Conversation
….6.2 - claude-flow: installed globally and via npm (v3.1.0-alpha.28) - agentic-qe: updated from 3.1.5 to 3.6.2 in Dockerfile.unified - agentic-qe mcp-config: updated version and dependency pins from 2.8.2 to 3.6.2 - Updated claude-flow version label in Dockerfile from 3.0 to 3.1 https://claude.ai/code/session_01MshJuJJ15YyUoi2TYXnqRJ
…-qe to 3.6.2" This reverts commit 01afb6f.
…rcadia parity audit Server-Authoritative Layout (Single Source of Truth): - Remove client-side force-directed physics from graph.worker.ts (~230 lines) - Remove computeForces(), applyForces(), domain clustering, spatial grid repulsion - All graph types (visionflow, logseq) now use server GPU physics exclusively - Client performs optimistic tweening only (interpolation toward server targets) - setUseServerPhysics() always enforces server mode - processBinaryData() accepts updates for all graph types (removed logseq-only guard) - graphDataManager.ts: removed graph-type guard on binary position updates WebGPU TSL Hardening: - GlassEdgeMaterial: added TSL animated flow emissive using time-driven positionLocal.y - Verified all material paths: GemNodeMaterial, CrystalOrbMaterial, AgentCapsuleMaterial - Confirmed drawIndexed(Infinity) protection via DataTexture metadata pipeline - Verified clean WebGL fallback (no hybrid path, no visual regression) Vircadia/Quest3 Parity Audit: - Documented server-authoritative position flow in CollaborativeGraphSync.applyOperation() - Identified disconnected sync layers (EntitySyncManager not wired to position flow) - Documented remaining work for bi-directional entity sync PRD-AFD-DDD Audit Documents: - PRD: Requirements for WebGPU TSL audit scope - AFD: Architecture for server-side layout with client optimistic tweening - DDD: Detailed rendering pipeline design with material feature matrix - Swarm findings: Complete audit report with fix status and remaining work https://claude.ai/code/session_01MshJuJJ15YyUoi2TYXnqRJ
… renderer capabilities WebGPU TSL Material Upgrades: - CrystalOrbMaterial: Add depth-pulsing emissive node (cosmic spectrum driven by pulseSpeed uniform) + per-instance vertexColor with Fresnel brightening - AgentCapsuleMaterial: Add bioluminescent heartbeat pulse emissive driven by activityLevel uniform + traveling wave via positionLocal.y + green rim tinting Client Tweening (Server-Authoritative): - Add ClientTweeningSettings interface (lerpBase, snapThreshold, maxDivergence) - graph.worker.ts: Configurable lerp base and snap threshold via setTweeningSettings() - graph.worker.ts: maxDivergence enforcement — force-snap when divergence exceeds limit - Settings interface: Add clientTweening and rendererCapabilities to Settings type Renderer Capabilities Report: - rendererFactory.ts: Export rendererCapabilities object populated at init time - Reports backend type, TSL status, bloom mode, GPU adapter name, max texture size - WebGPU path reads adapter info; WebGL path reads GL context parameters Physics Settings Documentation: - Annotated PhysicsSettings interface: all params route to server via PUT /api/settings - Added RendererCapabilities interface for settings panel display https://claude.ai/code/session_01MshJuJJ15YyUoi2TYXnqRJ
…sync Root cause: binary position frames from server piled up unbounded on fast LANs, overwhelming the worker pipeline and triggering permanent error modals. Data velocity management: - websocketStore: latest-only binary frame coalescing via queueMicrotask — intermediate frames are dropped, only the newest is processed - binaryProtocol: Protocol V4 frames now logged instead of silently returning empty arrays - graphWorkerProxy: removed logseq-only guard on binary processing (all graph types are server-authoritative), added consecutive error tracking and worker health monitoring to requestTick() - graphDataManager: increased worker init timeout from 3s to 12s with recovery attempt, wired transient error tracking - workerErrorStore: replaced permanent red screen with transient error counter — auto-clears when data resumes, only escalates after 30 consecutive failures Bi-directional Vircadia sync: - CollaborativeGraphSync now initializes EntitySyncManager - node_move operations forward positions to Vircadia entity layer - Registered onEntityUpdate callback for server→client reconciliation with 0.1-unit jitter threshold - Proper cleanup in dispose() https://claude.ai/code/session_01MshJuJJ15YyUoi2TYXnqRJ
… backoff BatchQueue: - Added exponential backoff (2^n * 100ms) on failed batch retries instead of immediate re-enqueue, preventing retry storms under load Debug UI cleanup: - GraphCanvas stats overlay: changed from alarming red (rgba 255,0,0) to neutral dark overlay — the red background was mistakable for an error state - CanvasErrorBoundary: softened colors from #ff6b6b to warm amber/grey, added retry button to recover without full page refresh Renderer factory (Quest3/Vircadia parity): - Added XR headset detection (Quest/Oculus/Pico) for pixel ratio capping — XR devices capped to 1.0 DPR to prevent GPU memory blowup on stereoscopic render targets - Added 5-second timeout on WebGPU renderer.init() — Quest 3 Oculus Browser can hang during GPU adapter negotiation, now falls back to WebGL gracefully - XR devices get descriptive GPU adapter name in capabilities panel https://claude.ai/code/session_01MshJuJJ15YyUoi2TYXnqRJ
Tab reorganization: - Dashboard (stub buttons): removed — integrated into Performance tab via PerformanceControlPanel + HealthDashboard - Auth: merged into System → Authentication subsection (was a 5-setting tab that didn't warrant its own top-level entry) - System Health: merged into Performance tab (co-located with perf metrics and quality gates) - Ontology Inference: merged into Analytics tab alongside SemanticAnalysisPanel and node filter settings - Quality Gates: surfaced in Performance tab (was defined in UI definition but had no dedicated tab) UX improvements: - Node Confidence Filter moved from buried Analytics → Visualization tab where users expect to control what they see - Removed duplicate per-graph physics subsections from Visualization (they duplicated the Physics & Layout tab's PhysicsControlPanel) - Fixed triple Brain icon: each tab now has a distinct icon (Eye, Zap, Activity, Network, HeartPulse, Smartphone, Settings, Brain, Code) Server connectivity: - Quality Gates settings (GPU acceleration, layout mode, ontology physics, semantic forces) now rendered inline in Performance tab, routed to PUT /api/settings/quality-gates via settingsApi - Node filter settings properly reach PUT /api/settings/node-filter from their new Visualization location https://claude.ai/code/session_01MshJuJJ15YyUoi2TYXnqRJ
- Delete 5 never-imported shell panels (AIPanel, AdvancedSettingsPanel, SystemPanel, XRPanel, VisualisationPanel) that only showed placeholder text - Wire GraphVircadiaBridge syncNodeToEntity/syncEdgeToEntity to EntitySyncManager + GraphEntityMapper for real Vircadia entity sync - Implement undo/redo in SettingsPanelRedesign with ref-based history stack (50 snapshots) replacing logger.info stubs - Clean up MainLayout: remove commented-out dev imports, unused React hooks https://claude.ai/code/session_01MshJuJJ15YyUoi2TYXnqRJ
… and turbo-whisper Four-plane audio architecture for real-time agent voice control + spatial chat: - Plane 1: PTT held → mic → turbo-whisper STT → agent commands (private per-user) - Plane 2: Agent response → Kokoro TTS (per-agent voice preset) → owner only - Plane 3: PTT released → mic → LiveKit SFU → spatial voice chat (all users) - Plane 4: Agent TTS → LiveKit virtual participant at agent's 3D position Backend: AudioRouter with user-scoped broadcast channels, TurboWhisper STT provider, AgentSpatialInfo/AudioTarget types, VoiceRoutingSettings config with LiveKit + per-agent presets. Frontend: PushToTalkService (push/toggle), LiveKitVoiceService (HRTF spatial panning from Vircadia positions), VoiceOrchestrator wiring PTT+VoiceWS+LiveKit+AudioInput. Infra: docker-compose.voice.yml with LiveKit SFU, turbo-whisper (faster-whisper CUDA), Kokoro TTS. Opus 48kHz mono throughout the entire pipeline. https://claude.ai/code/session_01MshJuJJ15YyUoi2TYXnqRJ
Add agent read/write paths for ontology data with Whelk reasoning, GitHub PR feedback loop for human-in-the-loop oversight, and MCP tool surface types for agent-callable ontology operations. New services: - OntologyQueryService: semantic discovery, enriched note reading, validated Cypher execution with OWL schema hints - OntologyMutationService: note creation/amendment proposals with Whelk consistency checks, Logseq markdown generation, GitHub PR - GitHubPRService: GitHub REST API integration for agent write-back (blob → tree → commit → branch → PR) New types: - ontology_tools.rs: full MCP tool input/output contracts for discover, read, query, traverse, propose, and validate operations Config: - OntologyAgentSettings: auto-merge threshold, confidence gates, per-user note paths, PR labels, consistency check toggles Design docs: - PRD: requirements and success metrics (+40% task accuracy target) - AFD: system architecture and data flow diagrams - DDD: detailed service design and Neo4j schema additions https://claude.ai/code/session_01MshJuJJ15YyUoi2TYXnqRJ
…ion tests - Wire OntologyQueryService, OntologyMutationService, and GitHubPRService into main.rs startup with proper Arc<> DI and app_data registration - Create ontology_agent_handler.rs with 7 REST endpoints: discover, read, query, traverse, propose, validate, status - Register 7 ontology MCP tools in mcp-server.js with VisionFlow API routing - Add 13 integration tests covering discovery, read, Cypher validation, proposal creation, markdown generation, amendment, and quality scoring - Tests use MockOntologyRepository + real WhelkInferenceEngine https://claude.ai/code/session_01MshJuJJ15YyUoi2TYXnqRJ
- Fix 11 broken links in docs/README.md (explanations/ → explanation/) - Replace SQLite/unified.db references with Neo4j across 15+ architecture docs - Convert ASCII box-drawing diagrams to Mermaid in services.md, github-sync-service-design.md, pipeline-integration.md, ontology-pipeline-integration.md - Rewrite project structure doc (02-project-structure.md) to reflect actual Rust codebase - Create new docs: ontology-agent-tools.md (7 MCP tools), voice-routing.md (4 audio planes) - Update README.md with Voice Routing and Ontology Agents capability rows - Update CHANGELOG.md with v1.2.0 (voice, ontology agents, docs overhaul) - Expand AGENTS.md with agent capabilities, MCP tools, and code conventions - Update system-overview.md: Neo4j migration, new services, corrected architecture - Update services.md: InMemoryOntologyRepository, correct storage layer - Update data-flow.md: Neo4j store, corrected mermaid diagrams - Update pipeline-operator-runbook.md: Neo4j maintenance commands https://claude.ai/code/session_01MshJuJJ15YyUoi2TYXnqRJ
The previous workflow referenced 7 shell scripts in docs/scripts/ that never existed, causing the PR comment step to crash on missing docs/reports/ directory. Replace with inline checks: internal link resolution, Mermaid diagram validation, stale reference detection (unified.db/SQLite repos), and Diataxis directory structure validation. https://claude.ai/code/session_01MshJuJJ15YyUoi2TYXnqRJ
Documentation Quality ReportInternal Link Validation
Result: 885/1445 links valid (61%) Mermaid Diagram ValidationResult: 617/617 diagrams valid (100%) Stale Reference Check
Result: 31 stale reference(s) found Directory Structure ValidationResult: Directory structure valid Quality Score: 13%
Generated by docs-ci |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This branch delivers four major work streams across 12 commits and 74 files (+8,352 / -1,603 lines):
1. Full-Stack Rendering & Physics Hardening
Server-Authoritative Physics (
ce325f8)graph.worker.tsprocessBinaryData()accepts updates for all graph types (removed logseq-only guard)WebGPU TSL Materials (
05dd0e4,ce325f8)activityLeveluniformpositionLocal.yData Velocity Management (
b8782df)queueMicrotask— prevents position frame pileup on fast LANsRenderer Hardening (
cbdf3d0)renderer.init()for Quest 3 Oculus Browser fallbackBi-directional Vircadia Sync (
b8782df,b31b44a)CollaborativeGraphSyncinitializesEntitySyncManager, forwards node_move to Vircadia entity layeronEntityUpdatecallback for server→client reconciliation with 0.1-unit jitter thresholdGraphVircadiaBridgewired toEntitySyncManager+GraphEntityMapperControl Center UX (
40056a4,b31b44a)2. Multi-User Voice-to-Voice Routing (
b92150b)Four-plane audio architecture:
Backend:
AudioRouterwith user-scoped broadcast channels,TurboWhisperSTTProvider, per-agent voice presets,VoiceRoutingSettingsconfigFrontend:
PushToTalkService,LiveKitVoiceService(HRTF spatial panning from Vircadia positions),VoiceOrchestratorInfra:
docker-compose.voice.ymlwith LiveKit SFU, turbo-whisper (faster-whisper CUDA), Kokoro TTSCodec: Opus 48kHz mono end-to-end
New files:
audio_router.rs,LiveKitVoiceService.ts,PushToTalkService.ts,VoiceOrchestrator.ts,docker-compose.voice.yml,config/livekit.yaml3. Ontology-Guided Agent Intelligence (
d856dfe,1bd5dc4)Services (Hexagonal Architecture):
OntologyQueryService— Semantic discovery (keyword → Whelk expansion → relationship fan-out → scoring), enriched note reading, validated Cypher execution with OWL schema hints and Levenshtein typo correctionOntologyMutationService— Proposal pipeline: Logseq markdown generation → OntologyParser round-trip validation → Whelk EL++ consistency check → quality scoring → stagingGitHubPRService— Agent write-back to GitHub: get base SHA → create blob → create tree → create commit → create branch ref → create PR7 MCP Tools (registered in
mcp-server.js):ontology_discover— Semantic keyword search with Whelk inference expansionontology_read— Read enriched note with axioms, relationships, schema contextontology_query— Validated Cypher execution with schema-aware label checkingontology_traverse— BFS graph traversal from starting IRIontology_propose— Create/amend notes with consistency checks + GitHub PRontology_validate— Axiom consistency check against Whelk reasonerontology_status— Service health and statistics7 REST Endpoints (
ontology_agent_handler.rs):POST /api/ontology-agent/{discover,read,query,traverse,propose,validate}GET /api/ontology-agent/statusConfig (
OntologyAgentSettings): auto-merge threshold, confidence gates, per-user note paths, PR labelsTests: 13 integration tests covering discovery, read, Cypher validation, proposals, markdown generation, amendments, quality scoring (
tests/ontology_agent_integration_test.rs)Design Docs: PRD, AFD, DDD in
docs/audit/4. Documentation Overhaul (
4451b45)23 files changed, 1,866 insertions, 885 deletions:
Critical Fixes:
docs/README.md(explanations/→explanation/)unified.db/SQLite references with Neo4j + In-Memory OntologyRepository across 15+ architecture docsSqliteKnowledgeGraphRepository→KnowledgeGraphRepository,SqliteOntologyRepository→OntologyRepositoryUnifiedOntologyRepository→InMemoryOntologyRepositoryComplete Rewrites:
02-project-structure.md: Was describing a Node.js/Vue.js/Express project (!). Now accurately documents Rust/Actix-web backend, React 19 + Three.js frontend, Neo4j, CUDA GPU kernels with full directory treesNew Documentation:
docs/how-to/agents/ontology-agent-tools.md— Complete guide for all 7 MCP tools with input/output schemasdocs/how-to/features/voice-routing.md— 4-plane audio routing architectureASCII → Mermaid Conversions:
services.md: Layered architecture diagramgithub-sync-service-design.md: Initialization and parser flowpipeline-integration.md: Data pipelineontology-pipeline-integration.md: 7-stage reasoning pipelineUpdated: README.md, CHANGELOG.md (v1.2.0), AGENTS.md (capabilities, MCP tools, conventions), system-overview.md, data-flow.md, pipeline-operator-runbook.md
Commits (12)
4451b451bd5dc4d856dfeb92150bb31b44a40056a4cbdf3d0b8782df05dd0e4ce325f8Stats