You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Configuration & Policy Authority: Consolidated configuration authority, single model selection authority, canonical provider registry, and explicit timeout/resource policies in m31a::config::canonical.
Sandbox Mode Vocabulary: Added canonical SandboxMode enum (Strict, Standard, Permissive, Disabled) with ordering semantics and TUI setup wizard mode cycling.
Model Invocations & Routing: Added structured InvocationOptions, model eligibility verification across TUI components, command handlers, and routed callers.
Task Graph Retrieval: Added latest active TaskGraph retrieval per workspace to prevent task graph snapshot leakage.
Evaluation Improvements: Added execution mode tracking in evaluation scenarios, scorecards, and cost tracking.
Fixed
Prompt Catalog Partitioning: Pruned deprecated prompt metadata and aligned workflow engine suites (genesis_runtime_integration, golden_project_genesis_workflow, workflow_engine, workflow_engine_failure, workflow_engine_recovery, workflow_execution_spine) with active canonical prompt catalog templates.
Agent Convergence Invariant: Preserved canonical autonomous objective fallback in AgentEngine::compile_context during interactive turns.
Database Migrations & Storage: Added channel-aware legacy path detection and fallback for database migration bootstrap and error recovery.
Cross-Platform Compatibility: Fixed installer script verification and test harnesses on Windows and macOS.
TUI Model Autocomplete: Reconciled model autocomplete suggestions with runtime catalog inventory without synthetic mock fallbacks.
Hermetic TUI Testing: Hermetically seeded cached model inventory in tui_autocomplete_repair integration test harness.