v0.2.0

0.2.0 (2026-08-27)

Bug Fixes

• agents: make agent cancellation interrupt execution (1f06c50)
• agents: wire AgentExecutor to use AgentLoop for real execution (c9b3f05)
• agents: wire Temporal workflow to execute AgentLoop via activity (1d6c5fd)
• analytics: fix mypy errors in ExperimentComparisonService (b0239f4)
• api: wire replay endpoints to app DI and build explicit response models (b3404a9)
• backend: add explicit retry policies to Temporal workflow activities (a66fcf7)
• backend: configure agent session factory and provider registry for Temporal activities (27ebb1a)
• backend: ensure APP_SECRET_KEY is set in test conftest for CI (d6dd55e)
• backend: register missing Temporal activities in DI container (fb6d4bf)
• backend: remove stale mypy type: ignore comments (6c575c4)
• backend: replace non-deterministic datetime.now() with workflow.now() in evaluation workflow (b1f87e2)
• backend: resolve all mypy strict type errors (28 errors in 6 files) (7494571)
• backend: resolve ruff format and lint errors in test and trajectory files (b364f30)
• ci: normalize Redis stream typing for strict mypy (e23e57b)
• ci: resolve Redis stream typing for strict mypy (3535d1d)
• config: remove dangerous secret defaults, fix lifecycle stop (d80febd)
• db: add agent trajectory migration (db635a4)
• evaluation: enforce tuple contracts in replay report value objects (8c74d4d)
• evaluation: fix mypy errors and ruff lint issues in workflow and activities (c9f356d)
• evaluation: formatting and lint fixes for regression module (a554137)
• evaluation: repair orchestration persistence stage to real repository contract (5405e76)
• evaluation: type redis trace repository and fix set slicing (a9f2b76)
• evaluation: use sa.delete() for mypy-compatible metric cleanup (6ca5097)
• evaluation: wire provider registry and temporal queue (e464753)
• event-bus: use explicit event types instead of wildcard subscriptions (a689d2a)
• event-bus: use per-event database sessions in audit and notification subscribers (2255de4)
• frontend: add .prettierignore to exclude build artifacts from format check (97cd2b0)
• frontend: attach bearer token to API requests and correct SSE paths (809b063)
• frontend: connect dashboard to real analytics data (1d43f07)
• frontend: resolve prettier formatting issues for CI (96b299d)
• frontend: resolve TypeScript errors in run pages (416b01c)
• frontend: wrap useSearchParams in Suspense boundary for /runs/compare (c16747f)
• infrastructure: stabilize application startup and repository tooling (1caeda7)
• metrics: align embedding metrics with real EmbeddingResponse contract (e3f4848)
• otel: fix TracerProvider overwrite bug (d759a98)
• prometheus: add middleware to increment request metrics (54adaa2)
• redteam: add evaluation_source and document INCONCLUSIVE/SCORE semantics (355634d)
• redteam: formatting and lint fixes for semantic judge (93e829f)
• redteam: remove mock data, wire cancel on attack run detail (e28e9e7)
• replay: clean up imports and formatting for B.9.1.1 (346fa42)
• replay: enforce authentication on all trace endpoints (1cb39ba)
• replay: load evaluation traces from persisted database records (cdcf919)
• ruff: fix import ordering in metric_result.py (005991a)
• temporal: add heartbeats to long-running activities (40b8781)
• temporal: enforce workflow execution timeouts (5386a8d)
• temporal: make evaluation item execution retry-safe (e6279da)
• tests: add prompt_injection and jailbreak to score contract canonical inputs (74466ad)
• trajectory: fix TrajectoryEvaluator conversation_history serialization (70ec222)
• types: normalize Redis event bus typing (cc514b5)

Features

• agents: add trajectory persistence layer (bb130cd)
• ai: implement real evaluation pipeline and AI evaluation core (acda741)
• analytics: add experiment comparison, metric distribution, pass/fail summary, and p50 latency (0bb2bf2)
• analytics: add Temporal export workflow and activity for large report exports (f6195b0)
• analytics: implement enterprise analytics and reporting (ba46170)
• api: add experiment and profile REST endpoints (687330c)
• api: add regression analysis endpoint and replay API methods (e75d53d)
• api: wire start_attack_run endpoint to RedTeamWorkflow via Temporal (bbe185e)
• attack-runs: add campaign_results JSON column and persist method (f2ccca7)
• b6: agent trajectory evaluation — domain model, tool execution, metrics, evaluator, integration tests (e4d271d)
• complete enterprise platform, production readiness, and agent runtime (83d74cb)
• composition: wire export workflow into container, services, and API (adb577a)
• composition: wire MetricRegistry into container + startup metric_definitions population (08467f7)
• composition: wire RedTeamWorkflow and activity into DI container (6199131)
• db: add Alembic migration 018 for experiments and evaluation_profiles tables (0672b8b)
• db: add metric_definition_id FK to metric_results + migration 020 + fix integrity activity metric versioning (6c43a23)
• db: add metric_definitions table with Alembic migration 019 (5ef4abe)
• evaluation: add regression analysis domain and CLI (1c812ea)
• evaluation: add reliability, reproducibility & benchmarking layer (B.4) (1990bb4)
• evaluation: complete Sprint 1.1 evaluation CRUD (38f2665)
• evaluation: establish explicit metric score contract (565095a)
• evaluation: execute selected metrics and persist results per item (b67c87f)
• evaluation: harden LLM judges against fabricated scores (7cfade1)
• evaluation: implement end-to-end evaluation pipeline (d3ebe04)
• evaluation: record provider provenance in every metric result (5d4afc8)
• evaluation: register built-in metrics into production MetricEngine (1f019c7)
• evaluation: wire embedding metrics to real provider boundary (1c08626)
• evaluation: wire trace recording, provenance, fingerprint, and threshold verdicts into evaluation workflow (e589cfb)
• evaluators: add evaluator abstraction layer with BaseEvaluatorAdapter, EvaluatorRegistry, and 5 adapter implementations (64f8a07)
• event-bus: implement Phase 5 event subscribers and composition wiring (85c9dd8)
• experiment: add Experiment domain entity, repository, and service (4680e36)
• frontend: add delete actions to evaluations and redteam definitions detail pages (275a2b8)
• frontend: add error boundary component and error.tsx files for main and root layouts (86b56d3)
• frontend: add experiment management pages (list/create/detail with edit and delete) (844af57)
• frontend: add missing API client functions and types for experiments, profiles, datasets, auth refresh, and analytics extras (9946144)
• frontend: add profile management pages (list/create/detail with edit and delete) (174babf)
• frontend: add regression analysis component (d99dbc2)
• frontend: add replay viewer component for run details (e1ccb8c)
• frontend: add responsive mobile sidebar with hamburger menu (93c277c)
• frontend: add run comparison page (b774ef0)
• frontend: add settings layout with provider credentials and team management pages (8639cd8)
• frontend: add skeleton loader components (Skeleton, ListSkeleton, TableSkeleton, CardSkeleton) (bc04b60)
• frontend: add sonner toast notification system with Toaster in root layout (2f4289d)
• frontend: complete agent runtime interface and React 19 compatibility (17439da)
• frontend: complete Phase 5 application foundation (9727d02)
• frontend: enhance run list and details with verdict, cancel, retry (f3e2f43)
• frontend: replace hardcoded projects page with real API integration and add create page (3144991)
• frontend: replace hardcoded reports page with real API integration for dashboard summary, safety trends, and generated reports (0035afb)
• frontend: replace simulated datasets page with functional file parser, preview table, and JSON export (36417e5)
• metrics: add CompositeMetric base class with AverageCompositeMetric and WeightedCompositeMetric (2d30709)
• metrics: add evaluator_type and required_inputs to all 26 metric implementations (2610dfc)
• metrics: add EvaluatorType enum and extend MetricDefinition with required_inputs, evaluator_type, plugin_module (a1df036)
• metrics: add MetricRegistry with importlib.metadata entry_points discovery (0159f6f)
• metrics: add PromptInjectionMetric and JailbreakMetric (e048812)
• observability: implement Phase 3 live monitoring (5b78215)
• plugins: add metric plugin template with example CustomAccuracyMetric (b9770f4)
• profile: add EvaluationProfile domain entity, repository, and service (63eeaa5)
• providers: implement real EmbeddingProvider boundary on OpenAIProvider (b1849ce)
• redteam: add FindingDetected and CampaignCompleted domain events (0db2a2b)
• redteam: add semantic effectiveness judge for attack evaluation (710ec13)
• redteam: add Temporal workflow and activity for red team campaigns (83ae964)
• redteam: implement B.5 adaptive red team campaign engine (da768f0)
• redteam: implement Phase 4 safety engine (b8882dc)
• runtime: integrate Temporal execution pipeline (b0916a5)
• tests: add tests for MetricRegistry, evaluator adapters, and composite metrics (8b05396)

Performance Improvements

• evaluation: reuse embeddings within a single item boundary (ff9d91c)