v1.11.0 - Modeling AI Agent Workspace

Changelog (1.10.4 -> 1.11.0)

Major Highlights

• Added the Modeling AI Agent workspace entry point, provider settings, readiness checks, and guided proposal controls.
• Added live OpenAI/Gemini provider proposal support with hardened JSON parsing, diagnostics, and local fallback proposals.
• Introduced the AI operation contract for controlled modeling mutations across add, move, update, route, delete, catalog, connector layout, and terminal-material workflows.
• Added editable-plan diffing so provider modifiedPlan responses become validated operation proposals before any state mutation.
• Added proposal preview, operation details, impact summaries, explicit apply/reject controls, and last-session rollback support.
• Expanded AI scope handling across active network, current selection, selected harness, and all networks, including network-scoped validation/apply for non-active networks.
• Hardened business validations for route locks, endpoint occupancy, technical-ID conflicts, delete impact, catalog references, and wire sizing recommendations.
• Polished the AI Agent proposal action row with compact Prepare, Apply, Reject, and Rollback controls.
• Added a post-1.11.0 AI Agent follow-up request covering instruction history, persisted panel settings, session review polish, and privacy-safe local reset behavior.
• Covered the AI workflow with focused unit and UI integration tests, plus a live OpenAI smoke test run locally against the configured .env.local key.

Version 1.11.0 - Modeling AI Agent Workspace

AI provider settings

• Added Settings controls for provider selection, API key, model, endpoint, timeout, strict mode, and experimental direct execution.
• Added provider readiness gating so the AI Agent entry is disabled until a valid local provider configuration exists.
• Added connection testing for OpenAI and Gemini without committing or requiring CI secrets.

Assisted proposal workflow

• Added the Modeling AI Agent panel with scoped instruction entry, permission toggles, assisted proposal generation, raw-response inspection, accepted/rejected/unsupported operation reporting, and apply/reject controls.
• Finalized the compact action row labels and ordering as Prepare, Apply, Reject, and Rollback.
• Added local fallback proposals when provider calls fail, keeping the workflow testable without network access.
• Added provider modifiedPlan handling and plan-diff conversion so the preferred response shape can edit a scoped JSON model while the app still applies only validated operations.

Operation contract and application

• Added validated AI operations for adding connectors, splices, nodes, segments, and wires.
• Added movement operations for single entities, relative placement, and batch moves.
• Added safe updates for catalog items, connectors, splices, nodes, segments, and wires.
• Added route operations for regenerating routes and locking wire routes.
• Added delete operations with existing cascade-impact protections.
• Added catalog-item creation, assignment to connectors/splices/wire protection, connector-layout updates, and connector terminal-material overrides.

Scope and safety

• Added AI contexts for active network, current selection, selected harness, and all networks.
• Updated selected-harness scope to use the displayed harness assembly instead of silently defaulting to the first assembly.
• Added explicit target-network support for multi-network AI validation and application, rejecting ambiguous all-network/selected-harness operations unless an existing entity or networkId resolves the target.
• Moved lock_wire_route behind route permission so disabling route mutations blocks both regeneration and manual route locks.

Preview, rollback, and diagnostics

• Added operation detail formatting in the AI panel so users can inspect the proposed mutation list before applying.
• Added impact preview counts by mutation class (add, update, move, route, delete).
• Added explicit AI session snapshots and rollback of the last applied AI session from the panel.
• Improved OpenAI diagnostics and provider JSON parsing to surface actionable errors instead of opaque failures.

Validation coverage

• Added focused tests for AI context building, provider client behavior, operation contract validation, plan diffing, proposal fallback, application, undo/rollback behavior, and UI provider workflows.
• Ran a local live OpenAI smoke test using .env.local to confirm a real provider proposal validates and applies without exposing the API key.

Follow-up planning and workflow polish

• Moved Clear terminal and seal overrides into the connector edit action row next to Cancel edit, matching the edit-action button style while keeping cleanup behavior unchanged.
• Added req_130_ai_agent_follow_up_memory_persistence_and_workflow_polish for instruction history, persisted Agent IA panel preferences, unsent draft preservation, session review, rollback clarity, and privacy/export boundaries.

Validation and Regression Evidence

• rtk npm run -s lint
• rtk npm run -s typecheck
• rtk npm test -- --run src/tests/ai-agent-context.spec.ts src/tests/ai-agent-operation-contract.spec.ts src/tests/ai-agent-apply.spec.ts src/tests/ai-agent-plan-diff.spec.ts src/tests/ai-agent-proposal.spec.ts src/tests/ai-agent-provider-client.spec.ts src/tests/app.ui.settings.spec.tsx src/tests/store.reducer.sync-invariant.spec.ts src/tests/app.ui.network-summary-bom-export.spec.tsx
• rtk npm test -- --run src/tests/app.ui.creation-flow-ergonomics.spec.tsx
• rtk npm run -s build
• rtk python3 -m logics_manager lint --require-status
• Local live provider smoke: rtk npm test -- --run src/tests/ai-agent-live-openai.local.spec.ts (temporary uncommitted test file removed after run)