anantham-v2 2.0.0-alpha.1 → 2.0.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/agents/subagent-handoff-protocol.d.ts +24 -0
- package/dist/agents/subagent-handoff-protocol.d.ts.map +1 -0
- package/dist/agents/subagent-handoff-protocol.js +48 -0
- package/dist/agents/subagent-handoff-protocol.js.map +1 -0
- package/dist/api/gateway-route-pipeline.d.ts +27 -0
- package/dist/api/gateway-route-pipeline.d.ts.map +1 -0
- package/dist/api/gateway-route-pipeline.js +58 -0
- package/dist/api/gateway-route-pipeline.js.map +1 -0
- package/dist/api/openapi-spec-generator.d.ts +23 -0
- package/dist/api/openapi-spec-generator.d.ts.map +1 -0
- package/dist/api/openapi-spec-generator.js +77 -0
- package/dist/api/openapi-spec-generator.js.map +1 -0
- package/dist/api/webhook-dlq-router.d.ts +24 -0
- package/dist/api/webhook-dlq-router.d.ts.map +1 -0
- package/dist/api/webhook-dlq-router.js +62 -0
- package/dist/api/webhook-dlq-router.js.map +1 -0
- package/dist/bin/anantham.js +17 -2
- package/dist/bin/anantham.js.map +1 -1
- package/dist/cli/cli-application.d.ts.map +1 -1
- package/dist/cli/cli-application.js +3 -0
- package/dist/cli/cli-application.js.map +1 -1
- package/dist/cli/command-parser.d.ts +5 -0
- package/dist/cli/command-parser.d.ts.map +1 -1
- package/dist/cli/command-parser.js +40 -9
- package/dist/cli/command-parser.js.map +1 -1
- package/dist/cli/command-registry.d.ts.map +1 -1
- package/dist/cli/command-registry.js +289 -9
- package/dist/cli/command-registry.js.map +1 -1
- package/dist/cli/commands/slash-migrate.d.ts +6 -0
- package/dist/cli/commands/slash-migrate.d.ts.map +1 -0
- package/dist/cli/commands/slash-migrate.js +6 -0
- package/dist/cli/commands/slash-migrate.js.map +1 -0
- package/dist/cli/error-handler.d.ts.map +1 -1
- package/dist/cli/error-handler.js +55 -1
- package/dist/cli/error-handler.js.map +1 -1
- package/dist/cli/interactive-shell.d.ts +27 -0
- package/dist/cli/interactive-shell.d.ts.map +1 -0
- package/dist/cli/interactive-shell.js +99 -0
- package/dist/cli/interactive-shell.js.map +1 -0
- package/dist/cli/keybindings-manager.d.ts +27 -0
- package/dist/cli/keybindings-manager.d.ts.map +1 -0
- package/dist/cli/keybindings-manager.js +43 -0
- package/dist/cli/keybindings-manager.js.map +1 -0
- package/dist/cli/notification-manager.d.ts +28 -0
- package/dist/cli/notification-manager.d.ts.map +1 -0
- package/dist/cli/notification-manager.js +47 -0
- package/dist/cli/notification-manager.js.map +1 -0
- package/dist/cli/shell-integration-generator.d.ts +13 -0
- package/dist/cli/shell-integration-generator.d.ts.map +1 -0
- package/dist/cli/shell-integration-generator.js +79 -0
- package/dist/cli/shell-integration-generator.js.map +1 -0
- package/dist/cli/slash-commands-registry.d.ts +30 -0
- package/dist/cli/slash-commands-registry.d.ts.map +1 -0
- package/dist/cli/slash-commands-registry.js +130 -0
- package/dist/cli/slash-commands-registry.js.map +1 -0
- package/dist/cli/slash-migrate.d.ts +28 -0
- package/dist/cli/slash-migrate.d.ts.map +1 -0
- package/dist/cli/slash-migrate.js +100 -0
- package/dist/cli/slash-migrate.js.map +1 -0
- package/dist/cli/terminal-theme-renderer.d.ts +27 -0
- package/dist/cli/terminal-theme-renderer.d.ts.map +1 -0
- package/dist/cli/terminal-theme-renderer.js +93 -0
- package/dist/cli/terminal-theme-renderer.js.map +1 -0
- package/dist/cli/terminal-title-manager.d.ts +13 -0
- package/dist/cli/terminal-title-manager.d.ts.map +1 -0
- package/dist/cli/terminal-title-manager.js +37 -0
- package/dist/cli/terminal-title-manager.js.map +1 -0
- package/dist/code-intel/code-index-engine.d.ts +31 -0
- package/dist/code-intel/code-index-engine.d.ts.map +1 -0
- package/dist/code-intel/code-index-engine.js +258 -0
- package/dist/code-intel/code-index-engine.js.map +1 -0
- package/dist/code-intel/code-intel-orchestrator.d.ts +28 -0
- package/dist/code-intel/code-intel-orchestrator.d.ts.map +1 -0
- package/dist/code-intel/code-intel-orchestrator.js +45 -0
- package/dist/code-intel/code-intel-orchestrator.js.map +1 -0
- package/dist/code-intel/fault-isolator.d.ts +19 -0
- package/dist/code-intel/fault-isolator.d.ts.map +1 -0
- package/dist/code-intel/fault-isolator.js +147 -0
- package/dist/code-intel/fault-isolator.js.map +1 -0
- package/dist/code-intel/file-watcher-sync.d.ts +29 -0
- package/dist/code-intel/file-watcher-sync.d.ts.map +1 -0
- package/dist/code-intel/file-watcher-sync.js +84 -0
- package/dist/code-intel/file-watcher-sync.js.map +1 -0
- package/dist/code-intel/incremental-indexer.d.ts +30 -0
- package/dist/code-intel/incremental-indexer.d.ts.map +1 -0
- package/dist/code-intel/incremental-indexer.js +90 -0
- package/dist/code-intel/incremental-indexer.js.map +1 -0
- package/dist/code-intel/index.d.ts +13 -0
- package/dist/code-intel/index.d.ts.map +1 -0
- package/dist/code-intel/index.js +13 -0
- package/dist/code-intel/index.js.map +1 -0
- package/dist/code-intel/lsp-bridge.d.ts +43 -0
- package/dist/code-intel/lsp-bridge.d.ts.map +1 -0
- package/dist/code-intel/lsp-bridge.js +103 -0
- package/dist/code-intel/lsp-bridge.js.map +1 -0
- package/dist/code-intel/parsers/multi-language-ast-parser.d.ts +22 -0
- package/dist/code-intel/parsers/multi-language-ast-parser.d.ts.map +1 -0
- package/dist/code-intel/parsers/multi-language-ast-parser.js +536 -0
- package/dist/code-intel/parsers/multi-language-ast-parser.js.map +1 -0
- package/dist/code-intel/project-knowledge-graph.d.ts +36 -0
- package/dist/code-intel/project-knowledge-graph.d.ts.map +1 -0
- package/dist/code-intel/project-knowledge-graph.js +79 -0
- package/dist/code-intel/project-knowledge-graph.js.map +1 -0
- package/dist/code-intel/types.d.ts +92 -0
- package/dist/code-intel/types.d.ts.map +1 -0
- package/dist/code-intel/types.js +6 -0
- package/dist/code-intel/types.js.map +1 -0
- package/dist/context/compaction-loss-evaluator.d.ts +135 -0
- package/dist/context/compaction-loss-evaluator.d.ts.map +1 -0
- package/dist/context/compaction-loss-evaluator.js +232 -0
- package/dist/context/compaction-loss-evaluator.js.map +1 -0
- package/dist/context/index.d.ts +1 -0
- package/dist/context/index.d.ts.map +1 -1
- package/dist/context/index.js +1 -0
- package/dist/context/index.js.map +1 -1
- package/dist/evaluation/ab-testing-engine.d.ts +28 -0
- package/dist/evaluation/ab-testing-engine.d.ts.map +1 -0
- package/dist/evaluation/ab-testing-engine.js +64 -0
- package/dist/evaluation/ab-testing-engine.js.map +1 -0
- package/dist/evaluation/agent-behavior-regression.d.ts +19 -0
- package/dist/evaluation/agent-behavior-regression.d.ts.map +1 -0
- package/dist/evaluation/agent-behavior-regression.js +45 -0
- package/dist/evaluation/agent-behavior-regression.js.map +1 -0
- package/dist/evaluation/anti-contamination-guard.d.ts +19 -0
- package/dist/evaluation/anti-contamination-guard.d.ts.map +1 -0
- package/dist/evaluation/anti-contamination-guard.js +40 -0
- package/dist/evaluation/anti-contamination-guard.js.map +1 -0
- package/dist/evaluation/benchmark-report-generator.d.ts +10 -0
- package/dist/evaluation/benchmark-report-generator.d.ts.map +1 -0
- package/dist/evaluation/benchmark-report-generator.js +57 -0
- package/dist/evaluation/benchmark-report-generator.js.map +1 -0
- package/dist/evaluation/continuous-eval-pipeline.d.ts +27 -0
- package/dist/evaluation/continuous-eval-pipeline.d.ts.map +1 -0
- package/dist/evaluation/continuous-eval-pipeline.js +44 -0
- package/dist/evaluation/continuous-eval-pipeline.js.map +1 -0
- package/dist/evaluation/eval-engine.d.ts +31 -0
- package/dist/evaluation/eval-engine.d.ts.map +1 -0
- package/dist/evaluation/eval-engine.js +57 -0
- package/dist/evaluation/eval-engine.js.map +1 -0
- package/dist/evaluation/ground-truth-comparator.d.ts +13 -0
- package/dist/evaluation/ground-truth-comparator.d.ts.map +1 -0
- package/dist/evaluation/ground-truth-comparator.js +56 -0
- package/dist/evaluation/ground-truth-comparator.js.map +1 -0
- package/dist/evaluation/hitl-review-dashboard.d.ts +25 -0
- package/dist/evaluation/hitl-review-dashboard.d.ts.map +1 -0
- package/dist/evaluation/hitl-review-dashboard.js +40 -0
- package/dist/evaluation/hitl-review-dashboard.js.map +1 -0
- package/dist/evaluation/humaneval-harness.d.ts +21 -0
- package/dist/evaluation/humaneval-harness.d.ts.map +1 -0
- package/dist/evaluation/humaneval-harness.js +47 -0
- package/dist/evaluation/humaneval-harness.js.map +1 -0
- package/dist/evaluation/index.d.ts +14 -0
- package/dist/evaluation/index.d.ts.map +1 -1
- package/dist/evaluation/index.js +14 -0
- package/dist/evaluation/index.js.map +1 -1
- package/dist/evaluation/model-arena.d.ts +33 -0
- package/dist/evaluation/model-arena.d.ts.map +1 -0
- package/dist/evaluation/model-arena.js +60 -0
- package/dist/evaluation/model-arena.js.map +1 -0
- package/dist/evaluation/multi-dimensional-grader.d.ts +22 -0
- package/dist/evaluation/multi-dimensional-grader.d.ts.map +1 -0
- package/dist/evaluation/multi-dimensional-grader.js +38 -0
- package/dist/evaluation/multi-dimensional-grader.js.map +1 -0
- package/dist/evaluation/swe-bench-harness.d.ts +25 -0
- package/dist/evaluation/swe-bench-harness.d.ts.map +1 -0
- package/dist/evaluation/swe-bench-harness.js +55 -0
- package/dist/evaluation/swe-bench-harness.js.map +1 -0
- package/dist/evaluation/synthetic-case-generator.d.ts +10 -0
- package/dist/evaluation/synthetic-case-generator.d.ts.map +1 -0
- package/dist/evaluation/synthetic-case-generator.js +49 -0
- package/dist/evaluation/synthetic-case-generator.js.map +1 -0
- package/dist/evaluation/workflow-profiler.d.ts +30 -0
- package/dist/evaluation/workflow-profiler.d.ts.map +1 -0
- package/dist/evaluation/workflow-profiler.js +47 -0
- package/dist/evaluation/workflow-profiler.js.map +1 -0
- package/dist/event-state/delta-snapshot-manager.d.ts +266 -0
- package/dist/event-state/delta-snapshot-manager.d.ts.map +1 -0
- package/dist/event-state/delta-snapshot-manager.js +446 -0
- package/dist/event-state/delta-snapshot-manager.js.map +1 -0
- package/dist/event-state/index.d.ts +1 -0
- package/dist/event-state/index.d.ts.map +1 -1
- package/dist/event-state/index.js +1 -0
- package/dist/event-state/index.js.map +1 -1
- package/dist/execution/index.d.ts +6 -15
- package/dist/execution/index.d.ts.map +1 -1
- package/dist/execution/index.js +6 -15
- package/dist/execution/index.js.map +1 -1
- package/dist/execution/process-bounds-manager.d.ts +24 -0
- package/dist/execution/process-bounds-manager.d.ts.map +1 -0
- package/dist/execution/process-bounds-manager.js +115 -0
- package/dist/execution/process-bounds-manager.js.map +1 -0
- package/dist/execution/risk-classifier.d.ts +15 -0
- package/dist/execution/risk-classifier.d.ts.map +1 -0
- package/dist/execution/risk-classifier.js +104 -0
- package/dist/execution/risk-classifier.js.map +1 -0
- package/dist/execution/sandbox-manager.d.ts +40 -0
- package/dist/execution/sandbox-manager.d.ts.map +1 -0
- package/dist/execution/sandbox-manager.js +78 -0
- package/dist/execution/sandbox-manager.js.map +1 -0
- package/dist/execution/tool-gateway-orchestrator.d.ts +28 -0
- package/dist/execution/tool-gateway-orchestrator.d.ts.map +1 -0
- package/dist/execution/tool-gateway-orchestrator.js +143 -0
- package/dist/execution/tool-gateway-orchestrator.js.map +1 -0
- package/dist/execution/types.d.ts +47 -0
- package/dist/execution/types.d.ts.map +1 -0
- package/dist/execution/types.js +6 -0
- package/dist/execution/types.js.map +1 -0
- package/dist/governance/air-gapped-policy-enforcer.d.ts +106 -0
- package/dist/governance/air-gapped-policy-enforcer.d.ts.map +1 -0
- package/dist/governance/air-gapped-policy-enforcer.js +243 -0
- package/dist/governance/air-gapped-policy-enforcer.js.map +1 -0
- package/dist/governance/data-retention-manager.d.ts +109 -0
- package/dist/governance/data-retention-manager.d.ts.map +1 -0
- package/dist/governance/data-retention-manager.js +167 -0
- package/dist/governance/data-retention-manager.js.map +1 -0
- package/dist/governance/index.d.ts +3 -0
- package/dist/governance/index.d.ts.map +1 -0
- package/dist/governance/index.js +3 -0
- package/dist/governance/index.js.map +1 -0
- package/dist/index.d.ts +1 -0
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js.map +1 -1
- package/dist/integrations/connector-dlq-router.d.ts +122 -0
- package/dist/integrations/connector-dlq-router.d.ts.map +1 -0
- package/dist/integrations/connector-dlq-router.js +259 -0
- package/dist/integrations/connector-dlq-router.js.map +1 -0
- package/dist/integrations/index.d.ts +1 -0
- package/dist/integrations/index.d.ts.map +1 -1
- package/dist/integrations/index.js +1 -0
- package/dist/integrations/index.js.map +1 -1
- package/dist/memory/index.d.ts +1 -0
- package/dist/memory/index.d.ts.map +1 -1
- package/dist/memory/index.js +1 -0
- package/dist/memory/index.js.map +1 -1
- package/dist/memory/memory-decay-engine.d.ts +103 -0
- package/dist/memory/memory-decay-engine.d.ts.map +1 -0
- package/dist/memory/memory-decay-engine.js +223 -0
- package/dist/memory/memory-decay-engine.js.map +1 -0
- package/dist/multimodal/audio-transcription-engine.d.ts +13 -0
- package/dist/multimodal/audio-transcription-engine.d.ts.map +1 -0
- package/dist/multimodal/audio-transcription-engine.js +32 -0
- package/dist/multimodal/audio-transcription-engine.js.map +1 -0
- package/dist/multimodal/document-pdf-extractor.d.ts +9 -0
- package/dist/multimodal/document-pdf-extractor.d.ts.map +1 -0
- package/dist/multimodal/document-pdf-extractor.js +60 -0
- package/dist/multimodal/document-pdf-extractor.js.map +1 -0
- package/dist/multimodal/image-processor.d.ts +16 -0
- package/dist/multimodal/image-processor.d.ts.map +1 -0
- package/dist/multimodal/image-processor.js +118 -0
- package/dist/multimodal/image-processor.js.map +1 -0
- package/dist/multimodal/index.d.ts +13 -0
- package/dist/multimodal/index.d.ts.map +1 -0
- package/dist/multimodal/index.js +13 -0
- package/dist/multimodal/index.js.map +1 -0
- package/dist/multimodal/multimodal-context-injector.d.ts +21 -0
- package/dist/multimodal/multimodal-context-injector.d.ts.map +1 -0
- package/dist/multimodal/multimodal-context-injector.js +73 -0
- package/dist/multimodal/multimodal-context-injector.js.map +1 -0
- package/dist/multimodal/multimodal-orchestrator.d.ts +29 -0
- package/dist/multimodal/multimodal-orchestrator.d.ts.map +1 -0
- package/dist/multimodal/multimodal-orchestrator.js +51 -0
- package/dist/multimodal/multimodal-orchestrator.js.map +1 -0
- package/dist/multimodal/ocr-pipeline.d.ts +18 -0
- package/dist/multimodal/ocr-pipeline.d.ts.map +1 -0
- package/dist/multimodal/ocr-pipeline.js +43 -0
- package/dist/multimodal/ocr-pipeline.js.map +1 -0
- package/dist/multimodal/perceptual-hasher.d.ts +24 -0
- package/dist/multimodal/perceptual-hasher.d.ts.map +1 -0
- package/dist/multimodal/perceptual-hasher.js +75 -0
- package/dist/multimodal/perceptual-hasher.js.map +1 -0
- package/dist/multimodal/screen-understanding.d.ts +18 -0
- package/dist/multimodal/screen-understanding.d.ts.map +1 -0
- package/dist/multimodal/screen-understanding.js +48 -0
- package/dist/multimodal/screen-understanding.js.map +1 -0
- package/dist/multimodal/types.d.ts +73 -0
- package/dist/multimodal/types.d.ts.map +1 -0
- package/dist/multimodal/types.js +6 -0
- package/dist/multimodal/types.js.map +1 -0
- package/dist/observability/compliance-audit-logger.d.ts +26 -0
- package/dist/observability/compliance-audit-logger.d.ts.map +1 -0
- package/dist/observability/compliance-audit-logger.js +52 -0
- package/dist/observability/compliance-audit-logger.js.map +1 -0
- package/dist/observability/log-streamer-sse.d.ts +28 -0
- package/dist/observability/log-streamer-sse.d.ts.map +1 -0
- package/dist/observability/log-streamer-sse.js +50 -0
- package/dist/observability/log-streamer-sse.js.map +1 -0
- package/dist/observability/metric-collector.d.ts +34 -0
- package/dist/observability/metric-collector.d.ts.map +1 -0
- package/dist/observability/metric-collector.js +60 -0
- package/dist/observability/metric-collector.js.map +1 -0
- package/dist/observability/otel-trace-exporter.d.ts +49 -0
- package/dist/observability/otel-trace-exporter.d.ts.map +1 -0
- package/dist/observability/otel-trace-exporter.js +47 -0
- package/dist/observability/otel-trace-exporter.js.map +1 -0
- package/dist/observability/token-cost-accounting.d.ts +35 -0
- package/dist/observability/token-cost-accounting.d.ts.map +1 -0
- package/dist/observability/token-cost-accounting.js +74 -0
- package/dist/observability/token-cost-accounting.js.map +1 -0
- package/dist/persistence/database-backup-manager.d.ts +342 -0
- package/dist/persistence/database-backup-manager.d.ts.map +1 -0
- package/dist/persistence/database-backup-manager.js +399 -0
- package/dist/persistence/database-backup-manager.js.map +1 -0
- package/dist/persistence/index.d.ts +3 -0
- package/dist/persistence/index.d.ts.map +1 -1
- package/dist/persistence/index.js +3 -0
- package/dist/persistence/index.js.map +1 -1
- package/dist/persistence/migrations/001_initial_core_schema.d.ts.map +1 -1
- package/dist/persistence/migrations/001_initial_core_schema.js +2 -0
- package/dist/persistence/migrations/001_initial_core_schema.js.map +1 -1
- package/dist/persistence/migrations/011_snapshots_and_wal_maintenance.d.ts +8 -0
- package/dist/persistence/migrations/011_snapshots_and_wal_maintenance.d.ts.map +1 -0
- package/dist/persistence/migrations/011_snapshots_and_wal_maintenance.js +129 -0
- package/dist/persistence/migrations/011_snapshots_and_wal_maintenance.js.map +1 -0
- package/dist/persistence/repositories/event-repository.d.ts +5 -0
- package/dist/persistence/repositories/event-repository.d.ts.map +1 -1
- package/dist/persistence/repositories/event-repository.js +21 -0
- package/dist/persistence/repositories/event-repository.js.map +1 -1
- package/dist/persistence/user-config-manager.d.ts +36 -0
- package/dist/persistence/user-config-manager.d.ts.map +1 -0
- package/dist/persistence/user-config-manager.js +141 -0
- package/dist/persistence/user-config-manager.js.map +1 -0
- package/dist/persistence/wal-checkpoint-scheduler.d.ts +51 -0
- package/dist/persistence/wal-checkpoint-scheduler.d.ts.map +1 -0
- package/dist/persistence/wal-checkpoint-scheduler.js +214 -0
- package/dist/persistence/wal-checkpoint-scheduler.js.map +1 -0
- package/dist/recovery/browser-zombie-reclaimer.d.ts +156 -0
- package/dist/recovery/browser-zombie-reclaimer.d.ts.map +1 -0
- package/dist/recovery/browser-zombie-reclaimer.js +225 -0
- package/dist/recovery/browser-zombie-reclaimer.js.map +1 -0
- package/dist/recovery/index.d.ts +1 -0
- package/dist/recovery/index.d.ts.map +1 -1
- package/dist/recovery/index.js +1 -0
- package/dist/recovery/index.js.map +1 -1
- package/dist/resume/external-lease-reconciler.d.ts +154 -0
- package/dist/resume/external-lease-reconciler.d.ts.map +1 -0
- package/dist/resume/external-lease-reconciler.js +260 -0
- package/dist/resume/external-lease-reconciler.js.map +1 -0
- package/dist/resume/fork-point-recovery-manager.d.ts +97 -0
- package/dist/resume/fork-point-recovery-manager.d.ts.map +1 -0
- package/dist/resume/fork-point-recovery-manager.js +275 -0
- package/dist/resume/fork-point-recovery-manager.js.map +1 -0
- package/dist/resume/index.d.ts +4 -0
- package/dist/resume/index.d.ts.map +1 -1
- package/dist/resume/index.js +4 -0
- package/dist/resume/index.js.map +1 -1
- package/dist/resume/model-migration-manager.d.ts +157 -0
- package/dist/resume/model-migration-manager.d.ts.map +1 -0
- package/dist/resume/model-migration-manager.js +203 -0
- package/dist/resume/model-migration-manager.js.map +1 -0
- package/dist/resume/process-reconnect-manager.d.ts +148 -0
- package/dist/resume/process-reconnect-manager.d.ts.map +1 -0
- package/dist/resume/process-reconnect-manager.js +325 -0
- package/dist/resume/process-reconnect-manager.js.map +1 -0
- package/dist/resume/resume-contract.d.ts +12 -12
- package/dist/saas/compliance-report-exporter.d.ts +20 -0
- package/dist/saas/compliance-report-exporter.d.ts.map +1 -0
- package/dist/saas/compliance-report-exporter.js +46 -0
- package/dist/saas/compliance-report-exporter.js.map +1 -0
- package/dist/saas/index.d.ts +10 -0
- package/dist/saas/index.d.ts.map +1 -0
- package/dist/saas/index.js +10 -0
- package/dist/saas/index.js.map +1 -0
- package/dist/saas/quota-rate-limiter.d.ts +25 -0
- package/dist/saas/quota-rate-limiter.d.ts.map +1 -0
- package/dist/saas/quota-rate-limiter.js +53 -0
- package/dist/saas/quota-rate-limiter.js.map +1 -0
- package/dist/saas/rbac-engine.d.ts +16 -0
- package/dist/saas/rbac-engine.d.ts.map +1 -0
- package/dist/saas/rbac-engine.js +45 -0
- package/dist/saas/rbac-engine.js.map +1 -0
- package/dist/saas/sso-authenticator.d.ts +22 -0
- package/dist/saas/sso-authenticator.d.ts.map +1 -0
- package/dist/saas/sso-authenticator.js +60 -0
- package/dist/saas/sso-authenticator.js.map +1 -0
- package/dist/saas/team-workspace-manager.d.ts +26 -0
- package/dist/saas/team-workspace-manager.d.ts.map +1 -0
- package/dist/saas/team-workspace-manager.js +44 -0
- package/dist/saas/team-workspace-manager.js.map +1 -0
- package/dist/saas/tenant-isolation-manager.d.ts +21 -0
- package/dist/saas/tenant-isolation-manager.d.ts.map +1 -0
- package/dist/saas/tenant-isolation-manager.js +30 -0
- package/dist/saas/tenant-isolation-manager.js.map +1 -0
- package/dist/search/index.d.ts +2 -0
- package/dist/search/index.d.ts.map +1 -0
- package/dist/search/index.js +2 -0
- package/dist/search/index.js.map +1 -0
- package/dist/search/universal-content-search-engine.d.ts +185 -0
- package/dist/search/universal-content-search-engine.d.ts.map +1 -0
- package/dist/search/universal-content-search-engine.js +243 -0
- package/dist/search/universal-content-search-engine.js.map +1 -0
- package/dist/security/dangerous-command-heuristics.d.ts +17 -0
- package/dist/security/dangerous-command-heuristics.d.ts.map +1 -0
- package/dist/security/dangerous-command-heuristics.js +71 -0
- package/dist/security/dangerous-command-heuristics.js.map +1 -0
- package/dist/security/policy-hierarchy-resolver.d.ts +35 -0
- package/dist/security/policy-hierarchy-resolver.d.ts.map +1 -0
- package/dist/security/policy-hierarchy-resolver.js +110 -0
- package/dist/security/policy-hierarchy-resolver.js.map +1 -0
- package/dist/security/windows-boundary-guard.d.ts +16 -0
- package/dist/security/windows-boundary-guard.d.ts.map +1 -0
- package/dist/security/windows-boundary-guard.js +64 -0
- package/dist/security/windows-boundary-guard.js.map +1 -0
- package/dist/security/zero-knowledge-enforcer.d.ts +32 -0
- package/dist/security/zero-knowledge-enforcer.d.ts.map +1 -0
- package/dist/security/zero-knowledge-enforcer.js +72 -0
- package/dist/security/zero-knowledge-enforcer.js.map +1 -0
- package/dist/tools/smart-git-generator.d.ts +28 -0
- package/dist/tools/smart-git-generator.d.ts.map +1 -0
- package/dist/tools/smart-git-generator.js +53 -0
- package/dist/tools/smart-git-generator.js.map +1 -0
- package/dist/tui/session-tree-visualizer.d.ts +16 -0
- package/dist/tui/session-tree-visualizer.d.ts.map +1 -0
- package/dist/tui/session-tree-visualizer.js +41 -0
- package/dist/tui/session-tree-visualizer.js.map +1 -0
- package/dist/tui/terminal-layout.d.ts.map +1 -1
- package/dist/tui/terminal-layout.js +12 -4
- package/dist/tui/terminal-layout.js.map +1 -1
- package/dist/tui/tui-application.d.ts +7 -0
- package/dist/tui/tui-application.d.ts.map +1 -1
- package/dist/tui/tui-application.js +110 -5
- package/dist/tui/tui-application.js.map +1 -1
- package/dist/tui/tui-controller.d.ts +77 -2
- package/dist/tui/tui-controller.d.ts.map +1 -1
- package/dist/tui/tui-controller.js +475 -23
- package/dist/tui/tui-controller.js.map +1 -1
- package/dist/tui/tui-dashboard.d.ts +31 -0
- package/dist/tui/tui-dashboard.d.ts.map +1 -0
- package/dist/tui/tui-dashboard.js +74 -0
- package/dist/tui/tui-dashboard.js.map +1 -0
- package/dist/tui/tui-renderer.d.ts +8 -1
- package/dist/tui/tui-renderer.d.ts.map +1 -1
- package/dist/tui/tui-renderer.js +74 -37
- package/dist/tui/tui-renderer.js.map +1 -1
- package/dist/workspace/ecosystem-adapters.d.ts +96 -0
- package/dist/workspace/ecosystem-adapters.d.ts.map +1 -0
- package/dist/workspace/ecosystem-adapters.js +432 -0
- package/dist/workspace/ecosystem-adapters.js.map +1 -0
- package/dist/workspace/ecosystem-importer.d.ts +6 -0
- package/dist/workspace/ecosystem-importer.d.ts.map +1 -0
- package/dist/workspace/ecosystem-importer.js +6 -0
- package/dist/workspace/ecosystem-importer.js.map +1 -0
- package/dist/workspace/index.d.ts +11 -0
- package/dist/workspace/index.d.ts.map +1 -0
- package/dist/workspace/index.js +11 -0
- package/dist/workspace/index.js.map +1 -0
- package/dist/workspace/instruction-loader.d.ts +45 -0
- package/dist/workspace/instruction-loader.d.ts.map +1 -0
- package/dist/workspace/instruction-loader.js +193 -0
- package/dist/workspace/instruction-loader.js.map +1 -0
- package/dist/workspace/monorepo-discovery.d.ts +31 -0
- package/dist/workspace/monorepo-discovery.d.ts.map +1 -0
- package/dist/workspace/monorepo-discovery.js +307 -0
- package/dist/workspace/monorepo-discovery.js.map +1 -0
- package/dist/workspace/multi-root-manager.d.ts +36 -0
- package/dist/workspace/multi-root-manager.d.ts.map +1 -0
- package/dist/workspace/multi-root-manager.js +91 -0
- package/dist/workspace/multi-root-manager.js.map +1 -0
- package/dist/workspace/project-deletion-safety.d.ts +26 -0
- package/dist/workspace/project-deletion-safety.d.ts.map +1 -0
- package/dist/workspace/project-deletion-safety.js +70 -0
- package/dist/workspace/project-deletion-safety.js.map +1 -0
- package/dist/workspace/tech-stack-discovery.d.ts +56 -0
- package/dist/workspace/tech-stack-discovery.d.ts.map +1 -0
- package/dist/workspace/tech-stack-discovery.js +489 -0
- package/dist/workspace/tech-stack-discovery.js.map +1 -0
- package/package.json +1 -1
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"compaction-loss-evaluator.js","sourceRoot":"","sources":["../../src/context/compaction-loss-evaluator.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,CAAC,IAAI,CAAC;IACvC,WAAW;IACX,qBAAqB;IACrB,YAAY;IACZ,UAAU;IACV,cAAc;IACd,kBAAkB;IAClB,oBAAoB;CACrB,CAAC,CAAC;AAGH,MAAM,CAAC,MAAM,oBAAoB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC3C,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACrB,QAAQ,EAAE,kBAAkB;IAC5B,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC9B,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAClC,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC/B,WAAW,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,UAAU,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAC;CAC3D,CAAC,CAAC;AAGH,MAAM,CAAC,MAAM,2BAA2B,GAAG,CAAC,CAAC,MAAM,CAAC;IAClD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC1B,QAAQ,EAAE,kBAAkB;IAC5B,WAAW,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,UAAU,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAC;IAC1D,MAAM,EAAE,CAAC,CAAC,OAAO,EAAE;IACnB,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACpC,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE;CACpB,CAAC,CAAC;AAGH,MAAM,CAAC,MAAM,2BAA2B,GAAG,CAAC,CAAC,MAAM,CAAC;IAClD,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,WAAW,EAAE;IAC3C,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,WAAW,EAAE;IAC5C,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,WAAW,EAAE;IAC5C,iBAAiB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,WAAW,EAAE;IACjD,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACvC,gBAAgB,EAAE,CAAC,CAAC,MAAM,CAAC,kBAAkB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;IACxE,mBAAmB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC7C,0BAA0B,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACpD,oBAAoB,EAAE,CAAC,CAAC,OAAO,EAAE;IACjC,WAAW,EAAE,CAAC,CAAC,KAAK,CAAC,2BAA2B,CAAC;IACjD,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;CAC/B,CAAC,CAAC;AAGH,MAAM,OAAO,2BAA4B,SAAQ,KAAK;IACpC,OAAO,CAAwB;IAE/C,YAAY,OAAe,EAAE,OAA8B;QACzD,KAAK,CAAC,OAAO,CAAC,CAAC;QACf,IAAI,CAAC,IAAI,GAAG,6BAA6B,CAAC;QAC1C,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC;IACzB,CAAC;CACF;AAED,MAAM,OAAO,uBAAuB;IAC1B,mBAAmB,CAAC,QAAgB,EAAE,IAAY;QACxD,MAAM,OAAO,GAAG,QAAQ,CAAC,WAAW,EAAE,CAAC,IAAI,EAAE,CAAC;QAC9C,MAAM,QAAQ,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;QAEpC,IAAI,QAAQ,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;YAC/B,OAAO,GAAG,CAAC;QACb,CAAC;QAED,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QAClE,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,GAAG,CAAC;QAEtC,IAAI,OAAO,GAAG,CAAC,CAAC;QAChB,KAAK,MAAM,IAAI,IAAI,QAAQ,EAAE,CAAC;YAC5B,IAAI,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC;gBAC5B,OAAO,EAAE,CAAC;YACZ,CAAC;QACH,CAAC;QAED,OAAO,OAAO,GAAG,QAAQ,CAAC,MAAM,CAAC;IACnC,CAAC;IAEM,aAAa,CAAC,OAAoB;QACvC,MAAM,MAAM,GAAqB,EAAE,CAAC;QACpC,IAAI,QAAQ,GAAG,CAAC,CAAC;QAEjB,KAAK,MAAM,IAAI,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;YACjC,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,IAAI,IAAI,CAAC,eAAe,IAAI,EAAE,CAAC;YAE3D,IAAI,IAAI,CAAC,QAAQ,KAAK,UAAU,EAAE,CAAC;gBACjC,MAAM,CAAC,IAAI,CAAC;oBACV,EAAE,EAAE,MAAM,GAAG,QAAQ,EAAE;oBACvB,QAAQ,EAAE,WAAW;oBACrB,WAAW,EAAE,+BAA+B;oBAC5C,eAAe,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC;oBACtC,YAAY,EAAE,IAAI,CAAC,EAAE;oBACrB,WAAW,EAAE,UAAU;iBACxB,CAAC,CAAC;YACL,CAAC;YAED,uCAAuC;YACvC,MAAM,eAAe,GAAG,OAAO,CAAC,KAAK,CAAC,oCAAoC,CAAC,CAAC;YAC5E,IAAI,eAAe,EAAE,CAAC;gBACpB,KAAK,MAAM,SAAS,IAAI,eAAe,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,CAAC;oBACpD,MAAM,CAAC,IAAI,CAAC;wBACV,EAAE,EAAE,MAAM,GAAG,QAAQ,EAAE;wBACvB,QAAQ,EAAE,cAAc;wBACxB,WAAW,EAAE,4BAA4B,GAAG,SAAS;wBACrD,eAAe,EAAE,SAAS;wBAC1B,YAAY,EAAE,IAAI,CAAC,EAAE;wBACrB,WAAW,EAAE,IAAI,CAAC,QAAQ,KAAK,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,MAAM;qBAChE,CAAC,CAAC;gBACL,CAAC;YACH,CAAC;YAED,mBAAmB;YACnB,IAAI,uBAAuB,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC;gBAC1C,MAAM,CAAC,IAAI,CAAC;oBACV,EAAE,EAAE,MAAM,GAAG,QAAQ,EAAE;oBACvB,QAAQ,EAAE,kBAAkB;oBAC5B,WAAW,EAAE,gCAAgC;oBAC7C,eAAe,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC;oBACrC,YAAY,EAAE,IAAI,CAAC,EAAE;oBACrB,WAAW,EAAE,MAAM;iBACpB,CAAC,CAAC;YACL,CAAC;YAED,sBAAsB;YACtB,IAAI,IAAI,CAAC,UAAU,KAAK,UAAU,EAAE,CAAC;gBACnC,MAAM,CAAC,IAAI,CAAC;oBACV,EAAE,EAAE,MAAM,GAAG,QAAQ,EAAE;oBACvB,QAAQ,EAAE,oBAAoB;oBAC9B,WAAW,EAAE,0BAA0B;oBACvC,eAAe,EAAE,IAAI,CAAC,QAAQ;oBAC9B,YAAY,EAAE,IAAI,CAAC,EAAE;oBACrB,WAAW,EAAE,MAAM;iBACpB,CAAC,CAAC;YACL,CAAC;YAED,sBAAsB;YACtB,IAAI,4CAA4C,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC;gBAC/D,MAAM,CAAC,IAAI,CAAC;oBACV,EAAE,EAAE,MAAM,GAAG,QAAQ,EAAE;oBACvB,QAAQ,EAAE,UAAU;oBACpB,WAAW,EAAE,8BAA8B;oBAC3C,eAAe,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC;oBACrC,YAAY,EAAE,IAAI,CAAC,EAAE;oBACrB,WAAW,EAAE,MAAM;iBACpB,CAAC,CAAC;YACL,CAAC;QACH,CAAC;QAED,OAAO,MAAM,CAAC;IAChB,CAAC;IAEM,sBAAsB,CAC3B,OAAoB,EACpB,OAA0B,EAC1B,QAAsB;QAEtB,MAAM,MAAM,GAAG,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC;QAE3C,yEAAyE;QACzE,MAAM,iBAAiB,GAAa;YAClC,OAAO,CAAC,SAAS;YACjB,OAAO,CAAC,YAAY;YACpB,GAAG,OAAO,CAAC,KAAK;YAChB,GAAG,OAAO,CAAC,SAAS;YACpB,GAAG,OAAO,CAAC,WAAW;YACtB,GAAG,OAAO,CAAC,UAAU;YACrB,GAAG,OAAO,CAAC,cAAc;YACzB,GAAG,OAAO,CAAC,kBAAkB,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,UAAU,GAAG,GAAG,GAAG,CAAC,CAAC,CAAC,GAAG,IAAI,EAAE,CAAC,GAAG,GAAG,GAAG,CAAC,CAAC,CAAC,WAAW,IAAI,EAAE,CAAC,CAAC,CAAC;SAC7G,CAAC;QAEF,IAAI,QAAQ,EAAE,CAAC;YACb,KAAK,MAAM,EAAE,IAAI,QAAQ,CAAC,KAAK,EAAE,CAAC;gBAChC,IAAI,EAAE,CAAC,OAAO;oBAAE,iBAAiB,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC,CAAC;gBACnD,IAAI,EAAE,CAAC,QAAQ;oBAAE,iBAAiB,CAAC,IAAI,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC;gBACrD,IAAI,EAAE,CAAC,eAAe;oBAAE,iBAAiB,CAAC,IAAI,CAAC,EAAE,CAAC,eAAe,CAAC,CAAC;YACrE,CAAC;QACH,CAAC;QAED,MAAM,kBAAkB,GAAG,iBAAiB,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAEvD,MAAM,WAAW,GAA4B,EAAE,CAAC;QAChD,IAAI,WAAW,GAAG,CAAC,CAAC;QACpB,IAAI,iBAAiB,GAAG,CAAC,CAAC;QAC1B,IAAI,gBAAgB,GAAG,CAAC,CAAC;QACzB,IAAI,WAAW,GAAG,CAAC,CAAC;QAEpB,MAAM,cAAc,GAAyD;YAC3E,SAAS,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE;YAC/B,mBAAmB,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE;YACzC,UAAU,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE;YAChC,QAAQ,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE;YAC9B,YAAY,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE;YAClC,gBAAgB,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE;YACtC,kBAAkB,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE;SACzC,CAAC;QAEF,MAAM,OAAO,GAAG,EAAE,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE,CAAC;QAE5D,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;YAC3B,MAAM,UAAU,GAAG,IAAI,CAAC,mBAAmB,CAAC,KAAK,CAAC,eAAe,EAAE,kBAAkB,CAAC,CAAC;YACvF,MAAM,MAAM,GAAG,UAAU,IAAI,IAAI,CAAC;YAElC,IAAI,MAAM,EAAE,CAAC;gBACX,WAAW,EAAE,CAAC;YAChB,CAAC;iBAAM,IAAI,KAAK,CAAC,WAAW,KAAK,UAAU,EAAE,CAAC;gBAC5C,iBAAiB,EAAE,CAAC;YACtB,CAAC;YAED,MAAM,MAAM,GAAG,OAAO,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC;YAC1C,gBAAgB,IAAI,UAAU,GAAG,MAAM,CAAC;YACxC,WAAW,IAAI,MAAM,CAAC;YAEtB,cAAc,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,KAAK,EAAE,CAAC;YACvC,cAAc,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,GAAG,IAAI,UAAU,CAAC;YAEjD,WAAW,CAAC,IAAI,CAAC;gBACf,OAAO,EAAE,KAAK,CAAC,EAAE;gBACjB,QAAQ,EAAE,KAAK,CAAC,QAAQ;gBACxB,WAAW,EAAE,KAAK,CAAC,WAAW;gBAC9B,MAAM;gBACN,UAAU;gBACV,OAAO,EAAE,MAAM;oBACb,CAAC,CAAC,0BAA0B,GAAG,UAAU,CAAC,OAAO,CAAC,CAAC,CAAC;oBACpD,CAAC,CAAC,uCAAuC,GAAG,KAAK,CAAC,eAAe,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC;aACjF,CAAC,CAAC;QACL,CAAC;QAED,MAAM,aAAa,GAAG,WAAW,GAAG,CAAC,CAAC,CAAC,CAAC,gBAAgB,GAAG,WAAW,CAAC,CAAC,CAAC,GAAG,CAAC;QAC7E,MAAM,oBAAoB,GAAG,iBAAiB,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,IAAI,aAAa,GAAG,IAAI,CAAC,CAAC;QAElG,MAAM,gBAAgB,GAAiC;YACrD,SAAS,EAAE,cAAc,CAAC,SAAS,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC,SAAS,CAAC,GAAG,GAAG,cAAc,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG;YACnH,mBAAmB,EAAE,cAAc,CAAC,mBAAmB,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC,mBAAmB,CAAC,GAAG,GAAG,cAAc,CAAC,mBAAmB,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG;YAC3J,UAAU,EAAE,cAAc,CAAC,UAAU,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC,UAAU,CAAC,GAAG,GAAG,cAAc,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG;YACvH,QAAQ,EAAE,cAAc,CAAC,QAAQ,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC,QAAQ,CAAC,GAAG,GAAG,cAAc,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG;YAC/G,YAAY,EAAE,cAAc,CAAC,YAAY,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC,YAAY,CAAC,GAAG,GAAG,cAAc,CAAC,YAAY,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG;YAC/H,gBAAgB,EAAE,cAAc,CAAC,gBAAgB,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC,gBAAgB,CAAC,GAAG,GAAG,cAAc,CAAC,gBAAgB,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG;YAC/I,kBAAkB,EAAE,cAAc,CAAC,kBAAkB,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC,kBAAkB,CAAC,GAAG,GAAG,cAAc,CAAC,kBAAkB,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG;SACxJ,CAAC;QAEF,MAAM,YAAY,GAAG,OAAO,CAAC,eAAe,CAAC;QAC7C,MAAM,WAAW,GAAG,QAAQ,EAAE,eAAe,IAAI,CAAC,CAAC;QACnD,MAAM,mBAAmB,GAAG,YAAY,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,YAAY,GAAG,WAAW,CAAC,GAAG,YAAY,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QAE5G,OAAO;YACL,WAAW,EAAE,MAAM,CAAC,MAAM;YAC1B,YAAY,EAAE,WAAW;YACzB,YAAY,EAAE,MAAM,CAAC,MAAM,GAAG,WAAW;YACzC,iBAAiB;YACjB,aAAa;YACb,gBAAgB;YAChB,mBAAmB;YACnB,0BAA0B,EAAE,mBAAmB;YAC/C,oBAAoB;YACpB,WAAW;YACX,WAAW,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;SACtC,CAAC;IACJ,CAAC;IAEM,0BAA0B,CAAC,OAA8B,EAAE,cAAsB,IAAI;QAC1F,IAAI,OAAO,CAAC,oBAAoB,EAAE,CAAC;YACjC,MAAM,IAAI,2BAA2B,CACnC,gEAAgE,GAAG,OAAO,CAAC,iBAAiB,GAAG,0BAA0B,EACzH,OAAO,CACR,CAAC;QACJ,CAAC;QACD,IAAI,OAAO,CAAC,aAAa,GAAG,WAAW,EAAE,CAAC;YACxC,MAAM,IAAI,2BAA2B,CACnC,4BAA4B,GAAG,OAAO,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,iCAAiC,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,EAC5H,OAAO,CACR,CAAC;QACJ,CAAC;IACH,CAAC;CACF"}
|
package/dist/context/index.d.ts
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/context/index.ts"],"names":[],"mappings":"AAAA,cAAc,yBAAyB,CAAC;AACxC,cAAc,qBAAqB,CAAC"}
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/context/index.ts"],"names":[],"mappings":"AAAA,cAAc,yBAAyB,CAAC;AACxC,cAAc,qBAAqB,CAAC;AACpC,cAAc,gCAAgC,CAAC"}
|
package/dist/context/index.js
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/context/index.ts"],"names":[],"mappings":"AAAA,cAAc,yBAAyB,CAAC;AACxC,cAAc,qBAAqB,CAAC"}
|
|
1
|
+
{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/context/index.ts"],"names":[],"mappings":"AAAA,cAAc,yBAAyB,CAAC;AACxC,cAAc,qBAAqB,CAAC;AACpC,cAAc,gCAAgC,CAAC"}
|
|
@@ -0,0 +1,28 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* A/B Testing Engine for Prompt Variants & Agent Configurations
|
|
3
|
+
* PRD-PART2-317: A/B Testing Engine for Prompt Variants & Agent Configurations
|
|
4
|
+
*/
|
|
5
|
+
export interface ExperimentVariant {
|
|
6
|
+
id: string;
|
|
7
|
+
name: string;
|
|
8
|
+
weight: number;
|
|
9
|
+
promptTemplate: string;
|
|
10
|
+
config: Record<string, unknown>;
|
|
11
|
+
}
|
|
12
|
+
export interface VariantMetrics {
|
|
13
|
+
impressions: number;
|
|
14
|
+
successes: number;
|
|
15
|
+
failures: number;
|
|
16
|
+
successRate: number;
|
|
17
|
+
}
|
|
18
|
+
export declare class AbTestingEngine {
|
|
19
|
+
private variants;
|
|
20
|
+
private metrics;
|
|
21
|
+
constructor(variants?: ExperimentVariant[]);
|
|
22
|
+
addVariant(variant: ExperimentVariant): void;
|
|
23
|
+
routeRequest(seed?: number): ExperimentVariant;
|
|
24
|
+
recordOutcome(variantId: string, success: boolean): void;
|
|
25
|
+
getVariantMetrics(variantId: string): VariantMetrics | undefined;
|
|
26
|
+
getAllMetrics(): Record<string, VariantMetrics>;
|
|
27
|
+
}
|
|
28
|
+
//# sourceMappingURL=ab-testing-engine.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"ab-testing-engine.d.ts","sourceRoot":"","sources":["../../src/evaluation/ab-testing-engine.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,MAAM,WAAW,iBAAiB;IAChC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,CAAC;IACf,cAAc,EAAE,MAAM,CAAC;IACvB,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACjC;AAED,MAAM,WAAW,cAAc;IAC7B,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;CACrB;AAED,qBAAa,eAAe;IAC1B,OAAO,CAAC,QAAQ,CAA2B;IAC3C,OAAO,CAAC,OAAO,CAA0C;gBAE7C,QAAQ,GAAE,iBAAiB,EAAO;IAMvC,UAAU,CAAC,OAAO,EAAE,iBAAiB,GAAG,IAAI;IAU5C,YAAY,CAAC,IAAI,SAAgB,GAAG,iBAAiB;IAsBrD,aAAa,CAAC,SAAS,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,GAAG,IAAI;IAcxD,iBAAiB,CAAC,SAAS,EAAE,MAAM,GAAG,cAAc,GAAG,SAAS;IAIhE,aAAa,IAAI,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC;CAOvD"}
|
|
@@ -0,0 +1,64 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* A/B Testing Engine for Prompt Variants & Agent Configurations
|
|
3
|
+
* PRD-PART2-317: A/B Testing Engine for Prompt Variants & Agent Configurations
|
|
4
|
+
*/
|
|
5
|
+
export class AbTestingEngine {
|
|
6
|
+
variants = [];
|
|
7
|
+
metrics = new Map();
|
|
8
|
+
constructor(variants = []) {
|
|
9
|
+
for (const v of variants) {
|
|
10
|
+
this.addVariant(v);
|
|
11
|
+
}
|
|
12
|
+
}
|
|
13
|
+
addVariant(variant) {
|
|
14
|
+
this.variants.push(variant);
|
|
15
|
+
this.metrics.set(variant.id, {
|
|
16
|
+
impressions: 0,
|
|
17
|
+
successes: 0,
|
|
18
|
+
failures: 0,
|
|
19
|
+
successRate: 0.0,
|
|
20
|
+
});
|
|
21
|
+
}
|
|
22
|
+
routeRequest(seed = Math.random()) {
|
|
23
|
+
if (this.variants.length === 0) {
|
|
24
|
+
throw new Error("No variants configured in A/B testing engine");
|
|
25
|
+
}
|
|
26
|
+
const totalWeight = this.variants.reduce((acc, v) => acc + v.weight, 0);
|
|
27
|
+
let randomVal = seed * totalWeight;
|
|
28
|
+
for (const v of this.variants) {
|
|
29
|
+
if (randomVal <= v.weight) {
|
|
30
|
+
const m = this.metrics.get(v.id);
|
|
31
|
+
m.impressions++;
|
|
32
|
+
return v;
|
|
33
|
+
}
|
|
34
|
+
randomVal -= v.weight;
|
|
35
|
+
}
|
|
36
|
+
const fallback = this.variants[0];
|
|
37
|
+
this.metrics.get(fallback.id).impressions++;
|
|
38
|
+
return fallback;
|
|
39
|
+
}
|
|
40
|
+
recordOutcome(variantId, success) {
|
|
41
|
+
const m = this.metrics.get(variantId);
|
|
42
|
+
if (!m)
|
|
43
|
+
return;
|
|
44
|
+
if (success) {
|
|
45
|
+
m.successes++;
|
|
46
|
+
}
|
|
47
|
+
else {
|
|
48
|
+
m.failures++;
|
|
49
|
+
}
|
|
50
|
+
const total = m.successes + m.failures;
|
|
51
|
+
m.successRate = total > 0 ? Number((m.successes / total).toFixed(4)) : 0;
|
|
52
|
+
}
|
|
53
|
+
getVariantMetrics(variantId) {
|
|
54
|
+
return this.metrics.get(variantId);
|
|
55
|
+
}
|
|
56
|
+
getAllMetrics() {
|
|
57
|
+
const res = {};
|
|
58
|
+
for (const [id, m] of this.metrics.entries()) {
|
|
59
|
+
res[id] = { ...m };
|
|
60
|
+
}
|
|
61
|
+
return res;
|
|
62
|
+
}
|
|
63
|
+
}
|
|
64
|
+
//# sourceMappingURL=ab-testing-engine.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"ab-testing-engine.js","sourceRoot":"","sources":["../../src/evaluation/ab-testing-engine.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAiBH,MAAM,OAAO,eAAe;IAClB,QAAQ,GAAwB,EAAE,CAAC;IACnC,OAAO,GAAgC,IAAI,GAAG,EAAE,CAAC;IAEzD,YAAY,WAAgC,EAAE;QAC5C,KAAK,MAAM,CAAC,IAAI,QAAQ,EAAE,CAAC;YACzB,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC;QACrB,CAAC;IACH,CAAC;IAEM,UAAU,CAAC,OAA0B;QAC1C,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAC5B,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,EAAE;YAC3B,WAAW,EAAE,CAAC;YACd,SAAS,EAAE,CAAC;YACZ,QAAQ,EAAE,CAAC;YACX,WAAW,EAAE,GAAG;SACjB,CAAC,CAAC;IACL,CAAC;IAEM,YAAY,CAAC,IAAI,GAAG,IAAI,CAAC,MAAM,EAAE;QACtC,IAAI,IAAI,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC/B,MAAM,IAAI,KAAK,CAAC,8CAA8C,CAAC,CAAC;QAClE,CAAC;QAED,MAAM,WAAW,GAAG,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;QACxE,IAAI,SAAS,GAAG,IAAI,GAAG,WAAW,CAAC;QAEnC,KAAK,MAAM,CAAC,IAAI,IAAI,CAAC,QAAQ,EAAE,CAAC;YAC9B,IAAI,SAAS,IAAI,CAAC,CAAC,MAAM,EAAE,CAAC;gBAC1B,MAAM,CAAC,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAE,CAAC;gBAClC,CAAC,CAAC,WAAW,EAAE,CAAC;gBAChB,OAAO,CAAC,CAAC;YACX,CAAC;YACD,SAAS,IAAI,CAAC,CAAC,MAAM,CAAC;QACxB,CAAC;QAED,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAE,CAAC;QACnC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAE,CAAC,WAAW,EAAE,CAAC;QAC7C,OAAO,QAAQ,CAAC;IAClB,CAAC;IAEM,aAAa,CAAC,SAAiB,EAAE,OAAgB;QACtD,MAAM,CAAC,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;QACtC,IAAI,CAAC,CAAC;YAAE,OAAO;QAEf,IAAI,OAAO,EAAE,CAAC;YACZ,CAAC,CAAC,SAAS,EAAE,CAAC;QAChB,CAAC;aAAM,CAAC;YACN,CAAC,CAAC,QAAQ,EAAE,CAAC;QACf,CAAC;QAED,MAAM,KAAK,GAAG,CAAC,CAAC,SAAS,GAAG,CAAC,CAAC,QAAQ,CAAC;QACvC,CAAC,CAAC,WAAW,GAAG,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,SAAS,GAAG,KAAK,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC3E,CAAC;IAEM,iBAAiB,CAAC,SAAiB;QACxC,OAAO,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;IACrC,CAAC;IAEM,aAAa;QAClB,MAAM,GAAG,GAAmC,EAAE,CAAC;QAC/C,KAAK,MAAM,CAAC,EAAE,EAAE,CAAC,CAAC,IAAI,IAAI,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC;YAC7C,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,CAAC;QACrB,CAAC;QACD,OAAO,GAAG,CAAC;IACb,CAAC;CACF"}
|
|
@@ -0,0 +1,19 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Regression Testing Framework for Agent Behaviors & Tool Calling
|
|
3
|
+
* PRD-PART2-313: Regression Testing Framework for Agent Behaviors & Tool Calling
|
|
4
|
+
*/
|
|
5
|
+
export interface ToolCallStep {
|
|
6
|
+
toolName: string;
|
|
7
|
+
action?: string;
|
|
8
|
+
arguments?: Record<string, unknown>;
|
|
9
|
+
}
|
|
10
|
+
export interface TrajectoryComparison {
|
|
11
|
+
passed: boolean;
|
|
12
|
+
expectedStepCount: number;
|
|
13
|
+
actualStepCount: number;
|
|
14
|
+
deviations: string[];
|
|
15
|
+
}
|
|
16
|
+
export declare class AgentBehaviorRegressionTester {
|
|
17
|
+
compareTrajectories(actual: ToolCallStep[], golden: ToolCallStep[], strictOrder?: boolean): TrajectoryComparison;
|
|
18
|
+
}
|
|
19
|
+
//# sourceMappingURL=agent-behavior-regression.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agent-behavior-regression.d.ts","sourceRoot":"","sources":["../../src/evaluation/agent-behavior-regression.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,MAAM,WAAW,YAAY;IAC3B,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACrC;AAED,MAAM,WAAW,oBAAoB;IACnC,MAAM,EAAE,OAAO,CAAC;IAChB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,eAAe,EAAE,MAAM,CAAC;IACxB,UAAU,EAAE,MAAM,EAAE,CAAC;CACtB;AAED,qBAAa,6BAA6B;IACjC,mBAAmB,CACxB,MAAM,EAAE,YAAY,EAAE,EACtB,MAAM,EAAE,YAAY,EAAE,EACtB,WAAW,UAAO,GACjB,oBAAoB;CAsCxB"}
|
|
@@ -0,0 +1,45 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Regression Testing Framework for Agent Behaviors & Tool Calling
|
|
3
|
+
* PRD-PART2-313: Regression Testing Framework for Agent Behaviors & Tool Calling
|
|
4
|
+
*/
|
|
5
|
+
export class AgentBehaviorRegressionTester {
|
|
6
|
+
compareTrajectories(actual, golden, strictOrder = true) {
|
|
7
|
+
const deviations = [];
|
|
8
|
+
if (strictOrder) {
|
|
9
|
+
const maxLen = Math.max(actual.length, golden.length);
|
|
10
|
+
for (let i = 0; i < maxLen; i++) {
|
|
11
|
+
const act = actual[i];
|
|
12
|
+
const gold = golden[i];
|
|
13
|
+
if (!gold && act) {
|
|
14
|
+
deviations.push(`Unexpected extra tool call at step ${i + 1}: '${act.toolName}'`);
|
|
15
|
+
}
|
|
16
|
+
else if (!act && gold) {
|
|
17
|
+
deviations.push(`Missing expected tool call at step ${i + 1}: '${gold.toolName}'`);
|
|
18
|
+
}
|
|
19
|
+
else if (act && gold) {
|
|
20
|
+
if (act.toolName !== gold.toolName) {
|
|
21
|
+
deviations.push(`Step ${i + 1} tool mismatch: expected '${gold.toolName}', got '${act.toolName}'`);
|
|
22
|
+
}
|
|
23
|
+
if (gold.action && act.action !== gold.action) {
|
|
24
|
+
deviations.push(`Step ${i + 1} action mismatch: expected '${gold.action}', got '${act.action}'`);
|
|
25
|
+
}
|
|
26
|
+
}
|
|
27
|
+
}
|
|
28
|
+
}
|
|
29
|
+
else {
|
|
30
|
+
// Non-strict order: check set containment
|
|
31
|
+
for (const gold of golden) {
|
|
32
|
+
if (!actual.some((act) => act.toolName === gold.toolName)) {
|
|
33
|
+
deviations.push(`Missing required tool call: '${gold.toolName}'`);
|
|
34
|
+
}
|
|
35
|
+
}
|
|
36
|
+
}
|
|
37
|
+
return {
|
|
38
|
+
passed: deviations.length === 0,
|
|
39
|
+
expectedStepCount: golden.length,
|
|
40
|
+
actualStepCount: actual.length,
|
|
41
|
+
deviations,
|
|
42
|
+
};
|
|
43
|
+
}
|
|
44
|
+
}
|
|
45
|
+
//# sourceMappingURL=agent-behavior-regression.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agent-behavior-regression.js","sourceRoot":"","sources":["../../src/evaluation/agent-behavior-regression.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAeH,MAAM,OAAO,6BAA6B;IACjC,mBAAmB,CACxB,MAAsB,EACtB,MAAsB,EACtB,WAAW,GAAG,IAAI;QAElB,MAAM,UAAU,GAAa,EAAE,CAAC;QAEhC,IAAI,WAAW,EAAE,CAAC;YAChB,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC;YACtD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBAChC,MAAM,GAAG,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;gBACtB,MAAM,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;gBAEvB,IAAI,CAAC,IAAI,IAAI,GAAG,EAAE,CAAC;oBACjB,UAAU,CAAC,IAAI,CAAC,sCAAsC,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,QAAQ,GAAG,CAAC,CAAC;gBACpF,CAAC;qBAAM,IAAI,CAAC,GAAG,IAAI,IAAI,EAAE,CAAC;oBACxB,UAAU,CAAC,IAAI,CAAC,sCAAsC,CAAC,GAAG,CAAC,MAAM,IAAI,CAAC,QAAQ,GAAG,CAAC,CAAC;gBACrF,CAAC;qBAAM,IAAI,GAAG,IAAI,IAAI,EAAE,CAAC;oBACvB,IAAI,GAAG,CAAC,QAAQ,KAAK,IAAI,CAAC,QAAQ,EAAE,CAAC;wBACnC,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,6BAA6B,IAAI,CAAC,QAAQ,WAAW,GAAG,CAAC,QAAQ,GAAG,CAAC,CAAC;oBACrG,CAAC;oBACD,IAAI,IAAI,CAAC,MAAM,IAAI,GAAG,CAAC,MAAM,KAAK,IAAI,CAAC,MAAM,EAAE,CAAC;wBAC9C,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,+BAA+B,IAAI,CAAC,MAAM,WAAW,GAAG,CAAC,MAAM,GAAG,CAAC,CAAC;oBACnG,CAAC;gBACH,CAAC;YACH,CAAC;QACH,CAAC;aAAM,CAAC;YACN,0CAA0C;YAC1C,KAAK,MAAM,IAAI,IAAI,MAAM,EAAE,CAAC;gBAC1B,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,QAAQ,KAAK,IAAI,CAAC,QAAQ,CAAC,EAAE,CAAC;oBAC1D,UAAU,CAAC,IAAI,CAAC,gCAAgC,IAAI,CAAC,QAAQ,GAAG,CAAC,CAAC;gBACpE,CAAC;YACH,CAAC;QACH,CAAC;QAED,OAAO;YACL,MAAM,EAAE,UAAU,CAAC,MAAM,KAAK,CAAC;YAC/B,iBAAiB,EAAE,MAAM,CAAC,MAAM;YAChC,eAAe,EAAE,MAAM,CAAC,MAAM;YAC9B,UAAU;SACX,CAAC;IACJ,CAAC;CACF"}
|
|
@@ -0,0 +1,19 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Evaluation Integrity & Anti-Contamination Invariants
|
|
3
|
+
* PRD-INV-005: Evaluation Integrity & Anti-Contamination Invariants
|
|
4
|
+
*/
|
|
5
|
+
export interface BenchmarkFingerprint {
|
|
6
|
+
benchmarkName: string;
|
|
7
|
+
testHashes: Set<string>;
|
|
8
|
+
}
|
|
9
|
+
export declare class AntiContaminationGuard {
|
|
10
|
+
private benchmarkFingerprints;
|
|
11
|
+
registerBenchmarkDataset(benchmarkName: string, testSamples: string[]): void;
|
|
12
|
+
detectContamination(benchmarkName: string, promptOrContext: string): {
|
|
13
|
+
contaminated: boolean;
|
|
14
|
+
matchedHash?: string;
|
|
15
|
+
reason?: string;
|
|
16
|
+
};
|
|
17
|
+
private normalizeText;
|
|
18
|
+
}
|
|
19
|
+
//# sourceMappingURL=anti-contamination-guard.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"anti-contamination-guard.d.ts","sourceRoot":"","sources":["../../src/evaluation/anti-contamination-guard.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAIH,MAAM,WAAW,oBAAoB;IACnC,aAAa,EAAE,MAAM,CAAC;IACtB,UAAU,EAAE,GAAG,CAAC,MAAM,CAAC,CAAC;CACzB;AAED,qBAAa,sBAAsB;IACjC,OAAO,CAAC,qBAAqB,CAAgD;IAEtE,wBAAwB,CAAC,aAAa,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,GAAG,IAAI;IAc5E,mBAAmB,CACxB,aAAa,EAAE,MAAM,EACrB,eAAe,EAAE,MAAM,GACtB;QAAE,YAAY,EAAE,OAAO,CAAC;QAAC,WAAW,CAAC,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE;IAoBnE,OAAO,CAAC,aAAa;CAGtB"}
|
|
@@ -0,0 +1,40 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Evaluation Integrity & Anti-Contamination Invariants
|
|
3
|
+
* PRD-INV-005: Evaluation Integrity & Anti-Contamination Invariants
|
|
4
|
+
*/
|
|
5
|
+
import { createHash } from "node:crypto";
|
|
6
|
+
export class AntiContaminationGuard {
|
|
7
|
+
benchmarkFingerprints = new Map();
|
|
8
|
+
registerBenchmarkDataset(benchmarkName, testSamples) {
|
|
9
|
+
const hashes = new Set();
|
|
10
|
+
for (const sample of testSamples) {
|
|
11
|
+
const normalized = this.normalizeText(sample);
|
|
12
|
+
const hash = createHash("sha256").update(normalized).digest("hex");
|
|
13
|
+
hashes.add(hash);
|
|
14
|
+
}
|
|
15
|
+
this.benchmarkFingerprints.set(benchmarkName, {
|
|
16
|
+
benchmarkName,
|
|
17
|
+
testHashes: hashes,
|
|
18
|
+
});
|
|
19
|
+
}
|
|
20
|
+
detectContamination(benchmarkName, promptOrContext) {
|
|
21
|
+
const fp = this.benchmarkFingerprints.get(benchmarkName);
|
|
22
|
+
if (!fp) {
|
|
23
|
+
return { contaminated: false };
|
|
24
|
+
}
|
|
25
|
+
const normPrompt = this.normalizeText(promptOrContext);
|
|
26
|
+
const promptHash = createHash("sha256").update(normPrompt).digest("hex");
|
|
27
|
+
if (fp.testHashes.has(promptHash)) {
|
|
28
|
+
return {
|
|
29
|
+
contaminated: true,
|
|
30
|
+
matchedHash: promptHash,
|
|
31
|
+
reason: `Exact test sample leakage detected in prompt context for benchmark '${benchmarkName}'`,
|
|
32
|
+
};
|
|
33
|
+
}
|
|
34
|
+
return { contaminated: false };
|
|
35
|
+
}
|
|
36
|
+
normalizeText(text) {
|
|
37
|
+
return text.trim().toLowerCase().replace(/\s+/g, " ");
|
|
38
|
+
}
|
|
39
|
+
}
|
|
40
|
+
//# sourceMappingURL=anti-contamination-guard.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"anti-contamination-guard.js","sourceRoot":"","sources":["../../src/evaluation/anti-contamination-guard.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAOzC,MAAM,OAAO,sBAAsB;IACzB,qBAAqB,GAAsC,IAAI,GAAG,EAAE,CAAC;IAEtE,wBAAwB,CAAC,aAAqB,EAAE,WAAqB;QAC1E,MAAM,MAAM,GAAG,IAAI,GAAG,EAAU,CAAC;QACjC,KAAK,MAAM,MAAM,IAAI,WAAW,EAAE,CAAC;YACjC,MAAM,UAAU,GAAG,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,CAAC;YAC9C,MAAM,IAAI,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YACnE,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QACnB,CAAC;QAED,IAAI,CAAC,qBAAqB,CAAC,GAAG,CAAC,aAAa,EAAE;YAC5C,aAAa;YACb,UAAU,EAAE,MAAM;SACnB,CAAC,CAAC;IACL,CAAC;IAEM,mBAAmB,CACxB,aAAqB,EACrB,eAAuB;QAEvB,MAAM,EAAE,GAAG,IAAI,CAAC,qBAAqB,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC;QACzD,IAAI,CAAC,EAAE,EAAE,CAAC;YACR,OAAO,EAAE,YAAY,EAAE,KAAK,EAAE,CAAC;QACjC,CAAC;QAED,MAAM,UAAU,GAAG,IAAI,CAAC,aAAa,CAAC,eAAe,CAAC,CAAC;QACvD,MAAM,UAAU,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QAEzE,IAAI,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC,UAAU,CAAC,EAAE,CAAC;YAClC,OAAO;gBACL,YAAY,EAAE,IAAI;gBAClB,WAAW,EAAE,UAAU;gBACvB,MAAM,EAAE,uEAAuE,aAAa,GAAG;aAChG,CAAC;QACJ,CAAC;QAED,OAAO,EAAE,YAAY,EAAE,KAAK,EAAE,CAAC;IACjC,CAAC;IAEO,aAAa,CAAC,IAAY;QAChC,OAAO,IAAI,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IACxD,CAAC;CACF"}
|
|
@@ -0,0 +1,10 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Automated Benchmark Report Generator (HTML/Markdown Dashboards)
|
|
3
|
+
* PRD-PART2-315: Automated Benchmark Report Generator with HTML/Markdown Dashboards
|
|
4
|
+
*/
|
|
5
|
+
import type { BenchmarkEvaluationReport } from "./eval-engine.js";
|
|
6
|
+
export declare class BenchmarkReportGenerator {
|
|
7
|
+
generateMarkdownReport(report: BenchmarkEvaluationReport): string;
|
|
8
|
+
generateHtmlReport(report: BenchmarkEvaluationReport): string;
|
|
9
|
+
}
|
|
10
|
+
//# sourceMappingURL=benchmark-report-generator.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"benchmark-report-generator.d.ts","sourceRoot":"","sources":["../../src/evaluation/benchmark-report-generator.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EAAE,yBAAyB,EAAE,MAAM,kBAAkB,CAAC;AAElE,qBAAa,wBAAwB;IAC5B,sBAAsB,CAAC,MAAM,EAAE,yBAAyB,GAAG,MAAM;IA2BjE,kBAAkB,CAAC,MAAM,EAAE,yBAAyB,GAAG,MAAM;CA0BrE"}
|
|
@@ -0,0 +1,57 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Automated Benchmark Report Generator (HTML/Markdown Dashboards)
|
|
3
|
+
* PRD-PART2-315: Automated Benchmark Report Generator with HTML/Markdown Dashboards
|
|
4
|
+
*/
|
|
5
|
+
export class BenchmarkReportGenerator {
|
|
6
|
+
generateMarkdownReport(report) {
|
|
7
|
+
const lines = [
|
|
8
|
+
`# Benchmark Evaluation Report: ${report.benchmarkName}`,
|
|
9
|
+
"",
|
|
10
|
+
`| Metric | Value |`,
|
|
11
|
+
`| :--- | :--- |`,
|
|
12
|
+
`| **Total Tests** | ${report.totalTests} |`,
|
|
13
|
+
`| **Passed** | ${report.passedCount} |`,
|
|
14
|
+
`| **Failed** | ${report.failedCount} |`,
|
|
15
|
+
`| **Pass Rate** | ${(report.passRate * 100).toFixed(1)}% |`,
|
|
16
|
+
`| **Avg Latency** | ${report.averageDurationMs} ms |`,
|
|
17
|
+
"",
|
|
18
|
+
"## Test Results Summary",
|
|
19
|
+
"",
|
|
20
|
+
"| Test Case | Status | Latency | Details |",
|
|
21
|
+
"| :--- | :--- | :--- | :--- |",
|
|
22
|
+
];
|
|
23
|
+
for (const r of report.results) {
|
|
24
|
+
const statusIcon = r.passed ? "PASS" : "FAIL";
|
|
25
|
+
const errorMsg = r.error ? `\`${r.error.slice(0, 40)}\`` : "-";
|
|
26
|
+
lines.push(`| \`${r.testCaseId}\` | **${statusIcon}** | ${r.durationMs}ms | ${errorMsg} |`);
|
|
27
|
+
}
|
|
28
|
+
return lines.join("\n");
|
|
29
|
+
}
|
|
30
|
+
generateHtmlReport(report) {
|
|
31
|
+
const passPercentage = (report.passRate * 100).toFixed(1);
|
|
32
|
+
return `<!DOCTYPE html>
|
|
33
|
+
<html>
|
|
34
|
+
<head>
|
|
35
|
+
<meta charset="utf-8">
|
|
36
|
+
<title>Benchmark: ${report.benchmarkName}</title>
|
|
37
|
+
<style>
|
|
38
|
+
body { font-family: system-ui, sans-serif; margin: 2rem; background: #0f172a; color: #f8fafc; }
|
|
39
|
+
.card { background: #1e293b; padding: 1.5rem; border-radius: 8px; margin-bottom: 1.5rem; }
|
|
40
|
+
.metric { font-size: 2rem; font-weight: bold; color: #38bdf8; }
|
|
41
|
+
table { width: 100%; border-collapse: collapse; margin-top: 1rem; }
|
|
42
|
+
th, td { padding: 0.75rem; text-align: left; border-bottom: 1px solid #334155; }
|
|
43
|
+
.pass { color: #4ade80; font-weight: bold; }
|
|
44
|
+
.fail { color: #f87171; font-weight: bold; }
|
|
45
|
+
</style>
|
|
46
|
+
</head>
|
|
47
|
+
<body>
|
|
48
|
+
<h1>Evaluation Report: ${report.benchmarkName}</h1>
|
|
49
|
+
<div class="card">
|
|
50
|
+
<div class="metric">${passPercentage}% Pass Rate</div>
|
|
51
|
+
<p>Total: ${report.totalTests} | Passed: ${report.passedCount} | Failed: ${report.failedCount} | Latency: ${report.averageDurationMs}ms</p>
|
|
52
|
+
</div>
|
|
53
|
+
</body>
|
|
54
|
+
</html>`;
|
|
55
|
+
}
|
|
56
|
+
}
|
|
57
|
+
//# sourceMappingURL=benchmark-report-generator.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"benchmark-report-generator.js","sourceRoot":"","sources":["../../src/evaluation/benchmark-report-generator.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAIH,MAAM,OAAO,wBAAwB;IAC5B,sBAAsB,CAAC,MAAiC;QAC7D,MAAM,KAAK,GAAa;YACtB,kCAAkC,MAAM,CAAC,aAAa,EAAE;YACxD,EAAE;YACF,oBAAoB;YACpB,iBAAiB;YACjB,uBAAuB,MAAM,CAAC,UAAU,IAAI;YAC5C,kBAAkB,MAAM,CAAC,WAAW,IAAI;YACxC,kBAAkB,MAAM,CAAC,WAAW,IAAI;YACxC,qBAAqB,CAAC,MAAM,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK;YAC5D,uBAAuB,MAAM,CAAC,iBAAiB,OAAO;YACtD,EAAE;YACF,yBAAyB;YACzB,EAAE;YACF,4CAA4C;YAC5C,+BAA+B;SAChC,CAAC;QAEF,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;YAC/B,MAAM,UAAU,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC;YAC9C,MAAM,QAAQ,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC;YAC/D,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,UAAU,UAAU,UAAU,QAAQ,CAAC,CAAC,UAAU,QAAQ,QAAQ,IAAI,CAAC,CAAC;QAC9F,CAAC;QAED,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC1B,CAAC;IAEM,kBAAkB,CAAC,MAAiC;QACzD,MAAM,cAAc,GAAG,CAAC,MAAM,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC1D,OAAO;;;;sBAIW,MAAM,CAAC,aAAa;;;;;;;;;;;;2BAYf,MAAM,CAAC,aAAa;;0BAErB,cAAc;gBACxB,MAAM,CAAC,UAAU,cAAc,MAAM,CAAC,WAAW,cAAc,MAAM,CAAC,WAAW,eAAe,MAAM,CAAC,iBAAiB;;;QAGhI,CAAC;IACP,CAAC;CACF"}
|
|
@@ -0,0 +1,27 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Continuous Evaluation & Regression Detection Pipeline
|
|
3
|
+
* PRD-EVAL-006: Continuous Evaluation & Regression Detection Pipeline
|
|
4
|
+
*/
|
|
5
|
+
export interface BenchmarkScoreRecord {
|
|
6
|
+
benchmarkName: string;
|
|
7
|
+
score: number;
|
|
8
|
+
commitHash: string;
|
|
9
|
+
timestamp: string;
|
|
10
|
+
}
|
|
11
|
+
export interface RegressionCheckResult {
|
|
12
|
+
passed: boolean;
|
|
13
|
+
benchmarkName: string;
|
|
14
|
+
currentScore: number;
|
|
15
|
+
baselineScore: number;
|
|
16
|
+
delta: number;
|
|
17
|
+
maxAllowedDrop: number;
|
|
18
|
+
message: string;
|
|
19
|
+
}
|
|
20
|
+
export declare class ContinuousEvalPipeline {
|
|
21
|
+
private baselines;
|
|
22
|
+
private maxAllowedDrop;
|
|
23
|
+
constructor(maxAllowedDrop?: number);
|
|
24
|
+
setBaseline(benchmarkName: string, baselineScore: number): void;
|
|
25
|
+
evaluateRun(benchmarkName: string, currentScore: number): RegressionCheckResult;
|
|
26
|
+
}
|
|
27
|
+
//# sourceMappingURL=continuous-eval-pipeline.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"continuous-eval-pipeline.d.ts","sourceRoot":"","sources":["../../src/evaluation/continuous-eval-pipeline.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,MAAM,WAAW,oBAAoB;IACnC,aAAa,EAAE,MAAM,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,qBAAqB;IACpC,MAAM,EAAE,OAAO,CAAC;IAChB,aAAa,EAAE,MAAM,CAAC;IACtB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,cAAc,EAAE,MAAM,CAAC;IACvB,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,qBAAa,sBAAsB;IACjC,OAAO,CAAC,SAAS,CAAkC;IACnD,OAAO,CAAC,cAAc,CAAS;gBAEnB,cAAc,SAAO;IAI1B,WAAW,CAAC,aAAa,EAAE,MAAM,EAAE,aAAa,EAAE,MAAM,GAAG,IAAI;IAI/D,WAAW,CAAC,aAAa,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,GAAG,qBAAqB;CAgCvF"}
|
|
@@ -0,0 +1,44 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Continuous Evaluation & Regression Detection Pipeline
|
|
3
|
+
* PRD-EVAL-006: Continuous Evaluation & Regression Detection Pipeline
|
|
4
|
+
*/
|
|
5
|
+
export class ContinuousEvalPipeline {
|
|
6
|
+
baselines = new Map();
|
|
7
|
+
maxAllowedDrop; // e.g. 0.02 (2% max drop)
|
|
8
|
+
constructor(maxAllowedDrop = 0.02) {
|
|
9
|
+
this.maxAllowedDrop = maxAllowedDrop;
|
|
10
|
+
}
|
|
11
|
+
setBaseline(benchmarkName, baselineScore) {
|
|
12
|
+
this.baselines.set(benchmarkName, baselineScore);
|
|
13
|
+
}
|
|
14
|
+
evaluateRun(benchmarkName, currentScore) {
|
|
15
|
+
const baseline = this.baselines.get(benchmarkName);
|
|
16
|
+
if (baseline === undefined) {
|
|
17
|
+
// First run becomes the baseline
|
|
18
|
+
this.baselines.set(benchmarkName, currentScore);
|
|
19
|
+
return {
|
|
20
|
+
passed: true,
|
|
21
|
+
benchmarkName,
|
|
22
|
+
currentScore,
|
|
23
|
+
baselineScore: currentScore,
|
|
24
|
+
delta: 0,
|
|
25
|
+
maxAllowedDrop: this.maxAllowedDrop,
|
|
26
|
+
message: `Baseline established at ${currentScore} for '${benchmarkName}'.`,
|
|
27
|
+
};
|
|
28
|
+
}
|
|
29
|
+
const delta = Number((currentScore - baseline).toFixed(4));
|
|
30
|
+
const passed = delta >= -this.maxAllowedDrop;
|
|
31
|
+
return {
|
|
32
|
+
passed,
|
|
33
|
+
benchmarkName,
|
|
34
|
+
currentScore,
|
|
35
|
+
baselineScore: baseline,
|
|
36
|
+
delta,
|
|
37
|
+
maxAllowedDrop: this.maxAllowedDrop,
|
|
38
|
+
message: passed
|
|
39
|
+
? `Score passed regression gate: ${currentScore} vs baseline ${baseline} (delta: ${delta >= 0 ? "+" : ""}${delta})`
|
|
40
|
+
: `CRITICAL: Benchmark regression detected! Current: ${currentScore}, Baseline: ${baseline}, Drop: ${delta} exceeds tolerance -${this.maxAllowedDrop}`,
|
|
41
|
+
};
|
|
42
|
+
}
|
|
43
|
+
}
|
|
44
|
+
//# sourceMappingURL=continuous-eval-pipeline.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"continuous-eval-pipeline.js","sourceRoot":"","sources":["../../src/evaluation/continuous-eval-pipeline.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAmBH,MAAM,OAAO,sBAAsB;IACzB,SAAS,GAAwB,IAAI,GAAG,EAAE,CAAC;IAC3C,cAAc,CAAS,CAAC,0BAA0B;IAE1D,YAAY,cAAc,GAAG,IAAI;QAC/B,IAAI,CAAC,cAAc,GAAG,cAAc,CAAC;IACvC,CAAC;IAEM,WAAW,CAAC,aAAqB,EAAE,aAAqB;QAC7D,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,aAAa,EAAE,aAAa,CAAC,CAAC;IACnD,CAAC;IAEM,WAAW,CAAC,aAAqB,EAAE,YAAoB;QAC5D,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC;QAEnD,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;YAC3B,iCAAiC;YACjC,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,aAAa,EAAE,YAAY,CAAC,CAAC;YAChD,OAAO;gBACL,MAAM,EAAE,IAAI;gBACZ,aAAa;gBACb,YAAY;gBACZ,aAAa,EAAE,YAAY;gBAC3B,KAAK,EAAE,CAAC;gBACR,cAAc,EAAE,IAAI,CAAC,cAAc;gBACnC,OAAO,EAAE,2BAA2B,YAAY,SAAS,aAAa,IAAI;aAC3E,CAAC;QACJ,CAAC;QAED,MAAM,KAAK,GAAG,MAAM,CAAC,CAAC,YAAY,GAAG,QAAQ,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC;QAC3D,MAAM,MAAM,GAAG,KAAK,IAAI,CAAC,IAAI,CAAC,cAAc,CAAC;QAE7C,OAAO;YACL,MAAM;YACN,aAAa;YACb,YAAY;YACZ,aAAa,EAAE,QAAQ;YACvB,KAAK;YACL,cAAc,EAAE,IAAI,CAAC,cAAc;YACnC,OAAO,EAAE,MAAM;gBACb,CAAC,CAAC,iCAAiC,YAAY,gBAAgB,QAAQ,YAAY,KAAK,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,KAAK,GAAG;gBACnH,CAAC,CAAC,qDAAqD,YAAY,eAAe,QAAQ,WAAW,KAAK,uBAAuB,IAAI,CAAC,cAAc,EAAE;SACzJ,CAAC;IACJ,CAAC;CACF"}
|
|
@@ -0,0 +1,31 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Core Evaluation Engine & Test Harness Runner
|
|
3
|
+
* PRD-EVAL-001: Core Evaluation Engine Architecture & Test Harness
|
|
4
|
+
*/
|
|
5
|
+
export interface TestCase {
|
|
6
|
+
id: string;
|
|
7
|
+
input: string;
|
|
8
|
+
expectedOutput?: string;
|
|
9
|
+
validator?: (actual: string) => boolean;
|
|
10
|
+
metadata?: Record<string, unknown>;
|
|
11
|
+
}
|
|
12
|
+
export interface TestCaseResult {
|
|
13
|
+
testCaseId: string;
|
|
14
|
+
passed: boolean;
|
|
15
|
+
durationMs: number;
|
|
16
|
+
actualOutput: string;
|
|
17
|
+
error?: string;
|
|
18
|
+
}
|
|
19
|
+
export interface BenchmarkEvaluationReport {
|
|
20
|
+
benchmarkName: string;
|
|
21
|
+
totalTests: number;
|
|
22
|
+
passedCount: number;
|
|
23
|
+
failedCount: number;
|
|
24
|
+
passRate: number;
|
|
25
|
+
averageDurationMs: number;
|
|
26
|
+
results: TestCaseResult[];
|
|
27
|
+
}
|
|
28
|
+
export declare class CoreEvaluationEngine {
|
|
29
|
+
runBenchmark(benchmarkName: string, testCases: TestCase[], runner: (input: string) => Promise<string> | string): Promise<BenchmarkEvaluationReport>;
|
|
30
|
+
}
|
|
31
|
+
//# sourceMappingURL=eval-engine.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"eval-engine.d.ts","sourceRoot":"","sources":["../../src/evaluation/eval-engine.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,MAAM,WAAW,QAAQ;IACvB,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,SAAS,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,KAAK,OAAO,CAAC;IACxC,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACpC;AAED,MAAM,WAAW,cAAc;IAC7B,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,OAAO,CAAC;IAChB,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,yBAAyB;IACxC,aAAa,EAAE,MAAM,CAAC;IACtB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,OAAO,EAAE,cAAc,EAAE,CAAC;CAC3B;AAED,qBAAa,oBAAoB;IAClB,YAAY,CACvB,aAAa,EAAE,MAAM,EACrB,SAAS,EAAE,QAAQ,EAAE,EACrB,MAAM,EAAE,CAAC,KAAK,EAAE,MAAM,KAAK,OAAO,CAAC,MAAM,CAAC,GAAG,MAAM,GAClD,OAAO,CAAC,yBAAyB,CAAC;CAqDtC"}
|
|
@@ -0,0 +1,57 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Core Evaluation Engine & Test Harness Runner
|
|
3
|
+
* PRD-EVAL-001: Core Evaluation Engine Architecture & Test Harness
|
|
4
|
+
*/
|
|
5
|
+
export class CoreEvaluationEngine {
|
|
6
|
+
async runBenchmark(benchmarkName, testCases, runner) {
|
|
7
|
+
const results = [];
|
|
8
|
+
let totalDuration = 0;
|
|
9
|
+
for (const tc of testCases) {
|
|
10
|
+
const start = Date.now();
|
|
11
|
+
try {
|
|
12
|
+
const actual = await runner(tc.input);
|
|
13
|
+
const durationMs = Date.now() - start;
|
|
14
|
+
totalDuration += durationMs;
|
|
15
|
+
let passed = false;
|
|
16
|
+
if (tc.validator) {
|
|
17
|
+
passed = tc.validator(actual);
|
|
18
|
+
}
|
|
19
|
+
else if (tc.expectedOutput !== undefined) {
|
|
20
|
+
passed = actual.trim() === tc.expectedOutput.trim();
|
|
21
|
+
}
|
|
22
|
+
results.push({
|
|
23
|
+
testCaseId: tc.id,
|
|
24
|
+
passed,
|
|
25
|
+
durationMs,
|
|
26
|
+
actualOutput: actual,
|
|
27
|
+
});
|
|
28
|
+
}
|
|
29
|
+
catch (err) {
|
|
30
|
+
const durationMs = Date.now() - start;
|
|
31
|
+
totalDuration += durationMs;
|
|
32
|
+
const errorMsg = err instanceof Error ? err.message : String(err);
|
|
33
|
+
results.push({
|
|
34
|
+
testCaseId: tc.id,
|
|
35
|
+
passed: false,
|
|
36
|
+
durationMs,
|
|
37
|
+
actualOutput: "",
|
|
38
|
+
error: errorMsg,
|
|
39
|
+
});
|
|
40
|
+
}
|
|
41
|
+
}
|
|
42
|
+
const passedCount = results.filter((r) => r.passed).length;
|
|
43
|
+
const totalTests = results.length;
|
|
44
|
+
const passRate = totalTests > 0 ? Number((passedCount / totalTests).toFixed(4)) : 0;
|
|
45
|
+
const averageDurationMs = totalTests > 0 ? Math.round(totalDuration / totalTests) : 0;
|
|
46
|
+
return {
|
|
47
|
+
benchmarkName,
|
|
48
|
+
totalTests,
|
|
49
|
+
passedCount,
|
|
50
|
+
failedCount: totalTests - passedCount,
|
|
51
|
+
passRate,
|
|
52
|
+
averageDurationMs,
|
|
53
|
+
results,
|
|
54
|
+
};
|
|
55
|
+
}
|
|
56
|
+
}
|
|
57
|
+
//# sourceMappingURL=eval-engine.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"eval-engine.js","sourceRoot":"","sources":["../../src/evaluation/eval-engine.ts"],"names":[],"mappings":"AAAA;;;GAGG;AA4BH,MAAM,OAAO,oBAAoB;IACxB,KAAK,CAAC,YAAY,CACvB,aAAqB,EACrB,SAAqB,EACrB,MAAmD;QAEnD,MAAM,OAAO,GAAqB,EAAE,CAAC;QACrC,IAAI,aAAa,GAAG,CAAC,CAAC;QAEtB,KAAK,MAAM,EAAE,IAAI,SAAS,EAAE,CAAC;YAC3B,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YACzB,IAAI,CAAC;gBACH,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,EAAE,CAAC,KAAK,CAAC,CAAC;gBACtC,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC;gBACtC,aAAa,IAAI,UAAU,CAAC;gBAE5B,IAAI,MAAM,GAAG,KAAK,CAAC;gBACnB,IAAI,EAAE,CAAC,SAAS,EAAE,CAAC;oBACjB,MAAM,GAAG,EAAE,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;gBAChC,CAAC;qBAAM,IAAI,EAAE,CAAC,cAAc,KAAK,SAAS,EAAE,CAAC;oBAC3C,MAAM,GAAG,MAAM,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,cAAc,CAAC,IAAI,EAAE,CAAC;gBACtD,CAAC;gBAED,OAAO,CAAC,IAAI,CAAC;oBACX,UAAU,EAAE,EAAE,CAAC,EAAE;oBACjB,MAAM;oBACN,UAAU;oBACV,YAAY,EAAE,MAAM;iBACrB,CAAC,CAAC;YACL,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC;gBACtC,aAAa,IAAI,UAAU,CAAC;gBAC5B,MAAM,QAAQ,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;gBAClE,OAAO,CAAC,IAAI,CAAC;oBACX,UAAU,EAAE,EAAE,CAAC,EAAE;oBACjB,MAAM,EAAE,KAAK;oBACb,UAAU;oBACV,YAAY,EAAE,EAAE;oBAChB,KAAK,EAAE,QAAQ;iBAChB,CAAC,CAAC;YACL,CAAC;QACH,CAAC;QAED,MAAM,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC;QAC3D,MAAM,UAAU,GAAG,OAAO,CAAC,MAAM,CAAC;QAClC,MAAM,QAAQ,GAAG,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,WAAW,GAAG,UAAU,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QACpF,MAAM,iBAAiB,GAAG,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QAEtF,OAAO;YACL,aAAa;YACb,UAAU;YACV,WAAW;YACX,WAAW,EAAE,UAAU,GAAG,WAAW;YACrC,QAAQ;YACR,iBAAiB;YACjB,OAAO;SACR,CAAC;IACJ,CAAC;CACF"}
|
|
@@ -0,0 +1,13 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Task Completion Verification & Ground Truth Comparator
|
|
3
|
+
* PRD-EVAL-004: Task Completion Verification & Ground Truth Comparator
|
|
4
|
+
*/
|
|
5
|
+
export interface ComparisonResult {
|
|
6
|
+
matched: boolean;
|
|
7
|
+
strategy: "exact" | "normalized" | "json_structural" | "numeric_tolerance";
|
|
8
|
+
diffDetails?: string;
|
|
9
|
+
}
|
|
10
|
+
export declare class GroundTruthComparator {
|
|
11
|
+
compare(actual: string, expected: string, strategy?: "exact" | "normalized" | "json_structural" | "numeric_tolerance", tolerance?: number): ComparisonResult;
|
|
12
|
+
}
|
|
13
|
+
//# sourceMappingURL=ground-truth-comparator.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"ground-truth-comparator.d.ts","sourceRoot":"","sources":["../../src/evaluation/ground-truth-comparator.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,MAAM,WAAW,gBAAgB;IAC/B,OAAO,EAAE,OAAO,CAAC;IACjB,QAAQ,EAAE,OAAO,GAAG,YAAY,GAAG,iBAAiB,GAAG,mBAAmB,CAAC;IAC3E,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAED,qBAAa,qBAAqB;IACzB,OAAO,CACZ,MAAM,EAAE,MAAM,EACd,QAAQ,EAAE,MAAM,EAChB,QAAQ,GAAE,OAAO,GAAG,YAAY,GAAG,iBAAiB,GAAG,mBAAkC,EACzF,SAAS,SAAQ,GAChB,gBAAgB;CAmDpB"}
|
|
@@ -0,0 +1,56 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Task Completion Verification & Ground Truth Comparator
|
|
3
|
+
* PRD-EVAL-004: Task Completion Verification & Ground Truth Comparator
|
|
4
|
+
*/
|
|
5
|
+
export class GroundTruthComparator {
|
|
6
|
+
compare(actual, expected, strategy = "normalized", tolerance = 0.001) {
|
|
7
|
+
switch (strategy) {
|
|
8
|
+
case "exact":
|
|
9
|
+
return {
|
|
10
|
+
matched: actual === expected,
|
|
11
|
+
strategy: "exact",
|
|
12
|
+
diffDetails: actual === expected ? undefined : `Actual length ${actual.length} vs Expected ${expected.length}`,
|
|
13
|
+
};
|
|
14
|
+
case "normalized": {
|
|
15
|
+
const normActual = actual.trim().replace(/\r\n/g, "\n").replace(/[ \t]+/g, " ");
|
|
16
|
+
const normExpected = expected.trim().replace(/\r\n/g, "\n").replace(/[ \t]+/g, " ");
|
|
17
|
+
return {
|
|
18
|
+
matched: normActual === normExpected,
|
|
19
|
+
strategy: "normalized",
|
|
20
|
+
};
|
|
21
|
+
}
|
|
22
|
+
case "json_structural": {
|
|
23
|
+
try {
|
|
24
|
+
const parsedActual = JSON.parse(actual);
|
|
25
|
+
const parsedExpected = JSON.parse(expected);
|
|
26
|
+
const match = JSON.stringify(parsedActual) === JSON.stringify(parsedExpected);
|
|
27
|
+
return {
|
|
28
|
+
matched: match,
|
|
29
|
+
strategy: "json_structural",
|
|
30
|
+
};
|
|
31
|
+
}
|
|
32
|
+
catch (err) {
|
|
33
|
+
return {
|
|
34
|
+
matched: false,
|
|
35
|
+
strategy: "json_structural",
|
|
36
|
+
diffDetails: `JSON parse error: ${err instanceof Error ? err.message : String(err)}`,
|
|
37
|
+
};
|
|
38
|
+
}
|
|
39
|
+
}
|
|
40
|
+
case "numeric_tolerance": {
|
|
41
|
+
const numActual = parseFloat(actual);
|
|
42
|
+
const numExpected = parseFloat(expected);
|
|
43
|
+
if (isNaN(numActual) || isNaN(numExpected)) {
|
|
44
|
+
return { matched: false, strategy: "numeric_tolerance", diffDetails: "NaN encountered" };
|
|
45
|
+
}
|
|
46
|
+
const delta = Math.abs(numActual - numExpected);
|
|
47
|
+
return {
|
|
48
|
+
matched: delta <= tolerance,
|
|
49
|
+
strategy: "numeric_tolerance",
|
|
50
|
+
diffDetails: `Delta: ${delta} (tolerance: ${tolerance})`,
|
|
51
|
+
};
|
|
52
|
+
}
|
|
53
|
+
}
|
|
54
|
+
}
|
|
55
|
+
}
|
|
56
|
+
//# sourceMappingURL=ground-truth-comparator.js.map
|