admin direct desktop control by default.
+- Moved blocked-system-prefix protection into `tools.local_write` itself so
+  local filesystem writes fail closed even when called outside the HTTP approval
+  route.
+- Narrowed the lazy `server_app` compatibility namespace by filtering
+  app-factory scratch imports and runtime wiring dictionaries while preserving
+  explicit legacy helpers, and added a typed `RuntimeBundle` migration target
+  behind the legacy `_RUNTIME_BUNDLE` dict.
+
+### Fixed
+- Added regression coverage for provider API-key lookup/storage behavior,
+  including keyring precedence, plaintext fallback gating, legacy plaintext
+  cleanup after keyring writes, and identity creation on plaintext fallback.
+- Added regression coverage for Computer Use policy enforcement, audit-safe
+  typed-text metadata, and direct local-write system-prefix blocking.
+- Fixed functional findings from the July 8 code review: file generation now
+  fails cleanly when no model is loaded, chat/document streams preserve a
+  terminal SSE event on generation errors, agent runs persist failed status on
+  executor exceptions, Brain delegation treats HTTP failures as failures, and
+  local permission expiry cleanup no longer corrupts the active token lookup.
+- Tightened non-security chat intent detection, Telegram bot server URL
+  configuration, LATTICE_TZ-aware runtime audit timestamps, local embedding
+  dimension consistency, and stale Brain UI version copy.
+- Paid down the remaining July 8 cleanup debt by moving duplicated JSON/ISO/hash
+  and setup detection helpers into shared modules, switching runtime audit
+  appends to JSONL while preserving legacy JSON reads, making the legacy runtime
+  namespace allowlist-based, clarifying static-vs-SPA design token ownership,
+  and consolidating duplicated frontend helper functions.
+- Reduced the remaining chat-router risk by extracting repeated chat history,
+  bridge notification, no-model, single-answer, direct-file, and agent-file
+  response paths out of the main `/chat` handler, with regression coverage for
+  the shared fast-path epilogue.
+
+## [8.9.0] - 2026-07-06
+
+### Added
+- Added authenticated user/workspace scoping to durable conversation history
+  reads and deletes.
+- Added workspace-aware Knowledge Graph search, traversal, relationship, node,
+  and chat-context reads.
+- Added direct HTTP/MCP Tool API policy enforcement for registry-governed tools.
+- Added permission approval queue hashing and atomic writes so raw tokens are
+  not persisted at rest.
+- Added confirmation-token guarded installer/process command plans with redacted
+  local process audit events.
+- Added regression coverage for scoped history, graph scoping, tool policy
+  gates, AgentRuntime approval semantics, permission tokens, session TTL
+  injection, and model-download runtime config.
+
+### Changed
+- AgentRuntime now requires explicit human approval for non-auto-approved plans
+  and rolls back git-governed tool results even when `success` is omitted.
+- Model download consent now uses configured runtime state instead of direct
+  environment-variable reads.
+- Frontend API base logic, CSS token/base rules, and i18n literal checks were
+  split into smaller maintainability seams.
+- Version bumped to 8.9.0 across Python, npm, VS Code extension, Tauri, static
+  metadata, readiness gates, release notes, and current-release documentation.
+- Documentation now states that SQLite is the live local Brain store; Postgres
+  remains optional scale/migration tooling.
+
+### Fixed
+- Conversation store migration now creates the workspace index only after the
+  scope columns exist.
+- Direct `write_file`/`edit_file` policy lookup now blocks system write
+  prefixes consistently with local-file approvals.
+- Workspace selection clearing now removes the persisted workspace id.
+- Tauri/local API fetches now include credentials for localhost backend
+  sessions.
+
 ## [8.8.0] - 2026-07-06
 
 ### Added
@@ -196,7 +308,7 @@ existed at that release.
 - Updated architecture and product readiness targets to 8.0.0.
 - Synchronized package/runtime/static/Tauri metadata to 8.0.0.
 - Updated current-release docs and exact artifact names to 8.0.0 while
-  preserving historical 7.x entries.
+  setting 8.0.0 as the oldest retained release-history entry.
 
 ### Fixed
 - Made logical Knowledge Graph `replace` imports transactional so malformed
@@ -205,239 +317,3 @@ existed at that release.
   `relationship_search`, and `traverse`.
 - Preserved colliding legacy edge labels during logical import/backfill while
   keeping native write-door synonym dedupe canonical.
-
-## [7.9.0] - 2026-06-23
-
-### Changed
-- Added `SingleAgentRuntime` as the explicit name for the legacy single-agent
-  state machine while preserving `AgentRuntime` as a compatibility alias.
-- Updated tool dispatch to build `SingleAgentRuntime` directly.
-- Moved single-agent git rollback behind an injected `rollback_file` port owned
-  by `ToolDispatchService`, keeping shell execution out of the core state
-  machine.
-- Added a shared `runtime-boundary/v1` descriptor so product and single-agent
-  runtime boundaries are machine-readable in config/tests.
-- Added `RuntimeBoundaryProtocol` as the minimal shared inspection surface for
-  runtime-boundary-aware dependency injection.
-- Updated architecture and product readiness targets to 7.9.0.
-
-## [7.8.0] - 2026-06-22
-
-### Changed
-- Rebuilt Brain Chat Home around immediate conversation: chat purpose, starter
-  prompts, and the composer now occupy the first viewport.
-- Collapsed source ingestion, readiness, proof, timeline, overview, model
-  continuity, and care controls behind one utility drawer.
-- Kept workspace navigation visible on the default Brain surface.
-- Hid default depth controls until the user intentionally travels deeper.
-- Integrated the six post-7.7 UX drafts into the canonical Brain experience:
-  first-run value cards, stronger recommendation affordances, product-toned
-  Brain Home copy, and routed legacy Brain conversation entry points to the
-  canonical Brain Home surface.
-- Removed obsolete Brain conversation and first-run guide components.
-- Moved draft onboarding polish out of inline styles and into the shared design
-  stylesheet with responsive behavior and bilingual copy.
-- Updated architecture and product readiness targets to 7.8.0.
-- Refreshed 7.8.0 release screenshots, walkthrough video/GIF, and capture
-  notes under `output/release/v7.8.0/`.
-
-## [7.7.0] - 2026-06-22
-
-> 7.7.0 marks the complete, finished product stage for Lattice AI.
-> After 7.6.0 architecture closure, this release polishes every surface so that anyone looking at the code, UI, docs, or running app immediately recognizes: "this is now a product".
-
-Lattice AI v7.7 delivers the Living Brain as the undeniable center, production-grade runtime contracts, stable ToolRegistry, full ingestion-to-graph flows, bilingual professional UX, and zero-beta signals. Classifiers moved to Production/Stable. All prior gates remain enforced under finished product contract.
-
-Package metadata, Tauri, frontend, Python all aligned to 7.7.0. UI/UX microcopy and signals updated to convey finished professional tool.
-
-### Productization Highlights
-- Extreme self + claude-code (pts_claudecode) used for polish, evaluation, iteration.
-- "This is a product" bar: clear durable knowledge ownership, no loose ends.
-- Validation: typecheck, unit, cargo, build scripts exercised.
-
-### Changed
-- Package/runtime/static metadata synchronized to 7.7.0.
-- Development status to Production/Stable.
-- All current-release references point to 7.7.0.
-
-## [7.6.0] - 2026-06-22
-
-> Brain-Centered UX & Architecture Closure. Incorporates the two local review
-> files into the release line with a Wake Brain first-run surface, memory rings
-> plus direct depth controls, and machine-checkable architecture readiness gates.
-
-### Added
-- Wake Brain first-run entry before owner/profile setup, reducing onboarding to
-  the product promise first and the setup mechanics second.
-- Concentric memory rings around Brain Home plus direct controls for Now, Memory,
-  Topics, Relationships, and Full Graph navigation.
-- `latticeai.services.architecture_readiness.architecture_readiness()` and
-  `tests/unit/test_v76_review_completion.py` to keep AgentRuntime, ToolRegistry,
-  Config, server decomposition, KG hardening, and Brain UX review closure under
-  test.
-
-### Changed
-- Package/runtime/static metadata is synchronized to 7.6.0; package publish and
-  deployment remain owner-run only.
-- README and release docs now describe 7.6.0 as the current release and point to
-  refreshed 7.6.0 screenshots, walkthrough GIF, and release evidence index under
-  `output/release/v7.6.0/`.
-
-## [7.5.0] - 2026-06-20
-
-> Runtime Debt Burn-down & Release Risk Cleanup. Turns the 7.4.0 contract
-> envelope into a consumed API surface, expands retrieval quality to a 250+
-> record local corpus fixture, and removes release/security warnings.
-
-### Added
-- `extract_contract`, `require_contract`, `contract_view`, and `contract_views`
-  helpers for consumers that need a surface-agnostic `agent-run-contract/v1`
-  projection.
-- AgentRuntime status/list/detail/events and realtime feed responses now expose
-  compact `contracts` views alongside legacy payloads.
-- Deterministic 250+ record retrieval benchmark corpus while keeping 12 judged
-  queries and real `KnowledgeGraphStore` + `SearchService` execution.
-- Refreshed README release evidence screenshots and walkthrough GIF under
-  `output/release/v7.5.0/`.
-
-### Changed
-- Tauri Rust/CLI dependencies are updated within the Tauri 2 line, removing the
-  old transitive `block v0.1.6` future-incompatibility warning.
-- npm dependency overrides move `js-yaml` to a non-vulnerable version; `npm
-  audit` reports 0 vulnerabilities.
-- CI lint compatibility is restored for the Brain quality gate script.
-- Local MLX model preparation now recognizes valid existing Hugging Face cache
-  snapshots, avoiding an unnecessary re-download when the model already exists
-  outside Lattice's managed `~/.ltcai/hf-models` directory.
-- Package/runtime/static metadata is synchronized to 7.5.0; package publish and
-  deployment remain owner-run only.
-
-## [7.4.0] - 2026-06-20
-
-> Runtime Contract Convergence & Corpus Retrieval. Completes the
-> agent-run-contract/v1 family across run storage, workflow execution, audit
-> events, realtime events, and a real corpus-scale retrieval quality gate.
-
-### Added
-- Persisted agent and workflow run rows now carry refreshed contract metadata
-  for queued, running, terminal, cancelled, and interrupted states.
-- Workflow engine results, replay payloads, audit log events, and realtime SSE
-  feed events now expose the same `agent-run-contract/v1` family envelope while
-  preserving existing top-level compatibility fields.
-- Corpus-scale retrieval fixture with 30+ documents, judged queries,
-  must-include expectations, and thresholds for recall, precision, NDCG, and
-  hit rate.
-- `scripts/brain_quality_eval.py` now exercises the real local
-  `KnowledgeGraphStore` + `SearchService` hybrid retrieval path before scoring.
-
-### Changed
-- `RetrievalBenchmarkRunner` reports dynamic metric aliases for the selected
-  `top_k` and a `must_include_hit_rate`.
-- Package/runtime/static metadata is synchronized to 7.4.0; package publish and
-  deployment remain owner-run only.
-
-## [7.3.0] - 2026-06-20
-
-> Runtime Contract & Retrieval Quality. Turns the next AgentRuntime extraction
-> step and the uploaded roadmap's hybrid-search quality goals into a small,
-> tested release: shared run contracts and deterministic recall regression.
-
-### Added
-- `lattice_brain.runtime.contracts.AgentRunContract`, a serializable
-  `agent-run-contract/v1` payload shared by single-agent and multi-agent
-  execution paths.
-- Multi-agent API result/run patches now include the shared contract with
-  runtime, mode, status, roles, retries, timeline, and terminal-state data.
-- Single-agent runtime exposes the same contract helper for UI/API/storage
-  convergence in the next extraction pass.
-- `scripts/brain_quality_eval.py` now runs deterministic hybrid recall/ranking
-  regression checks with recall and precision thresholds.
-
-### Changed
-- Package/runtime/static metadata is synchronized to 7.3.0; package publish and
-  deployment remain owner-run only.
-
-## [7.2.0] - 2026-06-20
-
-> Runtime Trust Baseline. Adds execution preview and registry diagnostics so
-> agent runs and tool permissions become inspectable contracts before action.
-
-### Added
-- `POST /agents/api/run/preview` for AgentRuntime readiness, role selection,
-  input keys, retry clamping, and blocking reasons without starting a run.
-- `GET /tools/registry` for the live ToolRegistry manifest across dispatch
-  handlers, governance policy, catalog descriptions, and permissions.
-- `GET /tools/registry/diagnostics` for a compact drift check suitable for CI,
-  admin views, and runtime health panels.
-- Unit coverage for AgentRuntime preview and ToolRegistry manifest diagnostics.
-
-### Changed
-- Tool governance now covers `read_document`, and the catalog describes
-  `create_web_project`, closing the current dispatch/governance/catalog drift.
-- Package/runtime/static metadata is synchronized to 7.2.0; package publish and
-  deployment remain owner-run only.
-
-## [7.1.0] - 2026-06-20
-
-> Brain Usability Completion. Completes the 7.1.0 first-run through editor-sync
-> usability pass: clear onboarding, visible ingestion, graph controls, inline
-> proof, workspace/admin discovery, feedback states, and VS Code sync status.
-
-### Added
-- Hardware visualization, expected timing, install timeline, and next-action
-  copy in first-run onboarding.
-- Brain Home ingestion stage disclosure and memory emergence timeline for file,
-  folder, note, and URL sources.
-- Knowledge Graph search suggestions, entity type filters, recent/all-time time
-  exploration, focus clearing, and neighbor highlighting.
-- Inline answer citation markers with keyboard-accessible proof cards.
-- Workspace/profile switcher, Admin Console gate, consent revoke feedback, and
-  shared empty/error feedback surfaces in the Brain shell.
-- VS Code extension heartbeat/status endpoint plus extension and main-app sync
-  indicators for connected/indexing/synced/offline states.
-
-### Changed
-- Package/runtime/static metadata is synchronized to 7.1.0; package publish and
-  deployment remain owner-run only.
-
-## [7.0.0] - 2026-06-18
-
-> Brain Productization Loop. Turns the Brain proof work into a first-five-minute
-> product flow: add sources, ask a question, see proof/citations, and verify the
-> same Brain evidence after switching models.
-
-### Added
-- Brain Home ingestion panel for files, local folder paths, notes, and web URLs,
-  all backed by existing workspace-scoped ingestion routes.
-- Answer-level Memory proof and source citation cards rendered under assistant
-  responses after Brain proof refreshes for the user's query.
-- Model-continuity demo strip that lets the user recheck the same Brain
-  evidence and jump to model switching from the Brain flow.
-- Deterministic `scripts/brain_quality_eval.py` recall/KG quality gate, wired
-  into CI after the unit suite.
-- Visual mock coverage for Brain proof, document upload, note ingest, folder
-  indexing, and web URL ingestion endpoints.
-
-### Changed
-- Brain Home is now ingestion-first instead of chat-first: first screen action
-  labels are files/folders/notes/web, with deeper graph/model/settings still
-  reachable from the shell.
-- Package/runtime/static metadata is synchronized to 7.0.0; package publish and
-  deployment remain owner-run only.
-
-- Completed ALL Recommended next refactor items from report #15 in this session:
-  - Server decomp wave (model_runtime globals/wiring): model_loading.py for prepare_and_load and stream, _MODEL_RUNTIME_STATE, model_engines.
-  - Deeper WorkspaceOSStore (timeline + plugins + snapshots + memory).
-  - KG embed: set_embed_dim.
-- All refactoring needed finished this session per AGENTS. 767 tests, builds, greps clean.
-
-### Added (this session features)
-- `vision_analyze` tool: new multimodal vision analysis tool using screenshot b64 + prompt. Leverages existing VLM support (image_data in generate). Added to computer-use agent prompt and general tools. Fits seamlessly with computer_use, agent runtime, tool registry, and VLM models without affecting text-only paths.
-- More recent multimodal models in user recommendations (Llama 3.2 11B Vision, Phi-3.5 Vision, Qwen2.5-VL 7B, Moondream2) + family order update in model_recommendation. Expanded curated list in model_capability_registry for better local VLM choices on Apple Silicon and other.
-- All additions checked for compatibility with existing KG (descriptions can be ingested), agents/tools dispatch, chat/computer_use (image pass-through), model rec logic, and non-multimodal fallbacks.
-
-- Perfect completion of report #15 recommended refactor:
-  - Server decomp wave: ModelRuntimeState class (not dict) for globals/wiring in model_runtime; sync_to_module_globals for compat; further clean.
-  - Deeper WorkspaceOSStore: timeline/plugins managers fully delegated and composed (record, has_permission etc all through).
-  - KG embed: set_embed_dim available for optional central handling.
-- All changes preserve legacy exactly, full composition, small modules.