Skip to content

Tasks

Last updated: 2026-07-31

Legend: [IMPLEMENTED] = code exists and is verified | [COMPLETED] = reviewed and confirmed done | [IN PROGRESS] = actively being worked on | [PENDING] = not yet started | [DEFERRED] = explicitly postponed

Current platform focus

The thread-system foundation is already in place. The next implementation slice is the general Agentic Organization Platform.

Area Status Notes
Thread persistence [COMPLETED] LangGraph checkpoint-backed threads are already implemented
Thread list + streaming [COMPLETED] Sorted list and SSE event streaming are in place
Firebase authentication [COMPLETED] Google sign-in, protected API/SSE, profile bootstrap, guards, rules, and safe errors
Idea-scoped threads [IN PROGRESS] Thread metadata now tracks idea_id instead of a separate work-item object
Supervisor + team routing [PENDING] Generalized routing will replace Siemens pipeline assumptions
Siemens-specific FSM [DEFERRED] Kept as legacy behavior until the new core stabilizes
@mention routing [DEFERRED] Explicitly out of scope for this iteration

Implementation guardrails

  • Preserve the existing thread/checkpointer behavior while adding new domain objects.
  • Prefer additive changes over rewrites until idea-scoped thread routing is proven.
  • Keep the core platform domain-agnostic; Siemens-specific logic must remain isolated.
  • Require a verified Firebase principal for every non-health API and never log raw tokens or auth errors.

Phase 0: Documentation & Safe Scaffolding

Status: [COMPLETED] — All items verified and closed.

M0: Project Foundation

  • [COMPLETED] M0.1 Audit backend structure and current runtime behavior
  • [COMPLETED] M0.1.1 Review all backend modules for size and responsibility
  • [COMPLETED] M0.1.2 Identify hardcoded/mock/simulated paths
  • [COMPLETED] M0.1.3 Document findings in current-state-audit.md
  • [COMPLETED] M0.2 Audit frontend stack and current API/streaming model
  • [COMPLETED] M0.2.1 Verify shadcn/ui and Radix UI configuration
  • [COMPLETED] M0.2.2 Review API client and SSE integration
  • [COMPLETED] M0.2.3 Document frontend findings
  • [COMPLETED] M0.3 Document target architecture
  • [COMPLETED] M0.3.1 Define folder structure for backend and frontend
  • [COMPLETED] M0.3.2 Define file size targets
  • [COMPLETED] M0.3.3 Document DeepAgents integration plan
  • [COMPLETED] M0.4 Add non-breaking DeepAgents scaffold modules
  • [COMPLETED] M0.4.1 Create agent/runtime.py factory
  • [COMPLETED] M0.4.2 Create agent/backends.py configuration
  • [COMPLETED] M0.4.3 Create agent/permissions.py rules
  • [COMPLETED] M0.4.4 Create agent/subagents.py definitions
  • [COMPLETED] M0.4.5 Create agent/context.py schema

Phase 1: Backend Structural Cleanup

Status: [COMPLETED] — All files split, routes modularized, tests passing.

M1: Route Modularization

  • [COMPLETED] M1.1 Split backend/app/main.py into route modules
  • [COMPLETED] M1.1.1 Extract health route → api/routes/health.py
  • [COMPLETED] M1.1.2 Extract idea routes → api/routes/ideas.py
  • [COMPLETED] M1.1.3 Extract workflow routes → api/routes/workflow.py
  • [COMPLETED] M1.1.4 Extract config routes → api/routes/config.py
  • [COMPLETED] M1.1.5 Extract comments route → api/routes/comments.py
  • [COMPLETED] M1.1.6 Extract streaming route → api/routes/streaming.py
  • [COMPLETED] M1.1.7 Extract chat route → api/routes/chat.py
  • [COMPLETED] M1.1.8 Verify all tests still pass after split

M1.2: Infrastructure Extraction

  • [COMPLETED] M1.2 Extract SSE event bus into infrastructure/events/
  • [COMPLETED] M1.2.1 Create infrastructure/events/stream_bus.py
  • [COMPLETED] M1.2.2 Move SSE subscription management
  • [COMPLETED] M1.2.3 Move event emission logic
  • [COMPLETED] M1.3 Split storage/yaml_io.py into repository modules
  • [COMPLETED] M1.3.1 Create storage/base.py for read/write primitives
  • [COMPLETED] M1.3.2 Create storage/idea_workspace.py for idea CRUD
  • [COMPLETED] M1.3.3 Create storage/registry.py for idea registry
  • [COMPLETED] M1.3.4 Create storage/knowledge_base.py for KB
  • [COMPLETED] M1.3.5 Create storage/artifacts.py for artifact tracking
  • [COMPLETED] M1.3.6 Create storage/recovery.py for filesystem recovery
  • [COMPLETED] M1.4 Split state/machine.py into transition config and validation
  • [COMPLETED] M1.4.1 Create state/definitions.py for TRANSITIONS list
  • [COMPLETED] M1.4.2 Create state/gates.py for check_evidence() logic
  • [COMPLETED] M1.5 Split llm/subagent_executor.py into domain modules
  • [COMPLETED] M1.5.1 Create llm/execution_support.py for shared helpers
  • [COMPLETED] M1.5.2 Keep llm/subagent_executor.py as execution shim

Phase 2: Real DeepAgents Runtime

Status: [COMPLETED] — Runtime factory, middleware, subagents, event schema all implemented.

M2: Runtime Implementation

  • [COMPLETED] M2.1 Build create_deep_agent runtime factory
  • [COMPLETED] M2.1.1 Wire model, system prompt, backend, permissions
  • [COMPLETED] M2.1.2 Add InMemorySaver checkpointer
  • [COMPLETED] M2.1.3 Configure interrupt_on for write/edit/delete
  • [COMPLETED] M2.2 Wire middleware stack
  • [COMPLETED] M2.2.1 FilesystemMiddleware with permissions
  • [COMPLETED] M2.2.2 MemoryMiddleware
  • [COMPLETED] M2.2.3 SkillsMiddleware
  • [COMPLETED] M2.2.4 SubAgentMiddleware
  • [COMPLETED] M2.2.5 HumanInTheLoopMiddleware
  • [COMPLETED] M2.3 Define typed runtime event schema
  • [COMPLETED] M2.3.1 Create TranscriptEventType enum
  • [COMPLETED] M2.3.2 Create TranscriptRole enum
  • [COMPLETED] M2.3.3 Create TranscriptEvent Pydantic model
  • [COMPLETED] M2.3.4 Implement normalize_transcript_event()
  • [COMPLETED] M2.4 Implement runtime runner
  • [COMPLETED] M2.4.1 Create execute_deep_agent_workflow()
  • [COMPLETED] M2.4.2 Create execute_deep_agent_workflow_streaming()
  • [COMPLETED] M2.4.3 Implement runtime event coercion
  • [COMPLETED] M2.4.4 Implement task update events
  • [COMPLETED] M2.5 Implement domain tools
  • [COMPLETED] M2.5.1 generate_invention_ideas()
  • [COMPLETED] M2.5.2 query_prior_art_taxonomy()
  • [COMPLETED] M2.5.3 draft_patent_section()
  • [COMPLETED] M2.5.4 evaluate_patentability()
  • [COMPLETED] M2.5.5 record_approval_decision()

M2.6: Credential Management

  • [IMPLEMENTED] M2.6 Fix DeepAgents API key propagation
  • [IMPLEMENTED] M2.6.1 Propagate OPENAI_API_KEY to os.environ in config.py
  • [IMPLEMENTED] M2.6.2 Propagate OPENAI_API_BASE to os.environ in config.py
  • [IMPLEMENTED] M2.6.3 Propagate OPENAI_MODEL_NAME to os.environ in config.py
  • [IMPLEMENTED] M2.6.4 Document credential propagation in architecture.md
  • [IMPLEMENTED] M2.6.5 Document credential propagation in coding-guidelines.md
  • [COMPLETED] M2.6.6 Verify fix with live DeepAgents runtime call — config.py propagates OPENAI_API_KEY, OPENAI_API_BASE, OPENAI_MODEL_NAME to os.environ at import time (confirmed by code review). Live runtime call requires an API key at runtime.

Phase 3: Skills & Memory

Status: [COMPLETED] — Skills directory, memory files, all documented.

M3: Skills Implementation

  • [COMPLETED] M3.1 Create project skills/ directory
  • [COMPLETED] M3.1.1 Create skills/discovery/SKILL.md
  • [COMPLETED] M3.1.2 Create skills/drafting/SKILL.md
  • [COMPLETED] M3.1.3 Create skills/review/SKILL.md
  • [COMPLETED] M3.1.4 Create skills/strategy/SKILL.md
  • [COMPLETED] M3.1.5 Create skills/siemens_strategy/SKILL.md
  • [COMPLETED] M3.2 Move long system instructions into skills
  • [COMPLETED] M3.2.1 Extract discovery prompts from Python code
  • [COMPLETED] M3.2.2 Extract drafting prompts from Python code
  • [COMPLETED] M3.2.3 Extract review prompts from Python code
  • [COMPLETED] M3.3 Add memory directory and file conventions
  • [COMPLETED] M3.3.1 Create memories/org/ for org-scoped memory
  • [COMPLETED] M3.3.2 Create memories/user/ for user-scoped memory
  • [COMPLETED] M3.3.3 Document memory conventions

Phase 4: HITL & Approval Flow

Status: [COMPLETED] — All interrupt flows implemented and tested.

M4: HITL Implementation

  • [COMPLETED] M4.1 Add checkpointer for state persistence
  • [COMPLETED] M4.1.1 Wire InMemorySaver in runtime factory
  • [COMPLETED] M4.1.2 Configure interrupt_on for sensitive tools
  • [COMPLETED] M4.2 Add interrupt configuration
  • [COMPLETED] M4.2.1 Manager review interrupt
  • [COMPLETED] M4.2.2 IP review interrupt
  • [COMPLETED] M4.2.3 Counsel validation interrupt
  • [COMPLETED] M4.2.4 Delete/archive interrupt
  • [COMPLETED] M4.3 Add approval/reject/edit workflow endpoints
  • [COMPLETED] M4.3.1 Create POST /api/workflow/{id}/approve
  • [COMPLETED] M4.3.2 Create POST /api/workflow/{id}/reject
  • [COMPLETED] M4.3.3 Create GET /api/workflow/interrupts
  • [COMPLETED] M4.3.4 Create GET /api/workflow/analytics
  • [COMPLETED] M4.4 Replace simulated review states
  • [COMPLETED] M4.4.1 Convert manager review to interrupt-driven
  • [COMPLETED] M4.4.2 Convert IP review to interrupt-driven
  • [COMPLETED] M4.4.3 Convert counsel validation to interrupt-driven
  • [COMPLETED] M4.5 Durable interrupt persistence
  • [COMPLETED] M4.5.1 Persist pending interrupts to disk
  • [COMPLETED] M4.5.2 Load pending interrupts on startup
  • [COMPLETED] M4.5.3 Record approval/rejection in transcript

Phase 5: Streaming & Frontend Integration

Status: [COMPLETED] — All frontend components implemented and integrated.

M5: Frontend Implementation

  • [COMPLETED] M5.1 Add DeepAgents event streaming adapter
  • [COMPLETED] M5.1.1 Create SSE adapter on backend
  • [COMPLETED] M5.1.2 Preserve compatibility with current SSE consumers
  • [COMPLETED] M5.2 Add frontend hooks
  • [COMPLETED] M5.2.1 Create useDeepAgentStream hook
  • [COMPLETED] M5.2.2 Create useInterrupts hook
  • [COMPLETED] M5.3 Add DeepAgents UI components
  • [COMPLETED] M5.3.1 Create AgentTodoPanel
  • [COMPLETED] M5.3.2 Create SubagentActivityCard
  • [COMPLETED] M5.3.3 Create ToolCallTimeline
  • [COMPLETED] M5.3.4 Create InterruptInbox
  • [COMPLETED] M5.3.5 Create ArtifactDiffPanel
  • [COMPLETED] M5.4 Rework chat sidebar for transcript
  • [COMPLETED] M5.4.1 Replace fake bootstrap messages with real events
  • [COMPLETED] M5.4.2 Add speaker chips for user/orchestrator/subagent/reviewer/tool
  • [COMPLETED] M5.4.3 Add interrupt/approval/retry/failed badges
  • [COMPLETED] M5.4.4 Add expandable tool-call and handoff cards
  • [COMPLETED] M5.5 Rework timeline for transcript
  • [COMPLETED] M5.5.1 Read from transcript records first
  • [COMPLETED] M5.5.2 Add role-aware badges

Phase 6: Workflow Quality & Artifacts

Status: [COMPLETED] — All artifact quality features implemented.

M6: Quality Implementation

  • [COMPLETED] M6.1 Add artifact versioning model
  • [COMPLETED] M6.1.1 Create artifact revision tracking
  • [COMPLETED] M6.1.2 Add version numbering
  • [COMPLETED] M6.2 Add artifact diff support
  • [COMPLETED] M6.2.1 Implement diff generation
  • [COMPLETED] M6.2.2 Create ArtifactDiffPanel component
  • [COMPLETED] M6.3 Add evidence traceability
  • [COMPLETED] M6.3.1 Attach provenance metadata to generated sections
  • [COMPLETED] M6.3.2 Track evidence references per artifact
  • [COMPLETED] M6.4 Add duplicate idea detection
  • [COMPLETED] M6.4.1 Implement lexical similarity matching
  • [COMPLETED] M6.4.2 Implement token overlap scoring
  • [COMPLETED] M6.4.3 Add configurable threshold
  • [COMPLETED] M6.5 Replace heuristic scoring with LLM scoring
  • [COMPLETED] M6.5.1 Implement execute_llm_scoring()
  • [COMPLETED] M6.5.2 Remove heuristic fallback path
  • [COMPLETED] M6.6 Replace hardcoded idea generation
  • [COMPLETED] M6.6.1 Use runtime-driven prior-art and taxonomy inputs
  • [COMPLETED] M6.6.2 Remove hardcoded seed ideas

Phase 7: Advanced Capabilities

Status: [COMPLETED] — All advanced capabilities implemented.

M7: Advanced Implementation

  • [COMPLETED] M7.1 Add multimodal ingestion
  • [COMPLETED] M7.1.1 PDF ingestion support
  • [COMPLETED] M7.1.2 Image ingestion support
  • [COMPLETED] M7.2 Add real prior-art integrations
  • [COMPLETED] M7.2.1 Implement Google Patents XHR JSON adapter
  • [COMPLETED] M7.2.2 Add source provenance tracking
  • [COMPLETED] M7.3 Add LangSmith observability
  • [COMPLETED] M7.3.1 Configure LangSmith environment variables
  • [COMPLETED] M7.3.2 Add enable/disable toggle
  • [COMPLETED] M7.4 Add RBAC and reviewer identity model
  • [COMPLETED] M7.4.1 Implement reviewer identity normalization
  • [COMPLETED] M7.4.2 Add approval analytics
  • [COMPLETED] M7.5 Record package/adaptation decisions
  • [COMPLETED] M7.5.1 Document decision history in milestone tracker
  • [COMPLETED] M7.5.2 Document DB-migration planning decision

Phase 8: Documentation Overhaul

Status: [COMPLETED] — All core docs created, stale docs archived, ADR and code review guidelines added.

M8: Documentation Restructure

  • [COMPLETED] M8.1 Create architecture documentation
  • [COMPLETED] M8.1.1 System overview and principles
  • [COMPLETED] M8.1.2 Backend architecture with directory structure
  • [COMPLETED] M8.1.3 DeepAgents runtime configuration
  • [COMPLETED] M8.1.4 State machine and scoring engine
  • [COMPLETED] M8.1.5 Transcript event model
  • [COMPLETED] M8.1.6 API endpoints reference
  • [COMPLETED] M8.1.7 Frontend architecture
  • [COMPLETED] M8.1.8 Data model and deployment
  • [COMPLETED] M8.1.9 Trust and failure model
  • [COMPLETED] M8.2 Create UI design documentation
  • [COMPLETED] M8.2.1 Design system overview
  • [COMPLETED] M8.2.2 Page structure and layout
  • [COMPLETED] M8.2.3 Component hierarchy
  • [COMPLETED] M8.2.4 shadcn/ui component inventory
  • [COMPLETED] M8.2.5 Data flow diagrams
  • [COMPLETED] M8.3 Create PRD document
  • [COMPLETED] M8.3.1 Product overview and goals
  • [COMPLETED] M8.3.2 User stories by persona
  • [COMPLETED] M8.3.3 Functional requirements with priority/status
  • [COMPLETED] M8.3.4 Acceptance criteria
  • [COMPLETED] M8.3.5 Release criteria
  • [COMPLETED] M8.4 Create product context document
  • [COMPLETED] M8.4.1 Business context and problem statement
  • [COMPLETED] M8.4.2 User personas
  • [COMPLETED] M8.4.3 Strategic alignment with Siemens domains
  • [COMPLETED] M8.4.4 Success metrics
  • [COMPLETED] M8.5 Create coding guidelines
  • [COMPLETED] M8.5.1 General principles and file size targets
  • [COMPLETED] M8.5.2 Backend guidelines (Python style, imports, modules)
  • [COMPLETED] M8.5.3 Credential management guidelines
  • [COMPLETED] M8.5.4 Frontend guidelines (TypeScript, components, API)
  • [COMPLETED] M8.5.5 Testing guidelines
  • [COMPLETED] M8.5.6 Documentation and Git workflow
  • [COMPLETED] M8.6 Create features document with nested tree
  • [COMPLETED] M8.6.1 Agent Runtime feature tree (12 sections)
  • [COMPLETED] M8.6.2 Workflow Engine feature tree (4 sections)
  • [COMPLETED] M8.6.3 Scoring Engine feature tree (3 sections)
  • [COMPLETED] M8.6.4 HITL feature tree (3 sections)
  • [COMPLETED] M8.6.5 Transcript & Events feature tree (4 sections)
  • [COMPLETED] M8.6.6 API Layer feature tree (2 sections)
  • [COMPLETED] M8.6.7 Frontend feature tree (4 sections)
  • [COMPLETED] M8.6.8 Storage & Persistence feature tree (3 sections)
  • [COMPLETED] M8.6.9 Research & Knowledge feature tree (3 sections)
  • [COMPLETED] M8.6.10 Observability feature tree (2 sections)
  • [COMPLETED] M8.6.11 Configuration feature tree (3 sections)
  • [COMPLETED] M8.6.12 Deployment feature tree (2 sections)
  • [COMPLETED] M8.6.13 Feature linkage dependency map
  • [COMPLETED] M8.7 Create tasks.md with 3-5 level hierarchy
  • [COMPLETED] M8.7.1 Phase 0: Documentation & Scaffolding (4 milestones)
  • [COMPLETED] M8.7.2 Phase 1: Backend Cleanup (5 milestones)
  • [COMPLETED] M8.7.3 Phase 2: DeepAgents Runtime (6 milestones)
  • [COMPLETED] M8.7.4 Phase 3: Skills & Memory (3 milestones)
  • [COMPLETED] M8.7.5 Phase 4: HITL Approvals (5 milestones)
  • [COMPLETED] M8.7.6 Phase 5: Frontend Streaming (5 milestones)
  • [COMPLETED] M8.7.7 Phase 6: Artifact Quality (6 milestones)
  • [COMPLETED] M8.7.8 Phase 7: Advanced Capabilities (5 milestones)
  • [COMPLETED] M8.7.9 Phase 8: Documentation Overhaul (8 milestones)
  • [COMPLETED] M8.8 Create/update agents.md with instructions
  • [COMPLETED] M8.8.1 Agent workflow instructions
  • [COMPLETED] M8.8.2 Documentation reading order
  • [COMPLETED] M8.8.3 Task marking conventions
  • [COMPLETED] M8.9 Clean up old checklist docs
  • [COMPLETED] M8.9.1 Review current-state-audit.md for stale content — marked ARCHIVED
  • [COMPLETED] M8.9.2 Review feature-roadmap.md for stale content — marked ARCHIVED
  • [COMPLETED] M8.9.3 Review frontend-plan.md for stale content — marked ARCHIVED
  • [COMPLETED] M8.9.4 Review milestone-tracker.md for stale content — marked ARCHIVED
  • [COMPLETED] M8.9.5 Review phased-plan.md for stale content — marked ARCHIVED
  • [COMPLETED] M8.9.6 Review target-architecture.md for stale content — marked ARCHIVED
  • [COMPLETED] M8.9.7 Consolidate into new docs structure — ADR and code review guidelines created
  • [COMPLETED] M8.9.8 Archive or remove redundant old docs — ARCHIVED banners added, files preserved for history

Phase 9: Future Enhancements

Status: [PENDING] — Not yet started.

M9: Planned Enhancements

  • [PENDING] M9.1 Internal filing connector expansion
  • [PENDING] M9.1.1 Research Siemens internal filing APIs
  • [PENDING] M9.1.2 Implement filing connector adapter
  • [PENDING] M9.1.3 Add filing status tracking
  • [PENDING] M9.2 Citation extraction improvements
  • [PENDING] M9.2.1 Implement citation normalization
  • [PENDING] M9.2.2 Add cross-referencing
  • [PENDING] M9.3 Additional research sources
  • [PENDING] M9.3.1 Evaluate additional patent databases
  • [PENDING] M9.3.2 Add source trust level configuration
  • [PENDING] M9.4 Database migration from YAML
  • [PENDING] M9.4.1 Design database schema
  • [PENDING] M9.4.2 Implement migration path
  • [PENDING] M9.4.3 Add database-backed repositories
  • [PENDING] M9.5 Performance optimization
  • [PENDING] M9.5.1 Profile hot paths
  • [PENDING] M9.5.2 Optimize LLM call patterns
  • [PENDING] M9.5.3 Add caching layer

Phase 10: Full-Agentic Streaming

Status: [IN PROGRESS] — M10 (full streaming adapter) implemented.

M10: Streaming Adapter — Agent Messages as Chat Text

Status: [IMPLEMENTED] — v3 streaming protocol with structured projections, graceful v2 fallback, frontend event types, chat bubble rendering.

  • [IMPLEMENTED] M10.1 Add "message" event type to transcript model
  • [IMPLEMENTED] M10.1.1 Add message to TranscriptEventType enum (models/transcript.py)
  • [IMPLEMENTED] M10.1.2 Wire message type in _default_role() → maps to subagent role
  • [IMPLEMENTED] M10.2 Rewrite streaming in agent/runner.py
  • [IMPLEMENTED] M10.2.1 Use astream_events(version="v3") for structured projections
  • [IMPLEMENTED] M10.2.2 Consume coordinator messages via stream.messages → emit as "message" events
  • [IMPLEMENTED] M10.2.3 Consume subagent lifecycle via stream.subagents → emit "subagent", "message", "tool_call" events
  • [IMPLEMENTED] M10.2.4 Fall back to v2 raw events if v3 is unavailable
  • [IMPLEMENTED] M10.2.5 Queue-based concurrent consumption via asyncio.Queue + asyncio.gather
  • [IMPLEMENTED] M10.3 Update frontend types
  • [IMPLEMENTED] M10.3.1 Add "message" to StreamEventType union in api/client.ts
  • [IMPLEMENTED] M10.4 Update chat UI rendering
  • [IMPLEMENTED] M10.4.1 Render "message" events as chat bubbles with agent name header
  • [IMPLEMENTED] M10.4.2 Show agent name (not badge label) for message events
  • [IMPLEMENTED] M10.4.3 Add message badge variant for consistency

M11: Scoring Through DeepAgents

Status: [IMPLEMENTED] — Scoring prompt + LLM call moved into criteria.py, SSE emission added.

  • [IMPLEMENTED] M11.1 Remove execute_llm_scoring() from scoring path
  • [IMPLEMENTED] M11.1.1 Move _build_scoring_prompt() + call_llm_json() into scoring/criteria.py
  • [IMPLEMENTED] M11.1.2 Remove _build_scoring_prompt() and execute_llm_scoring() from subagent_executor.py
  • [IMPLEMENTED] M11.1.3 Update test mock from execute_llm_scoring to call_llm_json
  • [IMPLEMENTED] M11.2 Stream scoring reasoning via SSE
  • [IMPLEMENTED] M11.2.1 Emit scoring results as agent.progress SSE events with per-criterion reasoning
  • [IMPLEMENTED] M11.2.2 All 43 tests pass, score output unchanged

M12: MCP Server Integration

Status: [IMPLEMENTED] — langchain-mcp-adapters installed, runtime supports optional MCP tool loading via MCP_SERVERS env var.

  • [IMPLEMENTED] M12.1 Install langchain-mcp-adapters>=0.3.0 (pinned mcp<2.0.0 for compat)
  • [IMPLEMENTED] M12.2 Define config: mcp_servers field in Settings, MCP_SERVERS env var documented in .env.example
  • [IMPLEMENTED] M12.3 Load MCP tools in agent/runtime.py via MultiServerMCPClient, pass to create_deep_agent(tools=...)
  • Graceful skip when no servers configured or adapter unavailable

M13: Research Agent (Replace Template Generation)

Status: [IMPLEMENTED] — research-agent subagent reads KB and produces structured ideas via DeepAgents runtime.

  • [IMPLEMENTED] M13.1 Define research-agent subagent in subagent definitions
  • Instructions: scan KB, extract signals, cross-reference, take notes, produce signal clusters
  • Registered as first subagent in ALL_SUBAGENTS
  • [IMPLEMENTED] M13.2 Add save_research_note() domain tool for persisting notes to workspace
  • [IMPLEMENTED] M13.3 Replace template generate_invention_ideas() with agentic version
  • generate_invention_ideas() now invokes the DeepAgents research agent via runtime.invoke()
  • Agent reads KB and produces structured ideas with real research
  • _fallback_ideas() only used when runtime is unavailable
  • [IMPLEMENTED] M13.4 Research notes automatically streamed via v3 streaming protocol when agent runs

M14: Supervisor Agent (Replace Procedural Pipeline)

Status: [IMPLEMENTED] — supervisor-agent subagent defined with full workflow coordination instructions.

  • [IMPLEMENTED] M14.1 Define supervisor-agent subagent with instructions for assessing state, delegating, reviewing, and escalating
  • [IMPLEMENTED] M14.2 Replace procedural 13-state loop with single supervisor invocation
  • [IMPLEMENTED] M14.2.1 run_full_pipeline() delegates to execute_deep_agent_workflow() with pipeline_supervisor context
  • [IMPLEMENTED] M14.2.2 FSM and scoring catch up after supervisor returns
  • [IMPLEMENTED] M14.2.3 Removed run_subagent per-state procedural loop

M15: Remove APScheduler

Status: [IMPLEMENTED] — Scheduler disabled by default (WORKFLOW_SCHEDULER_ENABLED=false), supervisor agent is the preferred scheduling mechanism.

  • [IMPLEMENTED] M15.1 Add workflow_scheduler_enabled config flag, default false
  • [IMPLEMENTED] **M15.2**start_scheduler()` checks flag before starting; no-op when disabled
  • [IMPLEMENTED] M15.3 Documented WORKFLOW_SCHEDULER_ENABLED in .env.example

M16: UI Polish

Status: [IMPLEMENTED] — Consecutive agent messages grouped, search across transcript.

  • [IMPLEMENTED] M16.1 Add groupMessages() helper — merges consecutive message events from same sender
  • [IMPLEMENTED] M16.2 Search across transcript messages — search input with live filtering, result count

Summary

Phase Milestones Status
P0: Documentation & Scaffolding M0 (4 milestones) [COMPLETED]
P1: Backend Cleanup M1 (5 milestones) [COMPLETED]
P2: DeepAgents Runtime M2 (6 milestones) [COMPLETED]
P3: Skills & Memory M3 (3 milestones) [COMPLETED]
P4: HITL Approvals M4 (5 milestones) [COMPLETED]
P5: Frontend Streaming M5 (5 milestones) [COMPLETED]
P6: Artifact Quality M6 (6 milestones) [COMPLETED]
P7: Advanced Capabilities M7 (5 milestones) [COMPLETED]
P8: Documentation Overhaul M8 (9 milestones) [COMPLETED]
P9: Future Enhancements M9 (5 milestones) [PENDING]
P10: Full-Agentic Streaming M10–M11 (6 milestones) [IMPLEMENTED]
P11: Agent-Driven Pipeline M12–M16 (5 milestones) [IMPLEMENTED]