@evalops/maestro 0.10.18 → 0.10.20
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/README.md +56 -1
- package/dist/agent/action-approval.d.ts +12 -2
- package/dist/agent/action-approval.d.ts.map +1 -1
- package/dist/agent/action-approval.js +19 -32
- package/dist/agent/action-approval.js.map +1 -1
- package/dist/agent/agent.d.ts +1 -0
- package/dist/agent/agent.d.ts.map +1 -1
- package/dist/agent/agent.js +26 -0
- package/dist/agent/agent.js.map +1 -1
- package/dist/agent/compaction.d.ts +2 -0
- package/dist/agent/compaction.d.ts.map +1 -1
- package/dist/agent/compaction.js +4 -2
- package/dist/agent/compaction.js.map +1 -1
- package/dist/agent/index.d.ts +1 -1
- package/dist/agent/index.d.ts.map +1 -1
- package/dist/agent/index.js +1 -1
- package/dist/agent/index.js.map +1 -1
- package/dist/agent/modes.d.ts +56 -4
- package/dist/agent/modes.d.ts.map +1 -1
- package/dist/agent/modes.js +163 -2
- package/dist/agent/modes.js.map +1 -1
- package/dist/agent/prompt-recovery.d.ts.map +1 -1
- package/dist/agent/prompt-recovery.js +1 -3
- package/dist/agent/prompt-recovery.js.map +1 -1
- package/dist/agent/providers/anthropic.d.ts.map +1 -1
- package/dist/agent/providers/anthropic.js +9 -1
- package/dist/agent/providers/anthropic.js.map +1 -1
- package/dist/agent/providers/codex-app-server.d.ts +12 -0
- package/dist/agent/providers/codex-app-server.d.ts.map +1 -0
- package/dist/agent/providers/codex-app-server.js +788 -0
- package/dist/agent/providers/codex-app-server.js.map +1 -0
- package/dist/agent/providers/google-gemini-cli.js +14 -3
- package/dist/agent/providers/google.js +22 -3
- package/dist/agent/providers/scripted.d.ts +11 -0
- package/dist/agent/providers/scripted.d.ts.map +1 -0
- package/dist/agent/providers/scripted.js +399 -0
- package/dist/agent/providers/scripted.js.map +1 -0
- package/dist/agent/providers/vertex.js +14 -3
- package/dist/agent/scenario-source.d.ts +19 -0
- package/dist/agent/scenario-source.d.ts.map +1 -0
- package/dist/agent/scenario-source.js +112 -0
- package/dist/agent/scenario-source.js.map +1 -0
- package/dist/agent/subagent-specs.d.ts +1 -1
- package/dist/agent/subagent-specs.js +1 -1
- package/dist/agent/subagent-specs.js.map +1 -1
- package/dist/agent/swarm/executor.d.ts +25 -0
- package/dist/agent/swarm/executor.d.ts.map +1 -1
- package/dist/agent/swarm/executor.js +795 -20
- package/dist/agent/swarm/executor.js.map +1 -1
- package/dist/agent/swarm/runtime-events.d.ts +11 -0
- package/dist/agent/swarm/runtime-events.d.ts.map +1 -0
- package/dist/agent/swarm/runtime-events.js +26 -0
- package/dist/agent/swarm/runtime-events.js.map +1 -0
- package/dist/agent/swarm/types.d.ts +78 -0
- package/dist/agent/swarm/types.d.ts.map +1 -1
- package/dist/agent/swarm/types.js.map +1 -1
- package/dist/agent/tracing.d.ts +1 -1
- package/dist/agent/tracing.d.ts.map +1 -1
- package/dist/agent/tracing.js +17 -9
- package/dist/agent/tracing.js.map +1 -1
- package/dist/agent/transport/create-provider-stream.d.ts.map +1 -1
- package/dist/agent/transport/create-provider-stream.js +14 -0
- package/dist/agent/transport/create-provider-stream.js.map +1 -1
- package/dist/agent/transport/stable-stringify.d.ts +2 -0
- package/dist/agent/transport/stable-stringify.d.ts.map +1 -0
- package/dist/agent/transport/stable-stringify.js +14 -0
- package/dist/agent/transport/stable-stringify.js.map +1 -0
- package/dist/agent/transport/tool-execution-bridge.d.ts.map +1 -1
- package/dist/agent/transport/tool-execution-bridge.js +2 -15
- package/dist/agent/transport/tool-execution-bridge.js.map +1 -1
- package/dist/agent/transport/tool-safety-pipeline.d.ts +5 -1
- package/dist/agent/transport/tool-safety-pipeline.d.ts.map +1 -1
- package/dist/agent/transport/tool-safety-pipeline.js +184 -36
- package/dist/agent/transport/tool-safety-pipeline.js.map +1 -1
- package/dist/agent/transport.d.ts +8 -0
- package/dist/agent/transport.d.ts.map +1 -1
- package/dist/agent/transport.js +1613 -68
- package/dist/agent/transport.js.map +1 -1
- package/dist/agent/types.d.ts +170 -2
- package/dist/agent/types.d.ts.map +1 -1
- package/dist/agent/types.js.map +1 -1
- package/dist/api/enterprise-routes.d.ts.map +1 -1
- package/dist/api/enterprise-routes.js +34 -3
- package/dist/api/enterprise-routes.js.map +1 -1
- package/dist/app-server/in-process-client.d.ts +32 -0
- package/dist/app-server/in-process-client.d.ts.map +1 -0
- package/dist/app-server/in-process-client.js +73 -0
- package/dist/app-server/in-process-client.js.map +1 -0
- package/dist/app-server/session-api.d.ts +36 -0
- package/dist/app-server/session-api.d.ts.map +1 -0
- package/dist/app-server/session-api.js +713 -0
- package/dist/app-server/session-api.js.map +1 -0
- package/dist/approvals/platform-action-approval.d.ts +3 -1
- package/dist/approvals/platform-action-approval.d.ts.map +1 -1
- package/dist/approvals/platform-action-approval.js +16 -9
- package/dist/approvals/platform-action-approval.js.map +1 -1
- package/dist/approvals/service-client.d.ts.map +1 -1
- package/dist/approvals/service-client.js +7 -7
- package/dist/approvals/service-client.js.map +1 -1
- package/dist/audit/logger.d.ts +4 -4
- package/dist/bootstrap/agent-creation-setup.d.ts +4 -0
- package/dist/bootstrap/agent-creation-setup.d.ts.map +1 -1
- package/dist/bootstrap/agent-creation-setup.js +3 -1
- package/dist/bootstrap/agent-creation-setup.js.map +1 -1
- package/dist/bootstrap/auth-setup.js +1 -1
- package/dist/bootstrap/auth-setup.js.map +1 -1
- package/dist/bootstrap/event-subscriptions-setup.d.ts +6 -0
- package/dist/bootstrap/event-subscriptions-setup.d.ts.map +1 -1
- package/dist/bootstrap/event-subscriptions-setup.js +26 -2
- package/dist/bootstrap/event-subscriptions-setup.js.map +1 -1
- package/dist/bootstrap/model-resolution-setup.d.ts.map +1 -1
- package/dist/bootstrap/model-resolution-setup.js +4 -1
- package/dist/bootstrap/model-resolution-setup.js.map +1 -1
- package/dist/bootstrap/tools-setup.d.ts +2 -1
- package/dist/bootstrap/tools-setup.d.ts.map +1 -1
- package/dist/bootstrap/tools-setup.js +5 -1
- package/dist/bootstrap/tools-setup.js.map +1 -1
- package/dist/cli/args.d.ts +9 -1
- package/dist/cli/args.d.ts.map +1 -1
- package/dist/cli/args.js +130 -5
- package/dist/cli/args.js.map +1 -1
- package/dist/cli/commands/a2a.d.ts +9 -0
- package/dist/cli/commands/a2a.d.ts.map +1 -0
- package/dist/cli/commands/a2a.js +1714 -0
- package/dist/cli/commands/a2a.js.map +1 -0
- package/dist/cli/commands/agents.d.ts +16 -2
- package/dist/cli/commands/agents.d.ts.map +1 -1
- package/dist/cli/commands/agents.js +223 -35
- package/dist/cli/commands/agents.js.map +1 -1
- package/dist/cli/commands/codex.d.ts +1 -1
- package/dist/cli/commands/codex.d.ts.map +1 -1
- package/dist/cli/commands/codex.js +145 -39
- package/dist/cli/commands/codex.js.map +1 -1
- package/dist/cli/commands/context.d.ts +10 -0
- package/dist/cli/commands/context.d.ts.map +1 -0
- package/dist/cli/commands/context.js +176 -0
- package/dist/cli/commands/context.js.map +1 -0
- package/dist/cli/commands/hosted-runner.d.ts +1 -0
- package/dist/cli/commands/hosted-runner.d.ts.map +1 -1
- package/dist/cli/commands/hosted-runner.js +9 -1
- package/dist/cli/commands/hosted-runner.js.map +1 -1
- package/dist/cli/commands/modes.d.ts +8 -0
- package/dist/cli/commands/modes.d.ts.map +1 -0
- package/dist/cli/commands/modes.js +151 -0
- package/dist/cli/commands/modes.js.map +1 -0
- package/dist/cli/commands/operating-plane.d.ts +14 -0
- package/dist/cli/commands/operating-plane.d.ts.map +1 -0
- package/dist/cli/commands/operating-plane.js +160 -0
- package/dist/cli/commands/operating-plane.js.map +1 -0
- package/dist/cli/commands/run.d.ts +87 -0
- package/dist/cli/commands/run.d.ts.map +1 -0
- package/dist/cli/commands/run.js +356 -0
- package/dist/cli/commands/run.js.map +1 -0
- package/dist/cli/commands/scenario.d.ts +7 -0
- package/dist/cli/commands/scenario.d.ts.map +1 -0
- package/dist/cli/commands/scenario.js +195 -0
- package/dist/cli/commands/scenario.js.map +1 -0
- package/dist/cli/commands/skill.d.ts +7 -0
- package/dist/cli/commands/skill.d.ts.map +1 -0
- package/dist/cli/commands/skill.js +286 -0
- package/dist/cli/commands/skill.js.map +1 -0
- package/dist/cli/headless-protocol.d.ts +55 -1
- package/dist/cli/headless-protocol.d.ts.map +1 -1
- package/dist/cli/headless-protocol.js +446 -7
- package/dist/cli/headless-protocol.js.map +1 -1
- package/dist/cli/headless-runtime-selection.d.ts +38 -0
- package/dist/cli/headless-runtime-selection.d.ts.map +1 -0
- package/dist/cli/headless-runtime-selection.js +48 -0
- package/dist/cli/headless-runtime-selection.js.map +1 -0
- package/dist/cli/headless.d.ts +5 -1
- package/dist/cli/headless.d.ts.map +1 -1
- package/dist/cli/headless.js +11 -1
- package/dist/cli/headless.js.map +1 -1
- package/dist/cli/help.d.ts +3 -1
- package/dist/cli/help.d.ts.map +1 -1
- package/dist/cli/help.js +135 -30
- package/dist/cli/help.js.map +1 -1
- package/dist/cli/system-prompt.d.ts +22 -4
- package/dist/cli/system-prompt.d.ts.map +1 -1
- package/dist/cli/system-prompt.js +224 -127
- package/dist/cli/system-prompt.js.map +1 -1
- package/dist/cli-tui/commands/a2a-handlers.d.ts +7 -0
- package/dist/cli-tui/commands/a2a-handlers.d.ts.map +1 -0
- package/dist/cli-tui/commands/a2a-handlers.js +160 -0
- package/dist/cli-tui/commands/a2a-handlers.js.map +1 -0
- package/dist/cli-tui/commands/command-catalog.d.ts.map +1 -1
- package/dist/cli-tui/commands/command-catalog.js +15 -0
- package/dist/cli-tui/commands/command-catalog.js.map +1 -1
- package/dist/cli-tui/commands/command-registry-adapter.js +10 -1
- package/dist/cli-tui/commands/command-registry-adapter.js.map +1 -1
- package/dist/cli-tui/commands/handlers/mode-handler.d.ts.map +1 -1
- package/dist/cli-tui/commands/handlers/mode-handler.js +22 -10
- package/dist/cli-tui/commands/handlers/mode-handler.js.map +1 -1
- package/dist/cli-tui/commands/mcp-handlers.d.ts.map +1 -1
- package/dist/cli-tui/commands/mcp-handlers.js +6 -5
- package/dist/cli-tui/commands/mcp-handlers.js.map +1 -1
- package/dist/cli-tui/commands/types.d.ts +1 -0
- package/dist/cli-tui/commands/types.d.ts.map +1 -1
- package/dist/cli-tui/commands/types.js.map +1 -1
- package/dist/cli-tui/commands/utility-handlers.d.ts +4 -0
- package/dist/cli-tui/commands/utility-handlers.d.ts.map +1 -1
- package/dist/cli-tui/commands/utility-handlers.js +45 -6
- package/dist/cli-tui/commands/utility-handlers.js.map +1 -1
- package/dist/cli-tui/run/run-shell-command.d.ts.map +1 -1
- package/dist/cli-tui/run/run-shell-command.js +14 -5
- package/dist/cli-tui/run/run-shell-command.js.map +1 -1
- package/dist/cli-tui/run/streaming-shell-command.d.ts.map +1 -1
- package/dist/cli-tui/run/streaming-shell-command.js +14 -5
- package/dist/cli-tui/run/streaming-shell-command.js.map +1 -1
- package/dist/cli-tui/selectors/model-selector.d.ts +1 -0
- package/dist/cli-tui/selectors/model-selector.d.ts.map +1 -1
- package/dist/cli-tui/selectors/model-selector.js +6 -2
- package/dist/cli-tui/selectors/model-selector.js.map +1 -1
- package/dist/cli-tui/tui-renderer/command-registry-options.d.ts.map +1 -1
- package/dist/cli-tui/tui-renderer/command-registry-options.js +5 -0
- package/dist/cli-tui/tui-renderer/command-registry-options.js.map +1 -1
- package/dist/cli-tui/utils/commands/review-prompt.d.ts +1 -1
- package/dist/cli-tui/utils/commands/review-prompt.d.ts.map +1 -1
- package/dist/cli-tui/utils/commands/review-prompt.js +8 -3
- package/dist/cli-tui/utils/commands/review-prompt.js.map +1 -1
- package/dist/cli-tui/utils/runtime-badges.d.ts +1 -0
- package/dist/cli-tui/utils/runtime-badges.d.ts.map +1 -1
- package/dist/cli-tui/utils/runtime-badges.js +7 -0
- package/dist/cli-tui/utils/runtime-badges.js.map +1 -1
- package/dist/cli-tui/utils/text-formatting.d.ts +1 -0
- package/dist/cli-tui/utils/text-formatting.d.ts.map +1 -1
- package/dist/cli-tui/utils/text-formatting.js +5 -0
- package/dist/cli-tui/utils/text-formatting.js.map +1 -1
- package/dist/cli.js +49158 -26161
- package/dist/cli.js.map +1 -1
- package/dist/codex/app-server-client.d.ts +139 -0
- package/dist/codex/app-server-client.d.ts.map +1 -0
- package/dist/codex/app-server-client.js +352 -0
- package/dist/codex/app-server-client.js.map +1 -0
- package/dist/codex/compatibility.d.ts +48 -0
- package/dist/codex/compatibility.d.ts.map +1 -0
- package/dist/codex/compatibility.js +344 -0
- package/dist/codex/compatibility.js.map +1 -0
- package/dist/codex/subagent-dispatch-table.d.ts +66 -0
- package/dist/codex/subagent-dispatch-table.d.ts.map +1 -0
- package/dist/codex/subagent-dispatch-table.js +127 -0
- package/dist/codex/subagent-dispatch-table.js.map +1 -0
- package/dist/codex/subagent-workgraph.d.ts +56 -0
- package/dist/codex/subagent-workgraph.d.ts.map +1 -0
- package/dist/codex/subagent-workgraph.js +89 -0
- package/dist/codex/subagent-workgraph.js.map +1 -0
- package/dist/config/constants.d.ts +3 -1
- package/dist/config/constants.d.ts.map +1 -1
- package/dist/config/constants.js +9 -0
- package/dist/config/constants.js.map +1 -1
- package/dist/config/feature-flags.d.ts +10 -1
- package/dist/config/feature-flags.d.ts.map +1 -1
- package/dist/config/feature-flags.js +17 -6
- package/dist/config/feature-flags.js.map +1 -1
- package/dist/config/index.d.ts +1 -1
- package/dist/config/index.d.ts.map +1 -1
- package/dist/config/index.js +1 -1
- package/dist/config/index.js.map +1 -1
- package/dist/config/toml-config.d.ts +32 -0
- package/dist/config/toml-config.d.ts.map +1 -1
- package/dist/config/toml-config.js +152 -43
- package/dist/config/toml-config.js.map +1 -1
- package/dist/connectors/service-client.d.ts.map +1 -1
- package/dist/connectors/service-client.js +4 -7
- package/dist/connectors/service-client.js.map +1 -1
- package/dist/context/manifest-types.d.ts +59 -0
- package/dist/context/manifest-types.d.ts.map +1 -0
- package/dist/context/manifest-types.js +2 -0
- package/dist/context/manifest-types.js.map +1 -0
- package/dist/context/manifest.d.ts +16 -0
- package/dist/context/manifest.d.ts.map +1 -0
- package/dist/context/manifest.js +503 -0
- package/dist/context/manifest.js.map +1 -0
- package/dist/db/schema.d.ts +10 -1
- package/dist/db/schema.d.ts.map +1 -1
- package/dist/db/schema.js.map +1 -1
- package/dist/evalops/env-aliases.d.ts +13 -0
- package/dist/evalops/env-aliases.d.ts.map +1 -0
- package/dist/evalops/env-aliases.js +60 -0
- package/dist/evalops/env-aliases.js.map +1 -0
- package/dist/evalops/managed-context.d.ts.map +1 -1
- package/dist/evalops/managed-context.js +28 -62
- package/dist/evalops/managed-context.js.map +1 -1
- package/dist/guardian/config.d.ts.map +1 -1
- package/dist/guardian/config.js +1 -0
- package/dist/guardian/config.js.map +1 -1
- package/dist/guardian/runner.d.ts +2 -0
- package/dist/guardian/runner.d.ts.map +1 -1
- package/dist/guardian/runner.js +105 -2
- package/dist/guardian/runner.js.map +1 -1
- package/dist/guardian/types.d.ts +1 -0
- package/dist/guardian/types.d.ts.map +1 -1
- package/dist/guardian/types.js.map +1 -1
- package/dist/index.d.ts +5 -1
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +5 -1
- package/dist/index.js.map +1 -1
- package/dist/main.d.ts.map +1 -1
- package/dist/main.js +267 -55
- package/dist/main.js.map +1 -1
- package/dist/mcp/config.d.ts.map +1 -1
- package/dist/mcp/config.js +89 -7
- package/dist/mcp/config.js.map +1 -1
- package/dist/mcp/manager.d.ts +5 -1
- package/dist/mcp/manager.d.ts.map +1 -1
- package/dist/mcp/manager.js +159 -4
- package/dist/mcp/manager.js.map +1 -1
- package/dist/mcp/platform-plugin.d.ts.map +1 -1
- package/dist/mcp/platform-plugin.js +9 -26
- package/dist/mcp/platform-plugin.js.map +1 -1
- package/dist/mcp/schema.d.ts +15 -0
- package/dist/mcp/schema.d.ts.map +1 -1
- package/dist/mcp/schema.js +24 -0
- package/dist/mcp/schema.js.map +1 -1
- package/dist/mcp/tool-bridge.d.ts +39 -10
- package/dist/mcp/tool-bridge.d.ts.map +1 -1
- package/dist/mcp/tool-bridge.js +101 -24
- package/dist/mcp/tool-bridge.js.map +1 -1
- package/dist/mcp/types.d.ts +27 -0
- package/dist/mcp/types.d.ts.map +1 -1
- package/dist/mcp/types.js.map +1 -1
- package/dist/mcp/workspace-trust.d.ts +10 -0
- package/dist/mcp/workspace-trust.d.ts.map +1 -0
- package/dist/mcp/workspace-trust.js +295 -0
- package/dist/mcp/workspace-trust.js.map +1 -0
- package/dist/memory/service-client.d.ts +1 -0
- package/dist/memory/service-client.d.ts.map +1 -1
- package/dist/memory/service-client.js +194 -9
- package/dist/memory/service-client.js.map +1 -1
- package/dist/models/builtin.d.ts.map +1 -1
- package/dist/models/builtin.js +66 -25
- package/dist/models/builtin.js.map +1 -1
- package/dist/models/config-loader.d.ts +10 -10
- package/dist/models/config-loader.d.ts.map +1 -1
- package/dist/models/config-loader.js +2 -0
- package/dist/models/config-loader.js.map +1 -1
- package/dist/models/registry.d.ts +1 -0
- package/dist/models/registry.d.ts.map +1 -1
- package/dist/models/registry.js +1 -0
- package/dist/models/registry.js.map +1 -1
- package/dist/oauth/device-identity.d.ts.map +1 -1
- package/dist/oauth/device-identity.js +1 -3
- package/dist/oauth/device-identity.js.map +1 -1
- package/dist/oauth/evalops.d.ts.map +1 -1
- package/dist/oauth/evalops.js +2 -5
- package/dist/oauth/evalops.js.map +1 -1
- package/dist/opentelemetry.d.ts.map +1 -1
- package/dist/opentelemetry.js +12 -12
- package/dist/opentelemetry.js.map +1 -1
- package/dist/package-version.d.ts +2 -0
- package/dist/package-version.d.ts.map +1 -0
- package/dist/package-version.js +11 -0
- package/dist/package-version.js.map +1 -0
- package/dist/packages/constants.d.ts +3 -0
- package/dist/packages/constants.d.ts.map +1 -0
- package/dist/packages/constants.js +3 -0
- package/dist/packages/constants.js.map +1 -0
- package/dist/packages/inspection.d.ts.map +1 -1
- package/dist/packages/inspection.js +2 -1
- package/dist/packages/inspection.js.map +1 -1
- package/dist/packages/search.d.ts.map +1 -1
- package/dist/packages/search.js +1 -1
- package/dist/packages/search.js.map +1 -1
- package/dist/platform/a2a-capability-market.d.ts +21 -0
- package/dist/platform/a2a-capability-market.d.ts.map +1 -0
- package/dist/platform/a2a-capability-market.js +176 -0
- package/dist/platform/a2a-capability-market.js.map +1 -0
- package/dist/platform/a2a-client.d.ts +119 -0
- package/dist/platform/a2a-client.d.ts.map +1 -1
- package/dist/platform/a2a-client.js +413 -24
- package/dist/platform/a2a-client.js.map +1 -1
- package/dist/platform/a2a-fleet.d.ts +41 -0
- package/dist/platform/a2a-fleet.d.ts.map +1 -0
- package/dist/platform/a2a-fleet.js +109 -0
- package/dist/platform/a2a-fleet.js.map +1 -0
- package/dist/platform/a2a-maestro-peer.d.ts +20 -0
- package/dist/platform/a2a-maestro-peer.d.ts.map +1 -0
- package/dist/platform/a2a-maestro-peer.js +176 -0
- package/dist/platform/a2a-maestro-peer.js.map +1 -0
- package/dist/platform/a2a-peer-pairing.d.ts +107 -0
- package/dist/platform/a2a-peer-pairing.d.ts.map +1 -0
- package/dist/platform/a2a-peer-pairing.js +489 -0
- package/dist/platform/a2a-peer-pairing.js.map +1 -0
- package/dist/platform/a2a-peer-registry.d.ts +68 -0
- package/dist/platform/a2a-peer-registry.d.ts.map +1 -0
- package/dist/platform/a2a-peer-registry.js +320 -0
- package/dist/platform/a2a-peer-registry.js.map +1 -0
- package/dist/platform/a2a-task-ledger.d.ts +86 -0
- package/dist/platform/a2a-task-ledger.d.ts.map +1 -0
- package/dist/platform/a2a-task-ledger.js +546 -0
- package/dist/platform/a2a-task-ledger.js.map +1 -0
- package/dist/platform/a2a-work-graph.d.ts +43 -0
- package/dist/platform/a2a-work-graph.d.ts.map +1 -0
- package/dist/platform/a2a-work-graph.js +267 -0
- package/dist/platform/a2a-work-graph.js.map +1 -0
- package/dist/platform/agent-registry-client.d.ts +327 -0
- package/dist/platform/agent-registry-client.d.ts.map +1 -0
- package/dist/platform/agent-registry-client.js +803 -0
- package/dist/platform/agent-registry-client.js.map +1 -0
- package/dist/platform/agent-runtime-client.d.ts +135 -1
- package/dist/platform/agent-runtime-client.d.ts.map +1 -1
- package/dist/platform/agent-runtime-client.js +467 -10
- package/dist/platform/agent-runtime-client.js.map +1 -1
- package/dist/platform/cerebro-facts-client.d.ts.map +1 -1
- package/dist/platform/cerebro-facts-client.js +4 -7
- package/dist/platform/cerebro-facts-client.js.map +1 -1
- package/dist/platform/client.d.ts.map +1 -1
- package/dist/platform/client.js +3 -12
- package/dist/platform/client.js.map +1 -1
- package/dist/platform/core-services.d.ts +68 -0
- package/dist/platform/core-services.d.ts.map +1 -1
- package/dist/platform/core-services.js +68 -0
- package/dist/platform/core-services.js.map +1 -1
- package/dist/platform/fermata-eval-client.d.ts +195 -0
- package/dist/platform/fermata-eval-client.d.ts.map +1 -0
- package/dist/platform/fermata-eval-client.js +146 -0
- package/dist/platform/fermata-eval-client.js.map +1 -0
- package/dist/platform/fermata-scenario-suite.d.ts +99 -0
- package/dist/platform/fermata-scenario-suite.d.ts.map +1 -0
- package/dist/platform/fermata-scenario-suite.js +459 -0
- package/dist/platform/fermata-scenario-suite.js.map +1 -0
- package/dist/platform/maestro-timeline-client.d.ts.map +1 -1
- package/dist/platform/maestro-timeline-client.js +5 -11
- package/dist/platform/maestro-timeline-client.js.map +1 -1
- package/dist/platform/operating-plane-client.d.ts +156 -0
- package/dist/platform/operating-plane-client.d.ts.map +1 -0
- package/dist/platform/operating-plane-client.js +131 -0
- package/dist/platform/operating-plane-client.js.map +1 -0
- package/dist/platform/operating-plane-summary.d.ts +47 -0
- package/dist/platform/operating-plane-summary.d.ts.map +1 -0
- package/dist/platform/operating-plane-summary.js +209 -0
- package/dist/platform/operating-plane-summary.js.map +1 -0
- package/dist/platform/tool-execution-client.d.ts.map +1 -1
- package/dist/platform/tool-execution-client.js +4 -11
- package/dist/platform/tool-execution-client.js.map +1 -1
- package/dist/prompts/service-client.d.ts.map +1 -1
- package/dist/prompts/service-client.js +3 -5
- package/dist/prompts/service-client.js.map +1 -1
- package/dist/prompts/system-prompt.d.ts +3 -0
- package/dist/prompts/system-prompt.d.ts.map +1 -1
- package/dist/prompts/system-prompt.js +13 -3
- package/dist/prompts/system-prompt.js.map +1 -1
- package/dist/prompts/types.d.ts +2 -0
- package/dist/prompts/types.d.ts.map +1 -1
- package/dist/prompts/types.js.map +1 -1
- package/dist/providers/api-keys.d.ts +1 -1
- package/dist/providers/api-keys.d.ts.map +1 -1
- package/dist/providers/api-keys.js +2 -1
- package/dist/providers/api-keys.js.map +1 -1
- package/dist/providers/auth.d.ts.map +1 -1
- package/dist/providers/auth.js +48 -25
- package/dist/providers/auth.js.map +1 -1
- package/dist/remote-runner/attach-client.d.ts.map +1 -1
- package/dist/remote-runner/attach-client.js +1 -0
- package/dist/remote-runner/attach-client.js.map +1 -1
- package/dist/remote-runner/client.d.ts.map +1 -1
- package/dist/remote-runner/client.js +4 -7
- package/dist/remote-runner/client.js.map +1 -1
- package/dist/rpc/rpc-types.d.ts.map +1 -1
- package/dist/rpc/rpc-types.js +1 -0
- package/dist/rpc/rpc-types.js.map +1 -1
- package/dist/safety/action-firewall.d.ts.map +1 -1
- package/dist/safety/action-firewall.js +69 -6
- package/dist/safety/action-firewall.js.map +1 -1
- package/dist/safety/attack-patterns.d.ts.map +1 -1
- package/dist/safety/attack-patterns.js +31 -14
- package/dist/safety/attack-patterns.js.map +1 -1
- package/dist/safety/bash-parser.d.ts.map +1 -1
- package/dist/safety/bash-parser.js +4 -3
- package/dist/safety/bash-parser.js.map +1 -1
- package/dist/safety/execpolicy.d.ts +15 -3
- package/dist/safety/execpolicy.d.ts.map +1 -1
- package/dist/safety/execpolicy.js +248 -37
- package/dist/safety/execpolicy.js.map +1 -1
- package/dist/safety/governance-service-client.d.ts.map +1 -1
- package/dist/safety/governance-service-client.js +4 -7
- package/dist/safety/governance-service-client.js.map +1 -1
- package/dist/safety/guarded-files.d.ts +34 -0
- package/dist/safety/guarded-files.d.ts.map +1 -0
- package/dist/safety/guarded-files.js +372 -0
- package/dist/safety/guarded-files.js.map +1 -0
- package/dist/safety/loop-detector.d.ts.map +1 -1
- package/dist/safety/loop-detector.js +6 -1
- package/dist/safety/loop-detector.js.map +1 -1
- package/dist/safety/permission-profile.d.ts +47 -0
- package/dist/safety/permission-profile.d.ts.map +1 -0
- package/dist/safety/permission-profile.js +349 -0
- package/dist/safety/permission-profile.js.map +1 -0
- package/dist/safety/safety-middleware.d.ts +5 -1
- package/dist/safety/safety-middleware.d.ts.map +1 -1
- package/dist/safety/safety-middleware.js +3 -2
- package/dist/safety/safety-middleware.js.map +1 -1
- package/dist/safety/suspicious-patterns.d.ts.map +1 -1
- package/dist/safety/suspicious-patterns.js +12 -7
- package/dist/safety/suspicious-patterns.js.map +1 -1
- package/dist/safety/tool-categorization.js +1 -1
- package/dist/safety/tool-categorization.js.map +1 -1
- package/dist/safety/validators/path-policy-validator.d.ts.map +1 -1
- package/dist/safety/validators/path-policy-validator.js +7 -0
- package/dist/safety/validators/path-policy-validator.js.map +1 -1
- package/dist/sandbox/index.d.ts +6 -1
- package/dist/sandbox/index.d.ts.map +1 -1
- package/dist/sandbox/index.js +30 -4
- package/dist/sandbox/index.js.map +1 -1
- package/dist/sandbox/native-sandbox.d.ts +3 -0
- package/dist/sandbox/native-sandbox.d.ts.map +1 -1
- package/dist/sandbox/native-sandbox.js +139 -18
- package/dist/sandbox/native-sandbox.js.map +1 -1
- package/dist/sentry.d.ts +17 -0
- package/dist/sentry.d.ts.map +1 -0
- package/dist/sentry.js +91 -0
- package/dist/sentry.js.map +1 -0
- package/dist/server/agent-runtime-ledger.d.ts +141 -0
- package/dist/server/agent-runtime-ledger.d.ts.map +1 -0
- package/dist/server/agent-runtime-ledger.js +334 -0
- package/dist/server/agent-runtime-ledger.js.map +1 -0
- package/dist/server/agent-trajectory-inspection.d.ts +104 -0
- package/dist/server/agent-trajectory-inspection.d.ts.map +1 -0
- package/dist/server/agent-trajectory-inspection.js +168 -0
- package/dist/server/agent-trajectory-inspection.js.map +1 -0
- package/dist/server/agent-trajectory-replay.d.ts +56 -0
- package/dist/server/agent-trajectory-replay.d.ts.map +1 -0
- package/dist/server/agent-trajectory-replay.js +214 -0
- package/dist/server/agent-trajectory-replay.js.map +1 -0
- package/dist/server/agent-trajectory-scenarios.d.ts +109 -0
- package/dist/server/agent-trajectory-scenarios.d.ts.map +1 -0
- package/dist/server/agent-trajectory-scenarios.js +822 -0
- package/dist/server/agent-trajectory-scenarios.js.map +1 -0
- package/dist/server/agent-trajectory-scorers.d.ts +61 -0
- package/dist/server/agent-trajectory-scorers.d.ts.map +1 -0
- package/dist/server/agent-trajectory-scorers.js +162 -0
- package/dist/server/agent-trajectory-scorers.js.map +1 -0
- package/dist/server/agent-trajectory-validation.d.ts +7 -0
- package/dist/server/agent-trajectory-validation.d.ts.map +1 -0
- package/dist/server/agent-trajectory-validation.js +101 -0
- package/dist/server/agent-trajectory-validation.js.map +1 -0
- package/dist/server/agent-trajectory.d.ts +50 -0
- package/dist/server/agent-trajectory.d.ts.map +1 -0
- package/dist/server/agent-trajectory.js +159 -0
- package/dist/server/agent-trajectory.js.map +1 -0
- package/dist/server/app-context.d.ts +18 -0
- package/dist/server/app-context.d.ts.map +1 -1
- package/dist/server/app-context.js.map +1 -1
- package/dist/server/handlers/headless-sessions.d.ts.map +1 -1
- package/dist/server/handlers/headless-sessions.js +64 -27
- package/dist/server/handlers/headless-sessions.js.map +1 -1
- package/dist/server/handlers/health.d.ts +6 -0
- package/dist/server/handlers/health.d.ts.map +1 -1
- package/dist/server/handlers/health.js +6 -0
- package/dist/server/handlers/health.js.map +1 -1
- package/dist/server/handlers/hosted-runner-drain.d.ts +69 -1
- package/dist/server/handlers/hosted-runner-drain.d.ts.map +1 -1
- package/dist/server/handlers/hosted-runner-drain.js +309 -18
- package/dist/server/handlers/hosted-runner-drain.js.map +1 -1
- package/dist/server/handlers/hosted-runner-identity.d.ts +16 -0
- package/dist/server/handlers/hosted-runner-identity.d.ts.map +1 -1
- package/dist/server/handlers/hosted-runner-identity.js +43 -0
- package/dist/server/handlers/hosted-runner-identity.js.map +1 -1
- package/dist/server/handlers/mode.d.ts.map +1 -1
- package/dist/server/handlers/mode.js +3 -1
- package/dist/server/handlers/mode.js.map +1 -1
- package/dist/server/handlers/pending-requests.d.ts.map +1 -1
- package/dist/server/handlers/pending-requests.js +336 -51
- package/dist/server/handlers/pending-requests.js.map +1 -1
- package/dist/server/handlers/platform-a2a-push.d.ts +6 -0
- package/dist/server/handlers/platform-a2a-push.d.ts.map +1 -0
- package/dist/server/handlers/platform-a2a-push.js +136 -0
- package/dist/server/handlers/platform-a2a-push.js.map +1 -0
- package/dist/server/handlers/sessions.d.ts.map +1 -1
- package/dist/server/handlers/sessions.js +21 -1
- package/dist/server/handlers/sessions.js.map +1 -1
- package/dist/server/headless-runtime-service.d.ts +9 -2
- package/dist/server/headless-runtime-service.d.ts.map +1 -1
- package/dist/server/headless-runtime-service.js +39 -0
- package/dist/server/headless-runtime-service.js.map +1 -1
- package/dist/server/hosted-agent-runtime-progress.d.ts +144 -0
- package/dist/server/hosted-agent-runtime-progress.d.ts.map +1 -0
- package/dist/server/hosted-agent-runtime-progress.js +1650 -0
- package/dist/server/hosted-agent-runtime-progress.js.map +1 -0
- package/dist/server/hosted-runner-lease.d.ts +41 -0
- package/dist/server/hosted-runner-lease.d.ts.map +1 -0
- package/dist/server/hosted-runner-lease.js +140 -0
- package/dist/server/hosted-runner-lease.js.map +1 -0
- package/dist/server/hosted-session-manager.d.ts +16 -8
- package/dist/server/hosted-session-manager.d.ts.map +1 -1
- package/dist/server/hosted-session-manager.js +98 -76
- package/dist/server/hosted-session-manager.js.map +1 -1
- package/dist/server/junit-xml.d.ts +2 -0
- package/dist/server/junit-xml.d.ts.map +1 -0
- package/dist/server/junit-xml.js +9 -0
- package/dist/server/junit-xml.js.map +1 -0
- package/dist/server/routes.d.ts.map +1 -1
- package/dist/server/routes.js +6 -0
- package/dist/server/routes.js.map +1 -1
- package/dist/server/scenario-recorder.d.ts +56 -0
- package/dist/server/scenario-recorder.d.ts.map +1 -0
- package/dist/server/scenario-recorder.js +109 -0
- package/dist/server/scenario-recorder.js.map +1 -0
- package/dist/server/scripted-scenario-runner.d.ts +46 -0
- package/dist/server/scripted-scenario-runner.d.ts.map +1 -0
- package/dist/server/scripted-scenario-runner.js +190 -0
- package/dist/server/scripted-scenario-runner.js.map +1 -0
- package/dist/server/server-middlewares.d.ts +6 -2
- package/dist/server/server-middlewares.d.ts.map +1 -1
- package/dist/server/server-middlewares.js +8 -2
- package/dist/server/server-middlewares.js.map +1 -1
- package/dist/server/server-request-manager.d.ts +9 -1
- package/dist/server/server-request-manager.d.ts.map +1 -1
- package/dist/server/server-request-manager.js +49 -4
- package/dist/server/server-request-manager.js.map +1 -1
- package/dist/services/intelligent-router/service.d.ts +2 -0
- package/dist/services/intelligent-router/service.d.ts.map +1 -1
- package/dist/services/intelligent-router/service.js +4 -0
- package/dist/services/intelligent-router/service.js.map +1 -1
- package/dist/services/service-authority.d.ts +25 -0
- package/dist/services/service-authority.d.ts.map +1 -0
- package/dist/services/service-authority.js +143 -0
- package/dist/services/service-authority.js.map +1 -0
- package/dist/services/workspace-config/service.d.ts +2 -0
- package/dist/services/workspace-config/service.d.ts.map +1 -1
- package/dist/services/workspace-config/service.js +4 -0
- package/dist/services/workspace-config/service.js.map +1 -1
- package/dist/session/manager.d.ts +8 -2
- package/dist/session/manager.d.ts.map +1 -1
- package/dist/session/manager.js +19 -4
- package/dist/session/manager.js.map +1 -1
- package/dist/session/session-branch.d.ts.map +1 -1
- package/dist/session/session-branch.js +5 -1
- package/dist/session/session-branch.js.map +1 -1
- package/dist/session/session-catalog.d.ts +7 -2
- package/dist/session/session-catalog.d.ts.map +1 -1
- package/dist/session/session-catalog.js +13 -2
- package/dist/session/session-catalog.js.map +1 -1
- package/dist/session/session-context.d.ts +6 -2
- package/dist/session/session-context.d.ts.map +1 -1
- package/dist/session/session-context.js +44 -4
- package/dist/session/session-context.js.map +1 -1
- package/dist/session/session-graph-projection.d.ts +32 -0
- package/dist/session/session-graph-projection.d.ts.map +1 -0
- package/dist/session/session-graph-projection.js +133 -0
- package/dist/session/session-graph-projection.js.map +1 -0
- package/dist/session/types.d.ts +18 -0
- package/dist/session/types.d.ts.map +1 -1
- package/dist/session/types.js +8 -0
- package/dist/session/types.js.map +1 -1
- package/dist/session/wire-format.generated.d.ts +3 -0
- package/dist/session/wire-format.generated.d.ts.map +1 -1
- package/dist/session/wire-format.generated.js +20 -0
- package/dist/session/wire-format.generated.js.map +1 -1
- package/dist/skills/eval-harness.d.ts +38 -0
- package/dist/skills/eval-harness.d.ts.map +1 -0
- package/dist/skills/eval-harness.js +112 -0
- package/dist/skills/eval-harness.js.map +1 -0
- package/dist/skills/index.d.ts +5 -1
- package/dist/skills/index.d.ts.map +1 -1
- package/dist/skills/index.js +5 -1
- package/dist/skills/index.js.map +1 -1
- package/dist/skills/linter.d.ts +36 -0
- package/dist/skills/linter.d.ts.map +1 -0
- package/dist/skills/linter.js +398 -0
- package/dist/skills/linter.js.map +1 -0
- package/dist/skills/loader.d.ts +36 -6
- package/dist/skills/loader.d.ts.map +1 -1
- package/dist/skills/loader.js +169 -86
- package/dist/skills/loader.js.map +1 -1
- package/dist/skills/package-contract.d.ts +45 -0
- package/dist/skills/package-contract.d.ts.map +1 -0
- package/dist/skills/package-contract.js +153 -0
- package/dist/skills/package-contract.js.map +1 -0
- package/dist/skills/runtime-activation.d.ts +68 -0
- package/dist/skills/runtime-activation.d.ts.map +1 -0
- package/dist/skills/runtime-activation.js +228 -0
- package/dist/skills/runtime-activation.js.map +1 -0
- package/dist/skills/service-client.d.ts.map +1 -1
- package/dist/skills/service-client.js +3 -4
- package/dist/skills/service-client.js.map +1 -1
- package/dist/skills/tool.d.ts.map +1 -1
- package/dist/skills/tool.js +2 -0
- package/dist/skills/tool.js.map +1 -1
- package/dist/skills/types.d.ts +14 -0
- package/dist/skills/types.d.ts.map +1 -1
- package/dist/skills/types.js.map +1 -1
- package/dist/telemetry/agent-operating-plane-context.d.ts +24 -0
- package/dist/telemetry/agent-operating-plane-context.d.ts.map +1 -0
- package/dist/telemetry/agent-operating-plane-context.js +54 -0
- package/dist/telemetry/agent-operating-plane-context.js.map +1 -0
- package/dist/telemetry/beacon.d.ts +24 -0
- package/dist/telemetry/beacon.d.ts.map +1 -0
- package/dist/telemetry/beacon.js +115 -0
- package/dist/telemetry/beacon.js.map +1 -0
- package/dist/telemetry/cli-command-aggregator.d.ts +36 -0
- package/dist/telemetry/cli-command-aggregator.d.ts.map +1 -0
- package/dist/telemetry/cli-command-aggregator.js +251 -0
- package/dist/telemetry/cli-command-aggregator.js.map +1 -0
- package/dist/telemetry/cli-startup.d.ts +21 -0
- package/dist/telemetry/cli-startup.d.ts.map +1 -0
- package/dist/telemetry/cli-startup.js +83 -0
- package/dist/telemetry/cli-startup.js.map +1 -0
- package/dist/telemetry/disablement.d.ts +13 -0
- package/dist/telemetry/disablement.d.ts.map +1 -0
- package/dist/telemetry/disablement.js +19 -0
- package/dist/telemetry/disablement.js.map +1 -0
- package/dist/telemetry/index.d.ts +6 -2
- package/dist/telemetry/index.d.ts.map +1 -1
- package/dist/telemetry/index.js +6 -2
- package/dist/telemetry/index.js.map +1 -1
- package/dist/telemetry/maestro-event-bus.d.ts +52 -0
- package/dist/telemetry/maestro-event-bus.d.ts.map +1 -1
- package/dist/telemetry/maestro-event-bus.js +179 -3
- package/dist/telemetry/maestro-event-bus.js.map +1 -1
- package/dist/telemetry/maestro-event-catalog.d.ts +3 -1
- package/dist/telemetry/maestro-event-catalog.d.ts.map +1 -1
- package/dist/telemetry/maestro-event-catalog.js +2 -0
- package/dist/telemetry/maestro-event-catalog.js.map +1 -1
- package/dist/telemetry/maestro-platform-replay-fixture.d.ts.map +1 -1
- package/dist/telemetry/maestro-platform-replay-fixture.js +74 -9
- package/dist/telemetry/maestro-platform-replay-fixture.js.map +1 -1
- package/dist/telemetry/maestro-publisher-conformance-fixture.d.ts +1 -1
- package/dist/telemetry/maestro-publisher-conformance-fixture.d.ts.map +1 -1
- package/dist/telemetry/maestro-publisher-conformance-fixture.js +46 -11
- package/dist/telemetry/maestro-publisher-conformance-fixture.js.map +1 -1
- package/dist/telemetry/mcp-beacon.d.ts +20 -0
- package/dist/telemetry/mcp-beacon.d.ts.map +1 -0
- package/dist/telemetry/mcp-beacon.js +54 -0
- package/dist/telemetry/mcp-beacon.js.map +1 -0
- package/dist/telemetry/metadata-normalization.d.ts +7 -0
- package/dist/telemetry/metadata-normalization.d.ts.map +1 -0
- package/dist/telemetry/metadata-normalization.js +189 -0
- package/dist/telemetry/metadata-normalization.js.map +1 -0
- package/dist/telemetry/meter-service-client.d.ts.map +1 -1
- package/dist/telemetry/meter-service-client.js +5 -10
- package/dist/telemetry/meter-service-client.js.map +1 -1
- package/dist/telemetry/metrics.d.ts +123 -0
- package/dist/telemetry/metrics.d.ts.map +1 -0
- package/dist/telemetry/metrics.js +184 -0
- package/dist/telemetry/metrics.js.map +1 -0
- package/dist/telemetry/turn-tracker.d.ts +1 -0
- package/dist/telemetry/turn-tracker.d.ts.map +1 -1
- package/dist/telemetry/turn-tracker.js +22 -1
- package/dist/telemetry/turn-tracker.js.map +1 -1
- package/dist/telemetry/wide-events.d.ts +31 -1
- package/dist/telemetry/wide-events.d.ts.map +1 -1
- package/dist/telemetry/wide-events.js +229 -1
- package/dist/telemetry/wide-events.js.map +1 -1
- package/dist/telemetry.d.ts +39 -3
- package/dist/telemetry.d.ts.map +1 -1
- package/dist/telemetry.js +160 -3
- package/dist/telemetry.js.map +1 -1
- package/dist/testing/test-verification-hook.d.ts.map +1 -1
- package/dist/testing/test-verification-hook.js +34 -21
- package/dist/testing/test-verification-hook.js.map +1 -1
- package/dist/tools/apply-patch-parser.d.ts +26 -0
- package/dist/tools/apply-patch-parser.d.ts.map +1 -0
- package/dist/tools/apply-patch-parser.js +146 -0
- package/dist/tools/apply-patch-parser.js.map +1 -0
- package/dist/tools/apply-patch.d.ts +42 -0
- package/dist/tools/apply-patch.d.ts.map +1 -0
- package/dist/tools/apply-patch.js +648 -0
- package/dist/tools/apply-patch.js.map +1 -0
- package/dist/tools/ask-user.d.ts +1 -0
- package/dist/tools/ask-user.d.ts.map +1 -1
- package/dist/tools/background/tool-handler.d.ts +1 -0
- package/dist/tools/background/tool-handler.d.ts.map +1 -1
- package/dist/tools/bash.d.ts +1 -0
- package/dist/tools/bash.d.ts.map +1 -1
- package/dist/tools/codesearch.d.ts +1 -0
- package/dist/tools/codesearch.d.ts.map +1 -1
- package/dist/tools/conductor-client.d.ts +1 -0
- package/dist/tools/conductor-client.d.ts.map +1 -1
- package/dist/tools/conductor-client.js +63 -0
- package/dist/tools/conductor-client.js.map +1 -1
- package/dist/tools/diff.d.ts +1 -0
- package/dist/tools/diff.d.ts.map +1 -1
- package/dist/tools/edit.d.ts +1 -0
- package/dist/tools/edit.d.ts.map +1 -1
- package/dist/tools/extract-document.d.ts +1 -0
- package/dist/tools/extract-document.d.ts.map +1 -1
- package/dist/tools/find.d.ts +10 -0
- package/dist/tools/find.d.ts.map +1 -1
- package/dist/tools/find.js +236 -27
- package/dist/tools/find.js.map +1 -1
- package/dist/tools/gh.d.ts +3 -0
- package/dist/tools/gh.d.ts.map +1 -1
- package/dist/tools/index.d.ts +50 -0
- package/dist/tools/index.d.ts.map +1 -1
- package/dist/tools/index.js +4 -0
- package/dist/tools/index.js.map +1 -1
- package/dist/tools/list.d.ts +1 -0
- package/dist/tools/list.d.ts.map +1 -1
- package/dist/tools/notebook.d.ts +1 -0
- package/dist/tools/notebook.d.ts.map +1 -1
- package/dist/tools/oracle.d.ts +1 -0
- package/dist/tools/oracle.d.ts.map +1 -1
- package/dist/tools/oracle.js +17 -32
- package/dist/tools/oracle.js.map +1 -1
- package/dist/tools/parallel-execution.d.ts +13 -2
- package/dist/tools/parallel-execution.d.ts.map +1 -1
- package/dist/tools/parallel-execution.js +178 -7
- package/dist/tools/parallel-execution.js.map +1 -1
- package/dist/tools/parallel-ripgrep.d.ts +1 -0
- package/dist/tools/parallel-ripgrep.d.ts.map +1 -1
- package/dist/tools/pipeline.d.ts +4 -0
- package/dist/tools/pipeline.d.ts.map +1 -1
- package/dist/tools/read.d.ts +1 -0
- package/dist/tools/read.d.ts.map +1 -1
- package/dist/tools/search.d.ts +1 -0
- package/dist/tools/search.d.ts.map +1 -1
- package/dist/tools/status.d.ts +1 -0
- package/dist/tools/status.d.ts.map +1 -1
- package/dist/tools/todo.d.ts +1 -0
- package/dist/tools/todo.d.ts.map +1 -1
- package/dist/tools/tool-dsl.d.ts +5 -1
- package/dist/tools/tool-dsl.d.ts.map +1 -1
- package/dist/tools/tool-dsl.js +1 -0
- package/dist/tools/tool-dsl.js.map +1 -1
- package/dist/tools/tool-result-cache.d.ts.map +1 -1
- package/dist/tools/tool-result-cache.js +1 -0
- package/dist/tools/tool-result-cache.js.map +1 -1
- package/dist/tools/typebox-tool.d.ts.map +1 -1
- package/dist/tools/typebox-tool.js +3 -0
- package/dist/tools/typebox-tool.js.map +1 -1
- package/dist/tools/webfetch.d.ts +1 -0
- package/dist/tools/webfetch.d.ts.map +1 -1
- package/dist/tools/websearch.d.ts +1 -0
- package/dist/tools/websearch.d.ts.map +1 -1
- package/dist/tools/write.d.ts +1 -0
- package/dist/tools/write.d.ts.map +1 -1
- package/dist/utils/abort-error.d.ts +2 -0
- package/dist/utils/abort-error.d.ts.map +1 -0
- package/dist/utils/abort-error.js +7 -0
- package/dist/utils/abort-error.js.map +1 -0
- package/dist/utils/downstream.d.ts.map +1 -1
- package/dist/utils/downstream.js +4 -0
- package/dist/utils/downstream.js.map +1 -1
- package/dist/utils/logger.d.ts +3 -0
- package/dist/utils/logger.d.ts.map +1 -1
- package/dist/utils/logger.js +14 -3
- package/dist/utils/logger.js.map +1 -1
- package/dist/utils/path-containment.d.ts +2 -0
- package/dist/utils/path-containment.d.ts.map +1 -0
- package/dist/utils/path-containment.js +14 -0
- package/dist/utils/path-containment.js.map +1 -0
- package/dist/utils/prompt-safe-text.d.ts +7 -0
- package/dist/utils/prompt-safe-text.d.ts.map +1 -0
- package/dist/utils/prompt-safe-text.js +44 -0
- package/dist/utils/prompt-safe-text.js.map +1 -0
- package/dist/version.json +2 -2
- package/dist/web-server.d.ts.map +1 -1
- package/dist/web-server.js +32 -7
- package/dist/web-server.js.map +1 -1
- package/dist/workspace/root-resolver.d.ts +4 -1
- package/dist/workspace/root-resolver.d.ts.map +1 -1
- package/dist/workspace/root-resolver.js +16 -6
- package/dist/workspace/root-resolver.js.map +1 -1
- package/node_modules/@evalops/contracts/dist/advisor-effort.d.ts +8 -0
- package/node_modules/@evalops/contracts/dist/advisor-effort.d.ts.map +1 -0
- package/node_modules/@evalops/contracts/dist/advisor-effort.js +35 -0
- package/node_modules/@evalops/contracts/dist/advisor-effort.js.map +1 -0
- package/node_modules/@evalops/contracts/dist/delegation-prompt.d.ts +10 -0
- package/node_modules/@evalops/contracts/dist/delegation-prompt.d.ts.map +1 -0
- package/node_modules/@evalops/contracts/dist/delegation-prompt.js +48 -0
- package/node_modules/@evalops/contracts/dist/delegation-prompt.js.map +1 -0
- package/node_modules/@evalops/contracts/dist/guarded-files-settings.d.ts +20 -0
- package/node_modules/@evalops/contracts/dist/guarded-files-settings.d.ts.map +1 -0
- package/node_modules/@evalops/contracts/dist/guarded-files-settings.js +147 -0
- package/node_modules/@evalops/contracts/dist/guarded-files-settings.js.map +1 -0
- package/node_modules/@evalops/contracts/dist/headless-protocol-generated.d.ts +2 -0
- package/node_modules/@evalops/contracts/dist/headless-protocol-generated.d.ts.map +1 -1
- package/node_modules/@evalops/contracts/dist/headless-protocol-generated.js +1 -0
- package/node_modules/@evalops/contracts/dist/headless-protocol-generated.js.map +1 -1
- package/node_modules/@evalops/contracts/dist/headless-protocol-schemas.generated.d.ts +280 -5
- package/node_modules/@evalops/contracts/dist/headless-protocol-schemas.generated.d.ts.map +1 -1
- package/node_modules/@evalops/contracts/dist/headless-protocol-schemas.generated.js +36 -1
- package/node_modules/@evalops/contracts/dist/headless-protocol-schemas.generated.js.map +1 -1
- package/node_modules/@evalops/contracts/dist/index.d.ts +15 -1
- package/node_modules/@evalops/contracts/dist/index.d.ts.map +1 -1
- package/node_modules/@evalops/contracts/dist/index.js +6 -0
- package/node_modules/@evalops/contracts/dist/index.js.map +1 -1
- package/node_modules/@evalops/contracts/dist/maestro-app-server.d.ts +593 -0
- package/node_modules/@evalops/contracts/dist/maestro-app-server.d.ts.map +1 -0
- package/node_modules/@evalops/contracts/dist/maestro-app-server.js +218 -0
- package/node_modules/@evalops/contracts/dist/maestro-app-server.js.map +1 -0
- package/node_modules/@evalops/contracts/dist/proto/maestro/v1/headless_pb.d.ts +91 -0
- package/node_modules/@evalops/contracts/dist/proto/maestro/v1/headless_pb.d.ts.map +1 -1
- package/node_modules/@evalops/contracts/dist/proto/maestro/v1/headless_pb.js +92 -61
- package/node_modules/@evalops/contracts/dist/proto/maestro/v1/headless_pb.js.map +1 -1
- package/node_modules/@evalops/contracts/dist/runtime-app-server.d.ts +3 -0
- package/node_modules/@evalops/contracts/dist/runtime-app-server.d.ts.map +1 -1
- package/node_modules/@evalops/contracts/dist/runtime-constraints.d.ts +28 -0
- package/node_modules/@evalops/contracts/dist/runtime-constraints.d.ts.map +1 -0
- package/node_modules/@evalops/contracts/dist/runtime-constraints.js +71 -0
- package/node_modules/@evalops/contracts/dist/runtime-constraints.js.map +1 -0
- package/node_modules/@evalops/contracts/dist/runtime-server-request.d.ts +7 -0
- package/node_modules/@evalops/contracts/dist/runtime-server-request.d.ts.map +1 -1
- package/node_modules/@evalops/contracts/dist/runtime-server-request.js +2 -0
- package/node_modules/@evalops/contracts/dist/runtime-server-request.js.map +1 -1
- package/node_modules/@evalops/contracts/dist/scenario.d.ts +191 -0
- package/node_modules/@evalops/contracts/dist/scenario.d.ts.map +1 -0
- package/node_modules/@evalops/contracts/dist/scenario.js +4 -0
- package/node_modules/@evalops/contracts/dist/scenario.js.map +1 -0
- package/node_modules/@evalops/contracts/dist/schemas.d.ts +17 -3
- package/node_modules/@evalops/contracts/dist/schemas.d.ts.map +1 -1
- package/node_modules/@evalops/contracts/dist/schemas.js +16 -0
- package/node_modules/@evalops/contracts/dist/schemas.js.map +1 -1
- package/node_modules/@evalops/contracts/package.json +1 -1
- package/node_modules/@evalops/tui/package.json +1 -1
- package/package.json +41 -7
- package/skills/incident-triage/SKILL.md +40 -0
- package/skills/incident-triage/mcp.json +14 -0
- package/skills/incident-triage/reference/triage.md +20 -0
- package/skills/incident-triage/toolbox/incident-timeline +22 -0
- package/skills/incident-triage/toolbox/incident-timeline.cmd +16 -0
- package/skills/pr-review/SKILL.md +40 -0
- package/skills/pr-review/mcp.json +14 -0
- package/skills/pr-review/reference/rubric.md +20 -0
- package/skills/pr-review/toolbox/review-summary +20 -0
- package/skills/pr-review/toolbox/review-summary.cmd +14 -0
- package/skills/release-verification/SKILL.md +40 -0
- package/skills/release-verification/mcp.json +15 -0
- package/skills/release-verification/reference/checklist.md +21 -0
- package/skills/release-verification/toolbox/release-readiness +18 -0
- package/skills/release-verification/toolbox/release-readiness.cmd +12 -0
|
@@ -0,0 +1,168 @@
|
|
|
1
|
+
export const AGENT_TRAJECTORY_INSPECTION_SCHEMA = "evalops.maestro.agent-trajectory-inspection.v1";
|
|
2
|
+
function redactedEvidence(evidence) {
|
|
3
|
+
return evidence
|
|
4
|
+
.map((anchor) => ({
|
|
5
|
+
...anchor,
|
|
6
|
+
redacted: true,
|
|
7
|
+
label: `${anchor.kind}:${anchor.id}`,
|
|
8
|
+
}))
|
|
9
|
+
.sort((left, right) => {
|
|
10
|
+
const kindDelta = left.kind.localeCompare(right.kind);
|
|
11
|
+
return kindDelta === 0 ? left.id.localeCompare(right.id) : kindDelta;
|
|
12
|
+
});
|
|
13
|
+
}
|
|
14
|
+
function timelineItemIdsForEvidence(evidence) {
|
|
15
|
+
return evidence
|
|
16
|
+
.filter((anchor) => anchor.kind === "timeline_item")
|
|
17
|
+
.map((anchor) => anchor.id)
|
|
18
|
+
.sort();
|
|
19
|
+
}
|
|
20
|
+
function timelineItemIdsForEvents(eventIds, eventsById) {
|
|
21
|
+
const ids = new Set();
|
|
22
|
+
for (const eventId of eventIds) {
|
|
23
|
+
const event = eventsById.get(eventId);
|
|
24
|
+
for (const timelineItemId of event?.timelineItemIds ?? []) {
|
|
25
|
+
ids.add(timelineItemId);
|
|
26
|
+
}
|
|
27
|
+
}
|
|
28
|
+
return [...ids].sort();
|
|
29
|
+
}
|
|
30
|
+
function metadataKeys(item) {
|
|
31
|
+
const metadata = item.metadata;
|
|
32
|
+
if (!metadata || typeof metadata !== "object" || Array.isArray(metadata)) {
|
|
33
|
+
return [];
|
|
34
|
+
}
|
|
35
|
+
return Object.keys(metadata).sort();
|
|
36
|
+
}
|
|
37
|
+
function inspectTimelineItem(item) {
|
|
38
|
+
return {
|
|
39
|
+
id: item.id,
|
|
40
|
+
timestamp: item.timestamp,
|
|
41
|
+
type: item.type,
|
|
42
|
+
status: item.status,
|
|
43
|
+
visibility: item.visibility,
|
|
44
|
+
source: item.source,
|
|
45
|
+
title: item.title,
|
|
46
|
+
...(item.summary ? { summary: item.summary } : {}),
|
|
47
|
+
...(item.role ? { role: item.role } : {}),
|
|
48
|
+
...(item.toolName ? { toolName: item.toolName } : {}),
|
|
49
|
+
...(item.pendingRequestKind
|
|
50
|
+
? { pendingRequestKind: item.pendingRequestKind }
|
|
51
|
+
: {}),
|
|
52
|
+
...(item.platformOperation
|
|
53
|
+
? { platformOperation: item.platformOperation }
|
|
54
|
+
: {}),
|
|
55
|
+
metadataKeys: metadataKeys(item),
|
|
56
|
+
redacted: true,
|
|
57
|
+
};
|
|
58
|
+
}
|
|
59
|
+
function inspectEvent(event) {
|
|
60
|
+
return {
|
|
61
|
+
id: event.id,
|
|
62
|
+
sequence: event.sequence,
|
|
63
|
+
timestamp: event.timestamp,
|
|
64
|
+
kind: event.kind,
|
|
65
|
+
phase: event.phase,
|
|
66
|
+
actor: event.actor,
|
|
67
|
+
type: event.type,
|
|
68
|
+
status: event.status,
|
|
69
|
+
visibility: event.visibility,
|
|
70
|
+
source: event.source,
|
|
71
|
+
title: event.title,
|
|
72
|
+
...(event.summary ? { summary: event.summary } : {}),
|
|
73
|
+
...(event.toolName ? { toolName: event.toolName } : {}),
|
|
74
|
+
relatedIds: event.relatedIds ?? [],
|
|
75
|
+
timelineItemIds: timelineItemIdsForEvidence(event.evidence),
|
|
76
|
+
evidence: redactedEvidence(event.evidence),
|
|
77
|
+
};
|
|
78
|
+
}
|
|
79
|
+
function finalAnswerFromEvents(events) {
|
|
80
|
+
const reversedEvents = [...events].reverse();
|
|
81
|
+
const event = reversedEvents.find((candidate) => candidate.actor === "assistant" &&
|
|
82
|
+
candidate.type === "message.assistant") ??
|
|
83
|
+
reversedEvents.find((candidate) => candidate.phase === "finish" || candidate.phase === "recover");
|
|
84
|
+
if (!event)
|
|
85
|
+
return undefined;
|
|
86
|
+
return {
|
|
87
|
+
eventId: event.id,
|
|
88
|
+
timelineItemIds: event.timelineItemIds,
|
|
89
|
+
title: event.title,
|
|
90
|
+
...(event.summary ? { summary: event.summary } : {}),
|
|
91
|
+
redacted: true,
|
|
92
|
+
};
|
|
93
|
+
}
|
|
94
|
+
export function buildAgentTrajectoryInspectionReport({ timelineItems, trajectory, replay, score, }) {
|
|
95
|
+
const events = trajectory.events.map(inspectEvent);
|
|
96
|
+
const finalAnswer = finalAnswerFromEvents(events);
|
|
97
|
+
const eventsById = new Map(events.map((event) => [event.id, event]));
|
|
98
|
+
const replayDeltas = replay.deltas.map((delta) => ({
|
|
99
|
+
id: delta.id,
|
|
100
|
+
severity: delta.severity,
|
|
101
|
+
ruleId: delta.ruleId,
|
|
102
|
+
message: delta.message,
|
|
103
|
+
...(delta.eventId ? { eventId: delta.eventId } : {}),
|
|
104
|
+
timelineItemIds: delta.eventId
|
|
105
|
+
? timelineItemIdsForEvents([delta.eventId], eventsById)
|
|
106
|
+
: timelineItemIdsForEvidence(delta.evidence),
|
|
107
|
+
evidence: redactedEvidence(delta.evidence),
|
|
108
|
+
}));
|
|
109
|
+
const scoreFindings = score.findings.map((finding) => ({
|
|
110
|
+
ruleId: finding.ruleId,
|
|
111
|
+
status: finding.status,
|
|
112
|
+
severity: finding.severity,
|
|
113
|
+
message: finding.message,
|
|
114
|
+
eventIds: finding.eventIds,
|
|
115
|
+
timelineItemIds: timelineItemIdsForEvents(finding.eventIds, eventsById),
|
|
116
|
+
evidence: redactedEvidence(finding.evidence),
|
|
117
|
+
remediation: finding.remediation,
|
|
118
|
+
}));
|
|
119
|
+
const jumpTargets = new Set();
|
|
120
|
+
for (const event of events) {
|
|
121
|
+
for (const timelineItemId of event.timelineItemIds) {
|
|
122
|
+
jumpTargets.add(`${event.id}->${timelineItemId}`);
|
|
123
|
+
}
|
|
124
|
+
}
|
|
125
|
+
for (const delta of replayDeltas) {
|
|
126
|
+
for (const timelineItemId of delta.timelineItemIds) {
|
|
127
|
+
jumpTargets.add(`${delta.id}->${timelineItemId}`);
|
|
128
|
+
}
|
|
129
|
+
}
|
|
130
|
+
for (const finding of scoreFindings) {
|
|
131
|
+
for (const timelineItemId of finding.timelineItemIds) {
|
|
132
|
+
jumpTargets.add(`${finding.ruleId}->${timelineItemId}`);
|
|
133
|
+
}
|
|
134
|
+
}
|
|
135
|
+
return {
|
|
136
|
+
schemaVersion: AGENT_TRAJECTORY_INSPECTION_SCHEMA,
|
|
137
|
+
trajectorySchemaVersion: trajectory.schemaVersion,
|
|
138
|
+
replaySchemaVersion: replay.schemaVersion,
|
|
139
|
+
scoreSchemaVersion: score.schemaVersion,
|
|
140
|
+
run: trajectory.run,
|
|
141
|
+
redaction: {
|
|
142
|
+
default: "redacted",
|
|
143
|
+
omitted: [
|
|
144
|
+
"raw prompts",
|
|
145
|
+
"raw tool arguments",
|
|
146
|
+
"raw tool outputs",
|
|
147
|
+
"full file diffs",
|
|
148
|
+
"timeline metadata values",
|
|
149
|
+
"secrets",
|
|
150
|
+
],
|
|
151
|
+
},
|
|
152
|
+
counts: {
|
|
153
|
+
timelineItems: timelineItems.length,
|
|
154
|
+
events: events.length,
|
|
155
|
+
replayDeltas: replayDeltas.length,
|
|
156
|
+
scoreFindings: scoreFindings.length,
|
|
157
|
+
scoreFailures: score.counts.failed,
|
|
158
|
+
scoreWarnings: score.counts.warnings,
|
|
159
|
+
jumpTargets: jumpTargets.size,
|
|
160
|
+
},
|
|
161
|
+
...(finalAnswer ? { finalAnswer } : {}),
|
|
162
|
+
timelineItems: timelineItems.map(inspectTimelineItem),
|
|
163
|
+
events,
|
|
164
|
+
replayDeltas,
|
|
165
|
+
scoreFindings,
|
|
166
|
+
};
|
|
167
|
+
}
|
|
168
|
+
//# sourceMappingURL=agent-trajectory-inspection.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agent-trajectory-inspection.js","sourceRoot":"","sources":["../../src/server/agent-trajectory-inspection.ts"],"names":[],"mappings":"AAcA,MAAM,CAAC,MAAM,kCAAkC,GAC9C,gDAAgD,CAAC;AA4GlD,SAAS,gBAAgB,CACxB,QAA0B;IAE1B,OAAO,QAAQ;SACb,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;QACjB,GAAG,MAAM;QACT,QAAQ,EAAE,IAAa;QACvB,KAAK,EAAE,GAAG,MAAM,CAAC,IAAI,IAAI,MAAM,CAAC,EAAE,EAAE;KACpC,CAAC,CAAC;SACF,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE;QACrB,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;QACtD,OAAO,SAAS,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACtE,CAAC,CAAC,CAAC;AACL,CAAC;AAED,SAAS,0BAA0B,CAAC,QAA0B;IAC7D,OAAO,QAAQ;SACb,MAAM,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,KAAK,eAAe,CAAC;SACnD,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,EAAE,CAAC;SAC1B,IAAI,EAAE,CAAC;AACV,CAAC;AAED,SAAS,wBAAwB,CAChC,QAAkB,EAClB,UAAuD;IAEvD,MAAM,GAAG,GAAG,IAAI,GAAG,EAAU,CAAC;IAC9B,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAChC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;QACtC,KAAK,MAAM,cAAc,IAAI,KAAK,EAAE,eAAe,IAAI,EAAE,EAAE,CAAC;YAC3D,GAAG,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC;QACzB,CAAC;IACF,CAAC;IACD,OAAO,CAAC,GAAG,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC;AACxB,CAAC;AAED,SAAS,YAAY,CAAC,IAA6B;IAClD,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,CAAC;IAC/B,IAAI,CAAC,QAAQ,IAAI,OAAO,QAAQ,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC;QAC1E,OAAO,EAAE,CAAC;IACX,CAAC;IACD,OAAO,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,IAAI,EAAE,CAAC;AACrC,CAAC;AAED,SAAS,mBAAmB,CAC3B,IAA6B;IAE7B,OAAO;QACN,EAAE,EAAE,IAAI,CAAC,EAAE;QACX,SAAS,EAAE,IAAI,CAAC,SAAS;QACzB,IAAI,EAAE,IAAI,CAAC,IAAI;QACf,MAAM,EAAE,IAAI,CAAC,MAAM;QACnB,UAAU,EAAE,IAAI,CAAC,UAAU;QAC3B,MAAM,EAAE,IAAI,CAAC,MAAM;QACnB,KAAK,EAAE,IAAI,CAAC,KAAK;QACjB,GAAG,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAClD,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACzC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACrD,GAAG,CAAC,IAAI,CAAC,kBAAkB;YAC1B,CAAC,CAAC,EAAE,kBAAkB,EAAE,IAAI,CAAC,kBAAkB,EAAE;YACjD,CAAC,CAAC,EAAE,CAAC;QACN,GAAG,CAAC,IAAI,CAAC,iBAAiB;YACzB,CAAC,CAAC,EAAE,iBAAiB,EAAE,IAAI,CAAC,iBAAiB,EAAE;YAC/C,CAAC,CAAC,EAAE,CAAC;QACN,YAAY,EAAE,YAAY,CAAC,IAAI,CAAC;QAChC,QAAQ,EAAE,IAAI;KACd,CAAC;AACH,CAAC;AAED,SAAS,YAAY,CACpB,KAA2B;IAE3B,OAAO;QACN,EAAE,EAAE,KAAK,CAAC,EAAE;QACZ,QAAQ,EAAE,KAAK,CAAC,QAAQ;QACxB,SAAS,EAAE,KAAK,CAAC,SAAS;QAC1B,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,MAAM,EAAE,KAAK,CAAC,MAAM;QACpB,UAAU,EAAE,KAAK,CAAC,UAAU;QAC5B,MAAM,EAAE,KAAK,CAAC,MAAM;QACpB,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACpD,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACvD,UAAU,EAAE,KAAK,CAAC,UAAU,IAAI,EAAE;QAClC,eAAe,EAAE,0BAA0B,CAAC,KAAK,CAAC,QAAQ,CAAC;QAC3D,QAAQ,EAAE,gBAAgB,CAAC,KAAK,CAAC,QAAQ,CAAC;KAC1C,CAAC;AACH,CAAC;AAED,SAAS,qBAAqB,CAC7B,MAAwC;IAExC,MAAM,cAAc,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,OAAO,EAAE,CAAC;IAC7C,MAAM,KAAK,GACV,cAAc,CAAC,IAAI,CAClB,CAAC,SAAS,EAAE,EAAE,CACb,SAAS,CAAC,KAAK,KAAK,WAAW;QAC/B,SAAS,CAAC,IAAI,KAAK,mBAAmB,CACvC;QACD,cAAc,CAAC,IAAI,CAClB,CAAC,SAAS,EAAE,EAAE,CACb,SAAS,CAAC,KAAK,KAAK,QAAQ,IAAI,SAAS,CAAC,KAAK,KAAK,SAAS,CAC9D,CAAC;IACH,IAAI,CAAC,KAAK;QAAE,OAAO,SAAS,CAAC;IAC7B,OAAO;QACN,OAAO,EAAE,KAAK,CAAC,EAAE;QACjB,eAAe,EAAE,KAAK,CAAC,eAAe;QACtC,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACpD,QAAQ,EAAE,IAAI;KACd,CAAC;AACH,CAAC;AAED,MAAM,UAAU,oCAAoC,CAAC,EACpD,aAAa,EACb,UAAU,EACV,MAAM,EACN,KAAK,GACwC;IAC7C,MAAM,MAAM,GAAG,UAAU,CAAC,MAAM,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC;IACnD,MAAM,WAAW,GAAG,qBAAqB,CAAC,MAAM,CAAC,CAAC;IAClD,MAAM,UAAU,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC;IACrE,MAAM,YAAY,GAAG,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;QAClD,EAAE,EAAE,KAAK,CAAC,EAAE;QACZ,QAAQ,EAAE,KAAK,CAAC,QAAQ;QACxB,MAAM,EAAE,KAAK,CAAC,MAAM;QACpB,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACpD,eAAe,EAAE,KAAK,CAAC,OAAO;YAC7B,CAAC,CAAC,wBAAwB,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,UAAU,CAAC;YACvD,CAAC,CAAC,0BAA0B,CAAC,KAAK,CAAC,QAAQ,CAAC;QAC7C,QAAQ,EAAE,gBAAgB,CAAC,KAAK,CAAC,QAAQ,CAAC;KAC1C,CAAC,CAAC,CAAC;IACJ,MAAM,aAAa,GAAG,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;QACtD,MAAM,EAAE,OAAO,CAAC,MAAM;QACtB,MAAM,EAAE,OAAO,CAAC,MAAM;QACtB,QAAQ,EAAE,OAAO,CAAC,QAAQ;QAC1B,OAAO,EAAE,OAAO,CAAC,OAAO;QACxB,QAAQ,EAAE,OAAO,CAAC,QAAQ;QAC1B,eAAe,EAAE,wBAAwB,CAAC,OAAO,CAAC,QAAQ,EAAE,UAAU,CAAC;QACvE,QAAQ,EAAE,gBAAgB,CAAC,OAAO,CAAC,QAAQ,CAAC;QAC5C,WAAW,EAAE,OAAO,CAAC,WAAW;KAChC,CAAC,CAAC,CAAC;IACJ,MAAM,WAAW,GAAG,IAAI,GAAG,EAAU,CAAC;IACtC,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;QAC5B,KAAK,MAAM,cAAc,IAAI,KAAK,CAAC,eAAe,EAAE,CAAC;YACpD,WAAW,CAAC,GAAG,CAAC,GAAG,KAAK,CAAC,EAAE,KAAK,cAAc,EAAE,CAAC,CAAC;QACnD,CAAC;IACF,CAAC;IACD,KAAK,MAAM,KAAK,IAAI,YAAY,EAAE,CAAC;QAClC,KAAK,MAAM,cAAc,IAAI,KAAK,CAAC,eAAe,EAAE,CAAC;YACpD,WAAW,CAAC,GAAG,CAAC,GAAG,KAAK,CAAC,EAAE,KAAK,cAAc,EAAE,CAAC,CAAC;QACnD,CAAC;IACF,CAAC;IACD,KAAK,MAAM,OAAO,IAAI,aAAa,EAAE,CAAC;QACrC,KAAK,MAAM,cAAc,IAAI,OAAO,CAAC,eAAe,EAAE,CAAC;YACtD,WAAW,CAAC,GAAG,CAAC,GAAG,OAAO,CAAC,MAAM,KAAK,cAAc,EAAE,CAAC,CAAC;QACzD,CAAC;IACF,CAAC;IAED,OAAO;QACN,aAAa,EAAE,kCAAkC;QACjD,uBAAuB,EAAE,UAAU,CAAC,aAAa;QACjD,mBAAmB,EAAE,MAAM,CAAC,aAAa;QACzC,kBAAkB,EAAE,KAAK,CAAC,aAAa;QACvC,GAAG,EAAE,UAAU,CAAC,GAAG;QACnB,SAAS,EAAE;YACV,OAAO,EAAE,UAAU;YACnB,OAAO,EAAE;gBACR,aAAa;gBACb,oBAAoB;gBACpB,kBAAkB;gBAClB,iBAAiB;gBACjB,0BAA0B;gBAC1B,SAAS;aACT;SACD;QACD,MAAM,EAAE;YACP,aAAa,EAAE,aAAa,CAAC,MAAM;YACnC,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,YAAY,EAAE,YAAY,CAAC,MAAM;YACjC,aAAa,EAAE,aAAa,CAAC,MAAM;YACnC,aAAa,EAAE,KAAK,CAAC,MAAM,CAAC,MAAM;YAClC,aAAa,EAAE,KAAK,CAAC,MAAM,CAAC,QAAQ;YACpC,WAAW,EAAE,WAAW,CAAC,IAAI;SAC7B;QACD,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACvC,aAAa,EAAE,aAAa,CAAC,GAAG,CAAC,mBAAmB,CAAC;QACrD,MAAM;QACN,YAAY;QACZ,aAAa;KACb,CAAC;AACH,CAAC","sourcesContent":["import type { ComposerRunTimelineItem } from \"@evalops/contracts\";\nimport type {\n\tAgentTrajectoryReplayDelta,\n\tAgentTrajectoryReplayReport,\n} from \"./agent-trajectory-replay.js\";\nimport type {\n\tAgentTrajectoryScoreFinding,\n\tAgentTrajectoryScoreReport,\n} from \"./agent-trajectory-scorers.js\";\nimport type {\n\tAgentTrajectoryEvent,\n\tAgentTrajectoryReport,\n} from \"./agent-trajectory.js\";\n\nexport const AGENT_TRAJECTORY_INSPECTION_SCHEMA =\n\t\"evalops.maestro.agent-trajectory-inspection.v1\";\n\ntype EvidenceAnchor = AgentTrajectoryEvent[\"evidence\"][number];\n\nexport interface AgentTrajectoryInspectionEvidenceAnchor\n\textends EvidenceAnchor {\n\tredacted: true;\n\tlabel: string;\n}\n\nexport interface AgentTrajectoryInspectionTimelineItem {\n\tid: string;\n\ttimestamp: string;\n\ttype: string;\n\tstatus: ComposerRunTimelineItem[\"status\"];\n\tvisibility: ComposerRunTimelineItem[\"visibility\"];\n\tsource: ComposerRunTimelineItem[\"source\"];\n\ttitle: string;\n\tsummary?: string;\n\trole?: ComposerRunTimelineItem[\"role\"];\n\ttoolName?: string;\n\tpendingRequestKind?: ComposerRunTimelineItem[\"pendingRequestKind\"];\n\tplatformOperation?: ComposerRunTimelineItem[\"platformOperation\"];\n\tmetadataKeys: string[];\n\tredacted: true;\n}\n\nexport interface AgentTrajectoryInspectionEvent {\n\tid: string;\n\tsequence: number;\n\ttimestamp: string;\n\tkind: AgentTrajectoryEvent[\"kind\"];\n\tphase: AgentTrajectoryEvent[\"phase\"];\n\tactor: AgentTrajectoryEvent[\"actor\"];\n\ttype: string;\n\tstatus: AgentTrajectoryEvent[\"status\"];\n\tvisibility: AgentTrajectoryEvent[\"visibility\"];\n\tsource: AgentTrajectoryEvent[\"source\"];\n\ttitle: string;\n\tsummary?: string;\n\ttoolName?: string;\n\trelatedIds: string[];\n\ttimelineItemIds: string[];\n\tevidence: AgentTrajectoryInspectionEvidenceAnchor[];\n}\n\nexport interface AgentTrajectoryInspectionReplayDelta {\n\tid: string;\n\tseverity: AgentTrajectoryReplayDelta[\"severity\"];\n\truleId: string;\n\tmessage: string;\n\teventId?: string;\n\ttimelineItemIds: string[];\n\tevidence: AgentTrajectoryInspectionEvidenceAnchor[];\n}\n\nexport interface AgentTrajectoryInspectionScoreFinding {\n\truleId: string;\n\tstatus: AgentTrajectoryScoreFinding[\"status\"];\n\tseverity: AgentTrajectoryScoreFinding[\"severity\"];\n\tmessage: string;\n\teventIds: string[];\n\ttimelineItemIds: string[];\n\tevidence: AgentTrajectoryInspectionEvidenceAnchor[];\n\tremediation: string;\n}\n\nexport interface AgentTrajectoryInspectionFinalAnswer {\n\teventId: string;\n\ttimelineItemIds: string[];\n\ttitle: string;\n\tsummary?: string;\n\tredacted: true;\n}\n\nexport interface AgentTrajectoryInspectionReport {\n\tschemaVersion: typeof AGENT_TRAJECTORY_INSPECTION_SCHEMA;\n\ttrajectorySchemaVersion: AgentTrajectoryReport[\"schemaVersion\"];\n\treplaySchemaVersion: AgentTrajectoryReplayReport[\"schemaVersion\"];\n\tscoreSchemaVersion: AgentTrajectoryScoreReport[\"schemaVersion\"];\n\trun: AgentTrajectoryReport[\"run\"];\n\tredaction: {\n\t\tdefault: \"redacted\";\n\t\tomitted: string[];\n\t};\n\tcounts: {\n\t\ttimelineItems: number;\n\t\tevents: number;\n\t\treplayDeltas: number;\n\t\tscoreFindings: number;\n\t\tscoreFailures: number;\n\t\tscoreWarnings: number;\n\t\tjumpTargets: number;\n\t};\n\tfinalAnswer?: AgentTrajectoryInspectionFinalAnswer;\n\ttimelineItems: AgentTrajectoryInspectionTimelineItem[];\n\tevents: AgentTrajectoryInspectionEvent[];\n\treplayDeltas: AgentTrajectoryInspectionReplayDelta[];\n\tscoreFindings: AgentTrajectoryInspectionScoreFinding[];\n}\n\nexport interface BuildAgentTrajectoryInspectionReportOptions {\n\ttimelineItems: ComposerRunTimelineItem[];\n\ttrajectory: AgentTrajectoryReport;\n\treplay: AgentTrajectoryReplayReport;\n\tscore: AgentTrajectoryScoreReport;\n}\n\nfunction redactedEvidence(\n\tevidence: EvidenceAnchor[],\n): AgentTrajectoryInspectionEvidenceAnchor[] {\n\treturn evidence\n\t\t.map((anchor) => ({\n\t\t\t...anchor,\n\t\t\tredacted: true as const,\n\t\t\tlabel: `${anchor.kind}:${anchor.id}`,\n\t\t}))\n\t\t.sort((left, right) => {\n\t\t\tconst kindDelta = left.kind.localeCompare(right.kind);\n\t\t\treturn kindDelta === 0 ? left.id.localeCompare(right.id) : kindDelta;\n\t\t});\n}\n\nfunction timelineItemIdsForEvidence(evidence: EvidenceAnchor[]): string[] {\n\treturn evidence\n\t\t.filter((anchor) => anchor.kind === \"timeline_item\")\n\t\t.map((anchor) => anchor.id)\n\t\t.sort();\n}\n\nfunction timelineItemIdsForEvents(\n\teventIds: string[],\n\teventsById: Map<string, AgentTrajectoryInspectionEvent>,\n): string[] {\n\tconst ids = new Set<string>();\n\tfor (const eventId of eventIds) {\n\t\tconst event = eventsById.get(eventId);\n\t\tfor (const timelineItemId of event?.timelineItemIds ?? []) {\n\t\t\tids.add(timelineItemId);\n\t\t}\n\t}\n\treturn [...ids].sort();\n}\n\nfunction metadataKeys(item: ComposerRunTimelineItem): string[] {\n\tconst metadata = item.metadata;\n\tif (!metadata || typeof metadata !== \"object\" || Array.isArray(metadata)) {\n\t\treturn [];\n\t}\n\treturn Object.keys(metadata).sort();\n}\n\nfunction inspectTimelineItem(\n\titem: ComposerRunTimelineItem,\n): AgentTrajectoryInspectionTimelineItem {\n\treturn {\n\t\tid: item.id,\n\t\ttimestamp: item.timestamp,\n\t\ttype: item.type,\n\t\tstatus: item.status,\n\t\tvisibility: item.visibility,\n\t\tsource: item.source,\n\t\ttitle: item.title,\n\t\t...(item.summary ? { summary: item.summary } : {}),\n\t\t...(item.role ? { role: item.role } : {}),\n\t\t...(item.toolName ? { toolName: item.toolName } : {}),\n\t\t...(item.pendingRequestKind\n\t\t\t? { pendingRequestKind: item.pendingRequestKind }\n\t\t\t: {}),\n\t\t...(item.platformOperation\n\t\t\t? { platformOperation: item.platformOperation }\n\t\t\t: {}),\n\t\tmetadataKeys: metadataKeys(item),\n\t\tredacted: true,\n\t};\n}\n\nfunction inspectEvent(\n\tevent: AgentTrajectoryEvent,\n): AgentTrajectoryInspectionEvent {\n\treturn {\n\t\tid: event.id,\n\t\tsequence: event.sequence,\n\t\ttimestamp: event.timestamp,\n\t\tkind: event.kind,\n\t\tphase: event.phase,\n\t\tactor: event.actor,\n\t\ttype: event.type,\n\t\tstatus: event.status,\n\t\tvisibility: event.visibility,\n\t\tsource: event.source,\n\t\ttitle: event.title,\n\t\t...(event.summary ? { summary: event.summary } : {}),\n\t\t...(event.toolName ? { toolName: event.toolName } : {}),\n\t\trelatedIds: event.relatedIds ?? [],\n\t\ttimelineItemIds: timelineItemIdsForEvidence(event.evidence),\n\t\tevidence: redactedEvidence(event.evidence),\n\t};\n}\n\nfunction finalAnswerFromEvents(\n\tevents: AgentTrajectoryInspectionEvent[],\n): AgentTrajectoryInspectionFinalAnswer | undefined {\n\tconst reversedEvents = [...events].reverse();\n\tconst event =\n\t\treversedEvents.find(\n\t\t\t(candidate) =>\n\t\t\t\tcandidate.actor === \"assistant\" &&\n\t\t\t\tcandidate.type === \"message.assistant\",\n\t\t) ??\n\t\treversedEvents.find(\n\t\t\t(candidate) =>\n\t\t\t\tcandidate.phase === \"finish\" || candidate.phase === \"recover\",\n\t\t);\n\tif (!event) return undefined;\n\treturn {\n\t\teventId: event.id,\n\t\ttimelineItemIds: event.timelineItemIds,\n\t\ttitle: event.title,\n\t\t...(event.summary ? { summary: event.summary } : {}),\n\t\tredacted: true,\n\t};\n}\n\nexport function buildAgentTrajectoryInspectionReport({\n\ttimelineItems,\n\ttrajectory,\n\treplay,\n\tscore,\n}: BuildAgentTrajectoryInspectionReportOptions): AgentTrajectoryInspectionReport {\n\tconst events = trajectory.events.map(inspectEvent);\n\tconst finalAnswer = finalAnswerFromEvents(events);\n\tconst eventsById = new Map(events.map((event) => [event.id, event]));\n\tconst replayDeltas = replay.deltas.map((delta) => ({\n\t\tid: delta.id,\n\t\tseverity: delta.severity,\n\t\truleId: delta.ruleId,\n\t\tmessage: delta.message,\n\t\t...(delta.eventId ? { eventId: delta.eventId } : {}),\n\t\ttimelineItemIds: delta.eventId\n\t\t\t? timelineItemIdsForEvents([delta.eventId], eventsById)\n\t\t\t: timelineItemIdsForEvidence(delta.evidence),\n\t\tevidence: redactedEvidence(delta.evidence),\n\t}));\n\tconst scoreFindings = score.findings.map((finding) => ({\n\t\truleId: finding.ruleId,\n\t\tstatus: finding.status,\n\t\tseverity: finding.severity,\n\t\tmessage: finding.message,\n\t\teventIds: finding.eventIds,\n\t\ttimelineItemIds: timelineItemIdsForEvents(finding.eventIds, eventsById),\n\t\tevidence: redactedEvidence(finding.evidence),\n\t\tremediation: finding.remediation,\n\t}));\n\tconst jumpTargets = new Set<string>();\n\tfor (const event of events) {\n\t\tfor (const timelineItemId of event.timelineItemIds) {\n\t\t\tjumpTargets.add(`${event.id}->${timelineItemId}`);\n\t\t}\n\t}\n\tfor (const delta of replayDeltas) {\n\t\tfor (const timelineItemId of delta.timelineItemIds) {\n\t\t\tjumpTargets.add(`${delta.id}->${timelineItemId}`);\n\t\t}\n\t}\n\tfor (const finding of scoreFindings) {\n\t\tfor (const timelineItemId of finding.timelineItemIds) {\n\t\t\tjumpTargets.add(`${finding.ruleId}->${timelineItemId}`);\n\t\t}\n\t}\n\n\treturn {\n\t\tschemaVersion: AGENT_TRAJECTORY_INSPECTION_SCHEMA,\n\t\ttrajectorySchemaVersion: trajectory.schemaVersion,\n\t\treplaySchemaVersion: replay.schemaVersion,\n\t\tscoreSchemaVersion: score.schemaVersion,\n\t\trun: trajectory.run,\n\t\tredaction: {\n\t\t\tdefault: \"redacted\",\n\t\t\tomitted: [\n\t\t\t\t\"raw prompts\",\n\t\t\t\t\"raw tool arguments\",\n\t\t\t\t\"raw tool outputs\",\n\t\t\t\t\"full file diffs\",\n\t\t\t\t\"timeline metadata values\",\n\t\t\t\t\"secrets\",\n\t\t\t],\n\t\t},\n\t\tcounts: {\n\t\t\ttimelineItems: timelineItems.length,\n\t\t\tevents: events.length,\n\t\t\treplayDeltas: replayDeltas.length,\n\t\t\tscoreFindings: scoreFindings.length,\n\t\t\tscoreFailures: score.counts.failed,\n\t\t\tscoreWarnings: score.counts.warnings,\n\t\t\tjumpTargets: jumpTargets.size,\n\t\t},\n\t\t...(finalAnswer ? { finalAnswer } : {}),\n\t\ttimelineItems: timelineItems.map(inspectTimelineItem),\n\t\tevents,\n\t\treplayDeltas,\n\t\tscoreFindings,\n\t};\n}\n"]}
|
|
@@ -0,0 +1,56 @@
|
|
|
1
|
+
import type { AgentTrajectoryEvent, AgentTrajectoryEventKind, AgentTrajectoryPhase, AgentTrajectoryReport } from "./agent-trajectory.js";
|
|
2
|
+
export declare const AGENT_TRAJECTORY_REPLAY_SCHEMA = "evalops.maestro.agent-trajectory-replay.v1";
|
|
3
|
+
export type AgentTrajectoryReplaySeverity = "error" | "warning";
|
|
4
|
+
export interface AgentTrajectoryReplayDelta {
|
|
5
|
+
id: string;
|
|
6
|
+
severity: AgentTrajectoryReplaySeverity;
|
|
7
|
+
ruleId: string;
|
|
8
|
+
sequence?: number;
|
|
9
|
+
eventId?: string;
|
|
10
|
+
phase?: AgentTrajectoryPhase;
|
|
11
|
+
kind?: AgentTrajectoryEventKind;
|
|
12
|
+
expected?: string;
|
|
13
|
+
observed?: string;
|
|
14
|
+
message: string;
|
|
15
|
+
evidence: AgentTrajectoryEvent["evidence"];
|
|
16
|
+
}
|
|
17
|
+
export interface AgentTrajectoryReplayPhaseSummary {
|
|
18
|
+
phase: AgentTrajectoryPhase;
|
|
19
|
+
events: number;
|
|
20
|
+
firstSequence: number;
|
|
21
|
+
lastSequence: number;
|
|
22
|
+
}
|
|
23
|
+
export interface AgentTrajectoryReplayToolSummary {
|
|
24
|
+
toolCallId: string;
|
|
25
|
+
toolName?: string;
|
|
26
|
+
requestedSequence?: number;
|
|
27
|
+
resultSequences: number[];
|
|
28
|
+
terminalStatus?: "completed" | "failed";
|
|
29
|
+
evidence: AgentTrajectoryEvent["evidence"];
|
|
30
|
+
}
|
|
31
|
+
export interface AgentTrajectoryReplayReport {
|
|
32
|
+
schemaVersion: typeof AGENT_TRAJECTORY_REPLAY_SCHEMA;
|
|
33
|
+
trajectorySchemaVersion: AgentTrajectoryReport["schemaVersion"];
|
|
34
|
+
run: AgentTrajectoryReport["run"];
|
|
35
|
+
deterministic: true;
|
|
36
|
+
counts: {
|
|
37
|
+
events: number;
|
|
38
|
+
deltas: number;
|
|
39
|
+
errors: number;
|
|
40
|
+
warnings: number;
|
|
41
|
+
toolCalls: number;
|
|
42
|
+
phases: number;
|
|
43
|
+
};
|
|
44
|
+
phases: AgentTrajectoryReplayPhaseSummary[];
|
|
45
|
+
toolCalls: AgentTrajectoryReplayToolSummary[];
|
|
46
|
+
deltas: AgentTrajectoryReplayDelta[];
|
|
47
|
+
}
|
|
48
|
+
export interface AgentTrajectoryReplayToolExpectation {
|
|
49
|
+
terminalStatus?: "completed" | "failed";
|
|
50
|
+
requiredArtifactIds?: string[];
|
|
51
|
+
}
|
|
52
|
+
export interface ReplayAgentTrajectoryOptions {
|
|
53
|
+
expectedTools?: Record<string, AgentTrajectoryReplayToolExpectation>;
|
|
54
|
+
}
|
|
55
|
+
export declare function replayAgentTrajectoryReport(report: AgentTrajectoryReport, options?: ReplayAgentTrajectoryOptions): AgentTrajectoryReplayReport;
|
|
56
|
+
//# sourceMappingURL=agent-trajectory-replay.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agent-trajectory-replay.d.ts","sourceRoot":"","sources":["../../src/server/agent-trajectory-replay.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EACX,oBAAoB,EACpB,wBAAwB,EACxB,oBAAoB,EACpB,qBAAqB,EACrB,MAAM,uBAAuB,CAAC;AAE/B,eAAO,MAAM,8BAA8B,+CACE,CAAC;AAE9C,MAAM,MAAM,6BAA6B,GAAG,OAAO,GAAG,SAAS,CAAC;AAEhE,MAAM,WAAW,0BAA0B;IAC1C,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,6BAA6B,CAAC;IACxC,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,oBAAoB,CAAC;IAC7B,IAAI,CAAC,EAAE,wBAAwB,CAAC;IAChC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,QAAQ,EAAE,oBAAoB,CAAC,UAAU,CAAC,CAAC;CAC3C;AAED,MAAM,WAAW,iCAAiC;IACjD,KAAK,EAAE,oBAAoB,CAAC;IAC5B,MAAM,EAAE,MAAM,CAAC;IACf,aAAa,EAAE,MAAM,CAAC;IACtB,YAAY,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,gCAAgC;IAChD,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,cAAc,CAAC,EAAE,WAAW,GAAG,QAAQ,CAAC;IACxC,QAAQ,EAAE,oBAAoB,CAAC,UAAU,CAAC,CAAC;CAC3C;AAED,MAAM,WAAW,2BAA2B;IAC3C,aAAa,EAAE,OAAO,8BAA8B,CAAC;IACrD,uBAAuB,EAAE,qBAAqB,CAAC,eAAe,CAAC,CAAC;IAChE,GAAG,EAAE,qBAAqB,CAAC,KAAK,CAAC,CAAC;IAClC,aAAa,EAAE,IAAI,CAAC;IACpB,MAAM,EAAE;QACP,MAAM,EAAE,MAAM,CAAC;QACf,MAAM,EAAE,MAAM,CAAC;QACf,MAAM,EAAE,MAAM,CAAC;QACf,QAAQ,EAAE,MAAM,CAAC;QACjB,SAAS,EAAE,MAAM,CAAC;QAClB,MAAM,EAAE,MAAM,CAAC;KACf,CAAC;IACF,MAAM,EAAE,iCAAiC,EAAE,CAAC;IAC5C,SAAS,EAAE,gCAAgC,EAAE,CAAC;IAC9C,MAAM,EAAE,0BAA0B,EAAE,CAAC;CACrC;AAED,MAAM,WAAW,oCAAoC;IACpD,cAAc,CAAC,EAAE,WAAW,GAAG,QAAQ,CAAC;IACxC,mBAAmB,CAAC,EAAE,MAAM,EAAE,CAAC;CAC/B;AAED,MAAM,WAAW,4BAA4B;IAC5C,aAAa,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,oCAAoC,CAAC,CAAC;CACrE;AA+HD,wBAAgB,2BAA2B,CAC1C,MAAM,EAAE,qBAAqB,EAC7B,OAAO,GAAE,4BAAiC,GACxC,2BAA2B,CAsI7B"}
|
|
@@ -0,0 +1,214 @@
|
|
|
1
|
+
import { validateAgentTrajectoryReport } from "./agent-trajectory-validation.js";
|
|
2
|
+
export const AGENT_TRAJECTORY_REPLAY_SCHEMA = "evalops.maestro.agent-trajectory-replay.v1";
|
|
3
|
+
function evidenceIds(event, kind) {
|
|
4
|
+
return event.evidence
|
|
5
|
+
.filter((anchor) => anchor.kind === kind)
|
|
6
|
+
.map((anchor) => anchor.id)
|
|
7
|
+
.sort();
|
|
8
|
+
}
|
|
9
|
+
function firstEvidenceId(event, kind) {
|
|
10
|
+
return evidenceIds(event, kind)[0];
|
|
11
|
+
}
|
|
12
|
+
function mergeEvidence(target, source) {
|
|
13
|
+
const seen = new Set(target.map((anchor) => `${anchor.kind}:${anchor.id}`));
|
|
14
|
+
for (const anchor of source) {
|
|
15
|
+
const key = `${anchor.kind}:${anchor.id}`;
|
|
16
|
+
if (seen.has(key))
|
|
17
|
+
continue;
|
|
18
|
+
seen.add(key);
|
|
19
|
+
target.push(anchor);
|
|
20
|
+
}
|
|
21
|
+
return target.sort((a, b) => {
|
|
22
|
+
const kindDelta = a.kind.localeCompare(b.kind);
|
|
23
|
+
return kindDelta === 0 ? a.id.localeCompare(b.id) : kindDelta;
|
|
24
|
+
});
|
|
25
|
+
}
|
|
26
|
+
function delta(deltas, params) {
|
|
27
|
+
deltas.push({
|
|
28
|
+
id: `delta:${String(deltas.length + 1).padStart(3, "0")}`,
|
|
29
|
+
...params,
|
|
30
|
+
});
|
|
31
|
+
}
|
|
32
|
+
function ensureToolState(tools, toolCallId, event) {
|
|
33
|
+
let state = tools.get(toolCallId);
|
|
34
|
+
if (!state) {
|
|
35
|
+
state = {
|
|
36
|
+
toolCallId,
|
|
37
|
+
toolName: event.toolName,
|
|
38
|
+
resultSequences: [],
|
|
39
|
+
evidence: [],
|
|
40
|
+
artifactIds: new Set(),
|
|
41
|
+
};
|
|
42
|
+
tools.set(toolCallId, state);
|
|
43
|
+
}
|
|
44
|
+
if (!state.toolName && event.toolName) {
|
|
45
|
+
state.toolName = event.toolName;
|
|
46
|
+
}
|
|
47
|
+
mergeEvidence(state.evidence, event.evidence);
|
|
48
|
+
return state;
|
|
49
|
+
}
|
|
50
|
+
function summarizePhases(events) {
|
|
51
|
+
const byPhase = new Map();
|
|
52
|
+
for (const event of events) {
|
|
53
|
+
const existing = byPhase.get(event.phase);
|
|
54
|
+
if (existing) {
|
|
55
|
+
existing.events += 1;
|
|
56
|
+
existing.lastSequence = event.sequence;
|
|
57
|
+
continue;
|
|
58
|
+
}
|
|
59
|
+
byPhase.set(event.phase, {
|
|
60
|
+
phase: event.phase,
|
|
61
|
+
events: 1,
|
|
62
|
+
firstSequence: event.sequence,
|
|
63
|
+
lastSequence: event.sequence,
|
|
64
|
+
});
|
|
65
|
+
}
|
|
66
|
+
return [...byPhase.values()].sort((a, b) => a.firstSequence - b.firstSequence);
|
|
67
|
+
}
|
|
68
|
+
function summarizeTools(tools) {
|
|
69
|
+
return [...tools.values()]
|
|
70
|
+
.map((tool) => ({
|
|
71
|
+
toolCallId: tool.toolCallId,
|
|
72
|
+
...(tool.toolName ? { toolName: tool.toolName } : {}),
|
|
73
|
+
...(tool.requestedSequence !== undefined
|
|
74
|
+
? { requestedSequence: tool.requestedSequence }
|
|
75
|
+
: {}),
|
|
76
|
+
resultSequences: [...tool.resultSequences].sort((a, b) => a - b),
|
|
77
|
+
...(tool.terminalStatus ? { terminalStatus: tool.terminalStatus } : {}),
|
|
78
|
+
evidence: [...tool.evidence],
|
|
79
|
+
}))
|
|
80
|
+
.sort((a, b) => {
|
|
81
|
+
const sequenceA = a.requestedSequence ?? Number.MAX_SAFE_INTEGER;
|
|
82
|
+
const sequenceB = b.requestedSequence ?? Number.MAX_SAFE_INTEGER;
|
|
83
|
+
if (sequenceA !== sequenceB)
|
|
84
|
+
return sequenceA - sequenceB;
|
|
85
|
+
return a.toolCallId.localeCompare(b.toolCallId);
|
|
86
|
+
});
|
|
87
|
+
}
|
|
88
|
+
export function replayAgentTrajectoryReport(report, options = {}) {
|
|
89
|
+
const deltas = [];
|
|
90
|
+
const validation = validateAgentTrajectoryReport(report);
|
|
91
|
+
for (const failure of validation.failures) {
|
|
92
|
+
delta(deltas, {
|
|
93
|
+
severity: "error",
|
|
94
|
+
ruleId: "trajectory.validation",
|
|
95
|
+
message: failure,
|
|
96
|
+
evidence: [],
|
|
97
|
+
});
|
|
98
|
+
}
|
|
99
|
+
const tools = new Map();
|
|
100
|
+
for (const event of report.events) {
|
|
101
|
+
const toolCallIds = evidenceIds(event, "tool_call");
|
|
102
|
+
const artifactIds = evidenceIds(event, "artifact");
|
|
103
|
+
if (event.type === "tool.requested") {
|
|
104
|
+
const toolCallId = firstEvidenceId(event, "tool_call");
|
|
105
|
+
if (!toolCallId)
|
|
106
|
+
continue;
|
|
107
|
+
const state = ensureToolState(tools, toolCallId, event);
|
|
108
|
+
if (state.requestedSequence !== undefined) {
|
|
109
|
+
delta(deltas, {
|
|
110
|
+
severity: "error",
|
|
111
|
+
ruleId: "tool.duplicate_request",
|
|
112
|
+
sequence: event.sequence,
|
|
113
|
+
eventId: event.id,
|
|
114
|
+
phase: event.phase,
|
|
115
|
+
kind: event.kind,
|
|
116
|
+
expected: "one tool.requested per tool call",
|
|
117
|
+
observed: `previous request at sequence ${state.requestedSequence}`,
|
|
118
|
+
message: `Tool call ${toolCallId} was requested more than once.`,
|
|
119
|
+
evidence: event.evidence,
|
|
120
|
+
});
|
|
121
|
+
}
|
|
122
|
+
state.requestedSequence = event.sequence;
|
|
123
|
+
}
|
|
124
|
+
if (event.type === "tool.completed" || event.type === "tool.failed") {
|
|
125
|
+
const toolCallId = firstEvidenceId(event, "tool_call");
|
|
126
|
+
if (!toolCallId)
|
|
127
|
+
continue;
|
|
128
|
+
const state = ensureToolState(tools, toolCallId, event);
|
|
129
|
+
const terminalStatus = event.type === "tool.completed" ? "completed" : "failed";
|
|
130
|
+
if (state.terminalStatus !== undefined &&
|
|
131
|
+
state.terminalStatus !== terminalStatus) {
|
|
132
|
+
delta(deltas, {
|
|
133
|
+
severity: "error",
|
|
134
|
+
ruleId: "tool.conflicting_terminal_status",
|
|
135
|
+
sequence: event.sequence,
|
|
136
|
+
eventId: event.id,
|
|
137
|
+
phase: event.phase,
|
|
138
|
+
kind: event.kind,
|
|
139
|
+
expected: state.terminalStatus,
|
|
140
|
+
observed: terminalStatus,
|
|
141
|
+
message: `Tool call ${toolCallId} has conflicting terminal statuses.`,
|
|
142
|
+
evidence: event.evidence,
|
|
143
|
+
});
|
|
144
|
+
}
|
|
145
|
+
state.terminalStatus = terminalStatus;
|
|
146
|
+
state.resultSequences.push(event.sequence);
|
|
147
|
+
}
|
|
148
|
+
for (const toolCallId of toolCallIds) {
|
|
149
|
+
const state = ensureToolState(tools, toolCallId, event);
|
|
150
|
+
for (const artifactId of artifactIds) {
|
|
151
|
+
state.artifactIds.add(artifactId);
|
|
152
|
+
}
|
|
153
|
+
}
|
|
154
|
+
}
|
|
155
|
+
for (const [toolCallId, expectation] of Object.entries(options.expectedTools ?? {}).sort(([left], [right]) => left.localeCompare(right))) {
|
|
156
|
+
const state = tools.get(toolCallId);
|
|
157
|
+
if (!state) {
|
|
158
|
+
delta(deltas, {
|
|
159
|
+
severity: "error",
|
|
160
|
+
ruleId: "tool.expected_missing",
|
|
161
|
+
expected: toolCallId,
|
|
162
|
+
observed: "missing",
|
|
163
|
+
message: `Expected tool call ${toolCallId} was not present in the trajectory.`,
|
|
164
|
+
evidence: [],
|
|
165
|
+
});
|
|
166
|
+
continue;
|
|
167
|
+
}
|
|
168
|
+
if (expectation.terminalStatus &&
|
|
169
|
+
state.terminalStatus !== expectation.terminalStatus) {
|
|
170
|
+
delta(deltas, {
|
|
171
|
+
severity: "error",
|
|
172
|
+
ruleId: "tool.terminal_status_mismatch",
|
|
173
|
+
expected: expectation.terminalStatus,
|
|
174
|
+
observed: state.terminalStatus ?? "missing",
|
|
175
|
+
message: `Tool call ${toolCallId} terminal status did not match replay expectation.`,
|
|
176
|
+
evidence: state.evidence,
|
|
177
|
+
});
|
|
178
|
+
}
|
|
179
|
+
for (const artifactId of expectation.requiredArtifactIds ?? []) {
|
|
180
|
+
if (state.artifactIds.has(artifactId))
|
|
181
|
+
continue;
|
|
182
|
+
delta(deltas, {
|
|
183
|
+
severity: "error",
|
|
184
|
+
ruleId: "tool.required_artifact_missing",
|
|
185
|
+
expected: artifactId,
|
|
186
|
+
observed: "missing",
|
|
187
|
+
message: `Tool call ${toolCallId} did not produce required artifact ${artifactId}.`,
|
|
188
|
+
evidence: state.evidence,
|
|
189
|
+
});
|
|
190
|
+
}
|
|
191
|
+
}
|
|
192
|
+
const errors = deltas.filter((item) => item.severity === "error").length;
|
|
193
|
+
const warnings = deltas.filter((item) => item.severity === "warning").length;
|
|
194
|
+
const phases = summarizePhases(report.events);
|
|
195
|
+
const toolCalls = summarizeTools(tools);
|
|
196
|
+
return {
|
|
197
|
+
schemaVersion: AGENT_TRAJECTORY_REPLAY_SCHEMA,
|
|
198
|
+
trajectorySchemaVersion: report.schemaVersion,
|
|
199
|
+
run: report.run,
|
|
200
|
+
deterministic: true,
|
|
201
|
+
counts: {
|
|
202
|
+
events: report.events.length,
|
|
203
|
+
deltas: deltas.length,
|
|
204
|
+
errors,
|
|
205
|
+
warnings,
|
|
206
|
+
toolCalls: toolCalls.length,
|
|
207
|
+
phases: phases.length,
|
|
208
|
+
},
|
|
209
|
+
phases,
|
|
210
|
+
toolCalls,
|
|
211
|
+
deltas,
|
|
212
|
+
};
|
|
213
|
+
}
|
|
214
|
+
//# sourceMappingURL=agent-trajectory-replay.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agent-trajectory-replay.js","sourceRoot":"","sources":["../../src/server/agent-trajectory-replay.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,6BAA6B,EAAE,MAAM,kCAAkC,CAAC;AAQjF,MAAM,CAAC,MAAM,8BAA8B,GAC1C,4CAA4C,CAAC;AAuE9C,SAAS,WAAW,CACnB,KAA2B,EAC3B,IAAsD;IAEtD,OAAO,KAAK,CAAC,QAAQ;SACnB,MAAM,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,KAAK,IAAI,CAAC;SACxC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,EAAE,CAAC;SAC1B,IAAI,EAAE,CAAC;AACV,CAAC;AAED,SAAS,eAAe,CACvB,KAA2B,EAC3B,IAAsD;IAEtD,OAAO,WAAW,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;AACpC,CAAC;AAED,SAAS,aAAa,CACrB,MAAwC,EACxC,MAAwC;IAExC,MAAM,IAAI,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,GAAG,MAAM,CAAC,IAAI,IAAI,MAAM,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;IAC5E,KAAK,MAAM,MAAM,IAAI,MAAM,EAAE,CAAC;QAC7B,MAAM,GAAG,GAAG,GAAG,MAAM,CAAC,IAAI,IAAI,MAAM,CAAC,EAAE,EAAE,CAAC;QAC1C,IAAI,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC;YAAE,SAAS;QAC5B,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACd,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IACrB,CAAC;IACD,OAAO,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;QAC3B,MAAM,SAAS,GAAG,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;QAC/C,OAAO,SAAS,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IAC/D,CAAC,CAAC,CAAC;AACJ,CAAC;AAED,SAAS,KAAK,CACb,MAAoC,EACpC,MAA8C;IAE9C,MAAM,CAAC,IAAI,CAAC;QACX,EAAE,EAAE,SAAS,MAAM,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE;QACzD,GAAG,MAAM;KACT,CAAC,CAAC;AACJ,CAAC;AAED,SAAS,eAAe,CACvB,KAA6B,EAC7B,UAAkB,EAClB,KAA2B;IAE3B,IAAI,KAAK,GAAG,KAAK,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC;IAClC,IAAI,CAAC,KAAK,EAAE,CAAC;QACZ,KAAK,GAAG;YACP,UAAU;YACV,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,eAAe,EAAE,EAAE;YACnB,QAAQ,EAAE,EAAE;YACZ,WAAW,EAAE,IAAI,GAAG,EAAU;SAC9B,CAAC;QACF,KAAK,CAAC,GAAG,CAAC,UAAU,EAAE,KAAK,CAAC,CAAC;IAC9B,CAAC;IACD,IAAI,CAAC,KAAK,CAAC,QAAQ,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC;QACvC,KAAK,CAAC,QAAQ,GAAG,KAAK,CAAC,QAAQ,CAAC;IACjC,CAAC;IACD,aAAa,CAAC,KAAK,CAAC,QAAQ,EAAE,KAAK,CAAC,QAAQ,CAAC,CAAC;IAC9C,OAAO,KAAK,CAAC;AACd,CAAC;AAED,SAAS,eAAe,CACvB,MAA8B;IAE9B,MAAM,OAAO,GAAG,IAAI,GAAG,EAGpB,CAAC;IACJ,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;QAC5B,MAAM,QAAQ,GAAG,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;QAC1C,IAAI,QAAQ,EAAE,CAAC;YACd,QAAQ,CAAC,MAAM,IAAI,CAAC,CAAC;YACrB,QAAQ,CAAC,YAAY,GAAG,KAAK,CAAC,QAAQ,CAAC;YACvC,SAAS;QACV,CAAC;QACD,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,EAAE;YACxB,KAAK,EAAE,KAAK,CAAC,KAAK;YAClB,MAAM,EAAE,CAAC;YACT,aAAa,EAAE,KAAK,CAAC,QAAQ;YAC7B,YAAY,EAAE,KAAK,CAAC,QAAQ;SAC5B,CAAC,CAAC;IACJ,CAAC;IACD,OAAO,CAAC,GAAG,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,IAAI,CAChC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,aAAa,GAAG,CAAC,CAAC,aAAa,CAC3C,CAAC;AACH,CAAC;AAED,SAAS,cAAc,CACtB,KAA6B;IAE7B,OAAO,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC;SACxB,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC;QACf,UAAU,EAAE,IAAI,CAAC,UAAU;QAC3B,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACrD,GAAG,CAAC,IAAI,CAAC,iBAAiB,KAAK,SAAS;YACvC,CAAC,CAAC,EAAE,iBAAiB,EAAE,IAAI,CAAC,iBAAiB,EAAE;YAC/C,CAAC,CAAC,EAAE,CAAC;QACN,eAAe,EAAE,CAAC,GAAG,IAAI,CAAC,eAAe,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC;QAChE,GAAG,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC,CAAC,EAAE,cAAc,EAAE,IAAI,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACvE,QAAQ,EAAE,CAAC,GAAG,IAAI,CAAC,QAAQ,CAAC;KAC5B,CAAC,CAAC;SACF,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;QACd,MAAM,SAAS,GAAG,CAAC,CAAC,iBAAiB,IAAI,MAAM,CAAC,gBAAgB,CAAC;QACjE,MAAM,SAAS,GAAG,CAAC,CAAC,iBAAiB,IAAI,MAAM,CAAC,gBAAgB,CAAC;QACjE,IAAI,SAAS,KAAK,SAAS;YAAE,OAAO,SAAS,GAAG,SAAS,CAAC;QAC1D,OAAO,CAAC,CAAC,UAAU,CAAC,aAAa,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC;IACjD,CAAC,CAAC,CAAC;AACL,CAAC;AAED,MAAM,UAAU,2BAA2B,CAC1C,MAA6B,EAC7B,UAAwC,EAAE;IAE1C,MAAM,MAAM,GAAiC,EAAE,CAAC;IAChD,MAAM,UAAU,GAAG,6BAA6B,CAAC,MAAM,CAAC,CAAC;IACzD,KAAK,MAAM,OAAO,IAAI,UAAU,CAAC,QAAQ,EAAE,CAAC;QAC3C,KAAK,CAAC,MAAM,EAAE;YACb,QAAQ,EAAE,OAAO;YACjB,MAAM,EAAE,uBAAuB;YAC/B,OAAO,EAAE,OAAO;YAChB,QAAQ,EAAE,EAAE;SACZ,CAAC,CAAC;IACJ,CAAC;IAED,MAAM,KAAK,GAAG,IAAI,GAAG,EAAqB,CAAC;IAC3C,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC;QACnC,MAAM,WAAW,GAAG,WAAW,CAAC,KAAK,EAAE,WAAW,CAAC,CAAC;QACpD,MAAM,WAAW,GAAG,WAAW,CAAC,KAAK,EAAE,UAAU,CAAC,CAAC;QACnD,IAAI,KAAK,CAAC,IAAI,KAAK,gBAAgB,EAAE,CAAC;YACrC,MAAM,UAAU,GAAG,eAAe,CAAC,KAAK,EAAE,WAAW,CAAC,CAAC;YACvD,IAAI,CAAC,UAAU;gBAAE,SAAS;YAC1B,MAAM,KAAK,GAAG,eAAe,CAAC,KAAK,EAAE,UAAU,EAAE,KAAK,CAAC,CAAC;YACxD,IAAI,KAAK,CAAC,iBAAiB,KAAK,SAAS,EAAE,CAAC;gBAC3C,KAAK,CAAC,MAAM,EAAE;oBACb,QAAQ,EAAE,OAAO;oBACjB,MAAM,EAAE,wBAAwB;oBAChC,QAAQ,EAAE,KAAK,CAAC,QAAQ;oBACxB,OAAO,EAAE,KAAK,CAAC,EAAE;oBACjB,KAAK,EAAE,KAAK,CAAC,KAAK;oBAClB,IAAI,EAAE,KAAK,CAAC,IAAI;oBAChB,QAAQ,EAAE,kCAAkC;oBAC5C,QAAQ,EAAE,gCAAgC,KAAK,CAAC,iBAAiB,EAAE;oBACnE,OAAO,EAAE,aAAa,UAAU,gCAAgC;oBAChE,QAAQ,EAAE,KAAK,CAAC,QAAQ;iBACxB,CAAC,CAAC;YACJ,CAAC;YACD,KAAK,CAAC,iBAAiB,GAAG,KAAK,CAAC,QAAQ,CAAC;QAC1C,CAAC;QAED,IAAI,KAAK,CAAC,IAAI,KAAK,gBAAgB,IAAI,KAAK,CAAC,IAAI,KAAK,aAAa,EAAE,CAAC;YACrE,MAAM,UAAU,GAAG,eAAe,CAAC,KAAK,EAAE,WAAW,CAAC,CAAC;YACvD,IAAI,CAAC,UAAU;gBAAE,SAAS;YAC1B,MAAM,KAAK,GAAG,eAAe,CAAC,KAAK,EAAE,UAAU,EAAE,KAAK,CAAC,CAAC;YACxD,MAAM,cAAc,GACnB,KAAK,CAAC,IAAI,KAAK,gBAAgB,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,QAAQ,CAAC;YAC1D,IACC,KAAK,CAAC,cAAc,KAAK,SAAS;gBAClC,KAAK,CAAC,cAAc,KAAK,cAAc,EACtC,CAAC;gBACF,KAAK,CAAC,MAAM,EAAE;oBACb,QAAQ,EAAE,OAAO;oBACjB,MAAM,EAAE,kCAAkC;oBAC1C,QAAQ,EAAE,KAAK,CAAC,QAAQ;oBACxB,OAAO,EAAE,KAAK,CAAC,EAAE;oBACjB,KAAK,EAAE,KAAK,CAAC,KAAK;oBAClB,IAAI,EAAE,KAAK,CAAC,IAAI;oBAChB,QAAQ,EAAE,KAAK,CAAC,cAAc;oBAC9B,QAAQ,EAAE,cAAc;oBACxB,OAAO,EAAE,aAAa,UAAU,qCAAqC;oBACrE,QAAQ,EAAE,KAAK,CAAC,QAAQ;iBACxB,CAAC,CAAC;YACJ,CAAC;YACD,KAAK,CAAC,cAAc,GAAG,cAAc,CAAC;YACtC,KAAK,CAAC,eAAe,CAAC,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;QAC5C,CAAC;QAED,KAAK,MAAM,UAAU,IAAI,WAAW,EAAE,CAAC;YACtC,MAAM,KAAK,GAAG,eAAe,CAAC,KAAK,EAAE,UAAU,EAAE,KAAK,CAAC,CAAC;YACxD,KAAK,MAAM,UAAU,IAAI,WAAW,EAAE,CAAC;gBACtC,KAAK,CAAC,WAAW,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC;YACnC,CAAC;QACF,CAAC;IACF,CAAC;IAED,KAAK,MAAM,CAAC,UAAU,EAAE,WAAW,CAAC,IAAI,MAAM,CAAC,OAAO,CACrD,OAAO,CAAC,aAAa,IAAI,EAAE,CAC3B,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC;QACxD,MAAM,KAAK,GAAG,KAAK,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC;QACpC,IAAI,CAAC,KAAK,EAAE,CAAC;YACZ,KAAK,CAAC,MAAM,EAAE;gBACb,QAAQ,EAAE,OAAO;gBACjB,MAAM,EAAE,uBAAuB;gBAC/B,QAAQ,EAAE,UAAU;gBACpB,QAAQ,EAAE,SAAS;gBACnB,OAAO,EAAE,sBAAsB,UAAU,qCAAqC;gBAC9E,QAAQ,EAAE,EAAE;aACZ,CAAC,CAAC;YACH,SAAS;QACV,CAAC;QACD,IACC,WAAW,CAAC,cAAc;YAC1B,KAAK,CAAC,cAAc,KAAK,WAAW,CAAC,cAAc,EAClD,CAAC;YACF,KAAK,CAAC,MAAM,EAAE;gBACb,QAAQ,EAAE,OAAO;gBACjB,MAAM,EAAE,+BAA+B;gBACvC,QAAQ,EAAE,WAAW,CAAC,cAAc;gBACpC,QAAQ,EAAE,KAAK,CAAC,cAAc,IAAI,SAAS;gBAC3C,OAAO,EAAE,aAAa,UAAU,oDAAoD;gBACpF,QAAQ,EAAE,KAAK,CAAC,QAAQ;aACxB,CAAC,CAAC;QACJ,CAAC;QACD,KAAK,MAAM,UAAU,IAAI,WAAW,CAAC,mBAAmB,IAAI,EAAE,EAAE,CAAC;YAChE,IAAI,KAAK,CAAC,WAAW,CAAC,GAAG,CAAC,UAAU,CAAC;gBAAE,SAAS;YAChD,KAAK,CAAC,MAAM,EAAE;gBACb,QAAQ,EAAE,OAAO;gBACjB,MAAM,EAAE,gCAAgC;gBACxC,QAAQ,EAAE,UAAU;gBACpB,QAAQ,EAAE,SAAS;gBACnB,OAAO,EAAE,aAAa,UAAU,sCAAsC,UAAU,GAAG;gBACnF,QAAQ,EAAE,KAAK,CAAC,QAAQ;aACxB,CAAC,CAAC;QACJ,CAAC;IACF,CAAC;IAED,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,KAAK,OAAO,CAAC,CAAC,MAAM,CAAC;IACzE,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,MAAM,CAAC;IAC7E,MAAM,MAAM,GAAG,eAAe,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;IAC9C,MAAM,SAAS,GAAG,cAAc,CAAC,KAAK,CAAC,CAAC;IACxC,OAAO;QACN,aAAa,EAAE,8BAA8B;QAC7C,uBAAuB,EAAE,MAAM,CAAC,aAAa;QAC7C,GAAG,EAAE,MAAM,CAAC,GAAG;QACf,aAAa,EAAE,IAAI;QACnB,MAAM,EAAE;YACP,MAAM,EAAE,MAAM,CAAC,MAAM,CAAC,MAAM;YAC5B,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,MAAM;YACN,QAAQ;YACR,SAAS,EAAE,SAAS,CAAC,MAAM;YAC3B,MAAM,EAAE,MAAM,CAAC,MAAM;SACrB;QACD,MAAM;QACN,SAAS;QACT,MAAM;KACN,CAAC;AACH,CAAC","sourcesContent":["import { validateAgentTrajectoryReport } from \"./agent-trajectory-validation.js\";\nimport type {\n\tAgentTrajectoryEvent,\n\tAgentTrajectoryEventKind,\n\tAgentTrajectoryPhase,\n\tAgentTrajectoryReport,\n} from \"./agent-trajectory.js\";\n\nexport const AGENT_TRAJECTORY_REPLAY_SCHEMA =\n\t\"evalops.maestro.agent-trajectory-replay.v1\";\n\nexport type AgentTrajectoryReplaySeverity = \"error\" | \"warning\";\n\nexport interface AgentTrajectoryReplayDelta {\n\tid: string;\n\tseverity: AgentTrajectoryReplaySeverity;\n\truleId: string;\n\tsequence?: number;\n\teventId?: string;\n\tphase?: AgentTrajectoryPhase;\n\tkind?: AgentTrajectoryEventKind;\n\texpected?: string;\n\tobserved?: string;\n\tmessage: string;\n\tevidence: AgentTrajectoryEvent[\"evidence\"];\n}\n\nexport interface AgentTrajectoryReplayPhaseSummary {\n\tphase: AgentTrajectoryPhase;\n\tevents: number;\n\tfirstSequence: number;\n\tlastSequence: number;\n}\n\nexport interface AgentTrajectoryReplayToolSummary {\n\ttoolCallId: string;\n\ttoolName?: string;\n\trequestedSequence?: number;\n\tresultSequences: number[];\n\tterminalStatus?: \"completed\" | \"failed\";\n\tevidence: AgentTrajectoryEvent[\"evidence\"];\n}\n\nexport interface AgentTrajectoryReplayReport {\n\tschemaVersion: typeof AGENT_TRAJECTORY_REPLAY_SCHEMA;\n\ttrajectorySchemaVersion: AgentTrajectoryReport[\"schemaVersion\"];\n\trun: AgentTrajectoryReport[\"run\"];\n\tdeterministic: true;\n\tcounts: {\n\t\tevents: number;\n\t\tdeltas: number;\n\t\terrors: number;\n\t\twarnings: number;\n\t\ttoolCalls: number;\n\t\tphases: number;\n\t};\n\tphases: AgentTrajectoryReplayPhaseSummary[];\n\ttoolCalls: AgentTrajectoryReplayToolSummary[];\n\tdeltas: AgentTrajectoryReplayDelta[];\n}\n\nexport interface AgentTrajectoryReplayToolExpectation {\n\tterminalStatus?: \"completed\" | \"failed\";\n\trequiredArtifactIds?: string[];\n}\n\nexport interface ReplayAgentTrajectoryOptions {\n\texpectedTools?: Record<string, AgentTrajectoryReplayToolExpectation>;\n}\n\ninterface ToolState {\n\ttoolCallId: string;\n\ttoolName?: string;\n\trequestedSequence?: number;\n\tresultSequences: number[];\n\tterminalStatus?: \"completed\" | \"failed\";\n\tevidence: AgentTrajectoryEvent[\"evidence\"];\n\tartifactIds: Set<string>;\n}\n\nfunction evidenceIds(\n\tevent: AgentTrajectoryEvent,\n\tkind: AgentTrajectoryEvent[\"evidence\"][number][\"kind\"],\n): string[] {\n\treturn event.evidence\n\t\t.filter((anchor) => anchor.kind === kind)\n\t\t.map((anchor) => anchor.id)\n\t\t.sort();\n}\n\nfunction firstEvidenceId(\n\tevent: AgentTrajectoryEvent,\n\tkind: AgentTrajectoryEvent[\"evidence\"][number][\"kind\"],\n): string | undefined {\n\treturn evidenceIds(event, kind)[0];\n}\n\nfunction mergeEvidence(\n\ttarget: AgentTrajectoryEvent[\"evidence\"],\n\tsource: AgentTrajectoryEvent[\"evidence\"],\n): AgentTrajectoryEvent[\"evidence\"] {\n\tconst seen = new Set(target.map((anchor) => `${anchor.kind}:${anchor.id}`));\n\tfor (const anchor of source) {\n\t\tconst key = `${anchor.kind}:${anchor.id}`;\n\t\tif (seen.has(key)) continue;\n\t\tseen.add(key);\n\t\ttarget.push(anchor);\n\t}\n\treturn target.sort((a, b) => {\n\t\tconst kindDelta = a.kind.localeCompare(b.kind);\n\t\treturn kindDelta === 0 ? a.id.localeCompare(b.id) : kindDelta;\n\t});\n}\n\nfunction delta(\n\tdeltas: AgentTrajectoryReplayDelta[],\n\tparams: Omit<AgentTrajectoryReplayDelta, \"id\">,\n): void {\n\tdeltas.push({\n\t\tid: `delta:${String(deltas.length + 1).padStart(3, \"0\")}`,\n\t\t...params,\n\t});\n}\n\nfunction ensureToolState(\n\ttools: Map<string, ToolState>,\n\ttoolCallId: string,\n\tevent: AgentTrajectoryEvent,\n): ToolState {\n\tlet state = tools.get(toolCallId);\n\tif (!state) {\n\t\tstate = {\n\t\t\ttoolCallId,\n\t\t\ttoolName: event.toolName,\n\t\t\tresultSequences: [],\n\t\t\tevidence: [],\n\t\t\tartifactIds: new Set<string>(),\n\t\t};\n\t\ttools.set(toolCallId, state);\n\t}\n\tif (!state.toolName && event.toolName) {\n\t\tstate.toolName = event.toolName;\n\t}\n\tmergeEvidence(state.evidence, event.evidence);\n\treturn state;\n}\n\nfunction summarizePhases(\n\tevents: AgentTrajectoryEvent[],\n): AgentTrajectoryReplayPhaseSummary[] {\n\tconst byPhase = new Map<\n\t\tAgentTrajectoryPhase,\n\t\tAgentTrajectoryReplayPhaseSummary\n\t>();\n\tfor (const event of events) {\n\t\tconst existing = byPhase.get(event.phase);\n\t\tif (existing) {\n\t\t\texisting.events += 1;\n\t\t\texisting.lastSequence = event.sequence;\n\t\t\tcontinue;\n\t\t}\n\t\tbyPhase.set(event.phase, {\n\t\t\tphase: event.phase,\n\t\t\tevents: 1,\n\t\t\tfirstSequence: event.sequence,\n\t\t\tlastSequence: event.sequence,\n\t\t});\n\t}\n\treturn [...byPhase.values()].sort(\n\t\t(a, b) => a.firstSequence - b.firstSequence,\n\t);\n}\n\nfunction summarizeTools(\n\ttools: Map<string, ToolState>,\n): AgentTrajectoryReplayToolSummary[] {\n\treturn [...tools.values()]\n\t\t.map((tool) => ({\n\t\t\ttoolCallId: tool.toolCallId,\n\t\t\t...(tool.toolName ? { toolName: tool.toolName } : {}),\n\t\t\t...(tool.requestedSequence !== undefined\n\t\t\t\t? { requestedSequence: tool.requestedSequence }\n\t\t\t\t: {}),\n\t\t\tresultSequences: [...tool.resultSequences].sort((a, b) => a - b),\n\t\t\t...(tool.terminalStatus ? { terminalStatus: tool.terminalStatus } : {}),\n\t\t\tevidence: [...tool.evidence],\n\t\t}))\n\t\t.sort((a, b) => {\n\t\t\tconst sequenceA = a.requestedSequence ?? Number.MAX_SAFE_INTEGER;\n\t\t\tconst sequenceB = b.requestedSequence ?? Number.MAX_SAFE_INTEGER;\n\t\t\tif (sequenceA !== sequenceB) return sequenceA - sequenceB;\n\t\t\treturn a.toolCallId.localeCompare(b.toolCallId);\n\t\t});\n}\n\nexport function replayAgentTrajectoryReport(\n\treport: AgentTrajectoryReport,\n\toptions: ReplayAgentTrajectoryOptions = {},\n): AgentTrajectoryReplayReport {\n\tconst deltas: AgentTrajectoryReplayDelta[] = [];\n\tconst validation = validateAgentTrajectoryReport(report);\n\tfor (const failure of validation.failures) {\n\t\tdelta(deltas, {\n\t\t\tseverity: \"error\",\n\t\t\truleId: \"trajectory.validation\",\n\t\t\tmessage: failure,\n\t\t\tevidence: [],\n\t\t});\n\t}\n\n\tconst tools = new Map<string, ToolState>();\n\tfor (const event of report.events) {\n\t\tconst toolCallIds = evidenceIds(event, \"tool_call\");\n\t\tconst artifactIds = evidenceIds(event, \"artifact\");\n\t\tif (event.type === \"tool.requested\") {\n\t\t\tconst toolCallId = firstEvidenceId(event, \"tool_call\");\n\t\t\tif (!toolCallId) continue;\n\t\t\tconst state = ensureToolState(tools, toolCallId, event);\n\t\t\tif (state.requestedSequence !== undefined) {\n\t\t\t\tdelta(deltas, {\n\t\t\t\t\tseverity: \"error\",\n\t\t\t\t\truleId: \"tool.duplicate_request\",\n\t\t\t\t\tsequence: event.sequence,\n\t\t\t\t\teventId: event.id,\n\t\t\t\t\tphase: event.phase,\n\t\t\t\t\tkind: event.kind,\n\t\t\t\t\texpected: \"one tool.requested per tool call\",\n\t\t\t\t\tobserved: `previous request at sequence ${state.requestedSequence}`,\n\t\t\t\t\tmessage: `Tool call ${toolCallId} was requested more than once.`,\n\t\t\t\t\tevidence: event.evidence,\n\t\t\t\t});\n\t\t\t}\n\t\t\tstate.requestedSequence = event.sequence;\n\t\t}\n\n\t\tif (event.type === \"tool.completed\" || event.type === \"tool.failed\") {\n\t\t\tconst toolCallId = firstEvidenceId(event, \"tool_call\");\n\t\t\tif (!toolCallId) continue;\n\t\t\tconst state = ensureToolState(tools, toolCallId, event);\n\t\t\tconst terminalStatus =\n\t\t\t\tevent.type === \"tool.completed\" ? \"completed\" : \"failed\";\n\t\t\tif (\n\t\t\t\tstate.terminalStatus !== undefined &&\n\t\t\t\tstate.terminalStatus !== terminalStatus\n\t\t\t) {\n\t\t\t\tdelta(deltas, {\n\t\t\t\t\tseverity: \"error\",\n\t\t\t\t\truleId: \"tool.conflicting_terminal_status\",\n\t\t\t\t\tsequence: event.sequence,\n\t\t\t\t\teventId: event.id,\n\t\t\t\t\tphase: event.phase,\n\t\t\t\t\tkind: event.kind,\n\t\t\t\t\texpected: state.terminalStatus,\n\t\t\t\t\tobserved: terminalStatus,\n\t\t\t\t\tmessage: `Tool call ${toolCallId} has conflicting terminal statuses.`,\n\t\t\t\t\tevidence: event.evidence,\n\t\t\t\t});\n\t\t\t}\n\t\t\tstate.terminalStatus = terminalStatus;\n\t\t\tstate.resultSequences.push(event.sequence);\n\t\t}\n\n\t\tfor (const toolCallId of toolCallIds) {\n\t\t\tconst state = ensureToolState(tools, toolCallId, event);\n\t\t\tfor (const artifactId of artifactIds) {\n\t\t\t\tstate.artifactIds.add(artifactId);\n\t\t\t}\n\t\t}\n\t}\n\n\tfor (const [toolCallId, expectation] of Object.entries(\n\t\toptions.expectedTools ?? {},\n\t).sort(([left], [right]) => left.localeCompare(right))) {\n\t\tconst state = tools.get(toolCallId);\n\t\tif (!state) {\n\t\t\tdelta(deltas, {\n\t\t\t\tseverity: \"error\",\n\t\t\t\truleId: \"tool.expected_missing\",\n\t\t\t\texpected: toolCallId,\n\t\t\t\tobserved: \"missing\",\n\t\t\t\tmessage: `Expected tool call ${toolCallId} was not present in the trajectory.`,\n\t\t\t\tevidence: [],\n\t\t\t});\n\t\t\tcontinue;\n\t\t}\n\t\tif (\n\t\t\texpectation.terminalStatus &&\n\t\t\tstate.terminalStatus !== expectation.terminalStatus\n\t\t) {\n\t\t\tdelta(deltas, {\n\t\t\t\tseverity: \"error\",\n\t\t\t\truleId: \"tool.terminal_status_mismatch\",\n\t\t\t\texpected: expectation.terminalStatus,\n\t\t\t\tobserved: state.terminalStatus ?? \"missing\",\n\t\t\t\tmessage: `Tool call ${toolCallId} terminal status did not match replay expectation.`,\n\t\t\t\tevidence: state.evidence,\n\t\t\t});\n\t\t}\n\t\tfor (const artifactId of expectation.requiredArtifactIds ?? []) {\n\t\t\tif (state.artifactIds.has(artifactId)) continue;\n\t\t\tdelta(deltas, {\n\t\t\t\tseverity: \"error\",\n\t\t\t\truleId: \"tool.required_artifact_missing\",\n\t\t\t\texpected: artifactId,\n\t\t\t\tobserved: \"missing\",\n\t\t\t\tmessage: `Tool call ${toolCallId} did not produce required artifact ${artifactId}.`,\n\t\t\t\tevidence: state.evidence,\n\t\t\t});\n\t\t}\n\t}\n\n\tconst errors = deltas.filter((item) => item.severity === \"error\").length;\n\tconst warnings = deltas.filter((item) => item.severity === \"warning\").length;\n\tconst phases = summarizePhases(report.events);\n\tconst toolCalls = summarizeTools(tools);\n\treturn {\n\t\tschemaVersion: AGENT_TRAJECTORY_REPLAY_SCHEMA,\n\t\ttrajectorySchemaVersion: report.schemaVersion,\n\t\trun: report.run,\n\t\tdeterministic: true,\n\t\tcounts: {\n\t\t\tevents: report.events.length,\n\t\t\tdeltas: deltas.length,\n\t\t\terrors,\n\t\t\twarnings,\n\t\t\ttoolCalls: toolCalls.length,\n\t\t\tphases: phases.length,\n\t\t},\n\t\tphases,\n\t\ttoolCalls,\n\t\tdeltas,\n\t};\n}\n"]}
|
|
@@ -0,0 +1,109 @@
|
|
|
1
|
+
import { MAESTRO_SCENARIO_SCHEMA, type MaestroScenario, type MaestroScenarioAssertion, type MaestroScenarioExternalRefs, type MaestroScenarioOutcome, type MaestroScenarioPlatformLink, type MaestroScenarioReleaseGate, type MaestroScenarioRequiredArtifact, type MaestroScenarioReviewLabel, type MaestroScenarioSeverity, type MaestroScenarioWorkspaceHydrationMode, type MaestroScenarioWorkspaceManifest } from "@evalops/contracts";
|
|
2
|
+
import type { AgentTrajectoryInspectionReport } from "./agent-trajectory-inspection.js";
|
|
3
|
+
import type { AgentTrajectoryReplayReport } from "./agent-trajectory-replay.js";
|
|
4
|
+
import type { AgentTrajectoryScoreReport } from "./agent-trajectory-scorers.js";
|
|
5
|
+
import type { AgentTrajectoryEvent, AgentTrajectoryReport } from "./agent-trajectory.js";
|
|
6
|
+
export declare const AGENT_TRAJECTORY_SCENARIO_RESULT_SCHEMA = "evalops.maestro.agent-trajectory-scenario-result.v1";
|
|
7
|
+
export { MAESTRO_SCENARIO_WORKSPACE_MANIFEST_SCHEMA, MAESTRO_SCENARIO_SCHEMA, type MaestroScenario, type MaestroScenarioAssertion, type MaestroScenarioExternalRefs, type MaestroScenarioGateTier, type MaestroScenarioOutcome, type MaestroScenarioPlatformLink, type MaestroScenarioReleaseGate, type MaestroScenarioReviewLabel, type MaestroScenarioRequiredArtifact, type MaestroScenarioSeverity, type MaestroScenarioWorkspaceManifest, type MaestroScenarioWorkspaceHydrationMode, } from "@evalops/contracts";
|
|
8
|
+
export type AgentTrajectoryScenarioStatus = "pass" | "fail" | "warn";
|
|
9
|
+
export interface AgentTrajectoryScenarioEvidence {
|
|
10
|
+
kind: string;
|
|
11
|
+
id: string;
|
|
12
|
+
source: "trajectory" | "replay" | "score" | "inspection" | "scenario";
|
|
13
|
+
label: string;
|
|
14
|
+
}
|
|
15
|
+
export interface AgentTrajectoryScenarioAssertionResult {
|
|
16
|
+
id: string;
|
|
17
|
+
kind: MaestroScenarioAssertion["kind"];
|
|
18
|
+
status: AgentTrajectoryScenarioStatus;
|
|
19
|
+
severity: MaestroScenarioSeverity;
|
|
20
|
+
message: string;
|
|
21
|
+
evidence: AgentTrajectoryScenarioEvidence[];
|
|
22
|
+
}
|
|
23
|
+
export interface AgentTrajectoryScenarioProvenanceStep {
|
|
24
|
+
eventId: string;
|
|
25
|
+
eventType: string;
|
|
26
|
+
phase: AgentTrajectoryEvent["phase"];
|
|
27
|
+
actor: AgentTrajectoryEvent["actor"];
|
|
28
|
+
evidence: AgentTrajectoryScenarioEvidence[];
|
|
29
|
+
}
|
|
30
|
+
export interface AgentTrajectoryScenarioDiff {
|
|
31
|
+
baselineRunId: string;
|
|
32
|
+
candidateRunId: string;
|
|
33
|
+
eventsDelta: number;
|
|
34
|
+
toolCallsDelta: number;
|
|
35
|
+
scoreFailuresDelta: number;
|
|
36
|
+
}
|
|
37
|
+
export interface AgentTrajectoryScenarioWorkspaceSummary {
|
|
38
|
+
manifestId: string;
|
|
39
|
+
source: MaestroScenarioWorkspaceManifest["source"];
|
|
40
|
+
recordedAt: string;
|
|
41
|
+
hydrationMode: MaestroScenarioWorkspaceHydrationMode;
|
|
42
|
+
files: number;
|
|
43
|
+
toolAdapters: number;
|
|
44
|
+
}
|
|
45
|
+
export interface AgentTrajectoryScenarioReleaseGateSummary extends MaestroScenarioReleaseGate {
|
|
46
|
+
satisfied: boolean;
|
|
47
|
+
missingArtifacts: MaestroScenarioRequiredArtifact[];
|
|
48
|
+
budgetViolations: string[];
|
|
49
|
+
policyViolations: string[];
|
|
50
|
+
}
|
|
51
|
+
export interface AgentTrajectoryScenarioResult {
|
|
52
|
+
schemaVersion: typeof AGENT_TRAJECTORY_SCENARIO_RESULT_SCHEMA;
|
|
53
|
+
scenarioSchemaVersion: typeof MAESTRO_SCENARIO_SCHEMA;
|
|
54
|
+
scenario: {
|
|
55
|
+
id: string;
|
|
56
|
+
title: string;
|
|
57
|
+
expectedOutcome: MaestroScenarioOutcome;
|
|
58
|
+
observedOutcome: MaestroScenarioOutcome;
|
|
59
|
+
reviewLabels: MaestroScenarioReviewLabel[];
|
|
60
|
+
};
|
|
61
|
+
externalRefs?: MaestroScenarioExternalRefs;
|
|
62
|
+
run: AgentTrajectoryReport["run"] & {
|
|
63
|
+
scenarioId: string;
|
|
64
|
+
replay: true;
|
|
65
|
+
};
|
|
66
|
+
counts: {
|
|
67
|
+
assertions: number;
|
|
68
|
+
passed: number;
|
|
69
|
+
failed: number;
|
|
70
|
+
warnings: number;
|
|
71
|
+
events: number;
|
|
72
|
+
toolCalls: number;
|
|
73
|
+
replayDeltas: number;
|
|
74
|
+
scoreFailures: number;
|
|
75
|
+
scoreWarnings: number;
|
|
76
|
+
workspaceFiles: number;
|
|
77
|
+
toolAdapters: number;
|
|
78
|
+
};
|
|
79
|
+
platform: MaestroScenarioPlatformLink & {
|
|
80
|
+
evidenceEventType: "maestro.events.eval.scored";
|
|
81
|
+
};
|
|
82
|
+
releaseGate?: AgentTrajectoryScenarioReleaseGateSummary;
|
|
83
|
+
workspace?: AgentTrajectoryScenarioWorkspaceSummary;
|
|
84
|
+
assumptions: MaestroScenario["assumptions"];
|
|
85
|
+
assertions: AgentTrajectoryScenarioAssertionResult[];
|
|
86
|
+
provenance: AgentTrajectoryScenarioProvenanceStep[];
|
|
87
|
+
diff?: AgentTrajectoryScenarioDiff;
|
|
88
|
+
}
|
|
89
|
+
interface ScenarioInputs {
|
|
90
|
+
trajectory: AgentTrajectoryReport;
|
|
91
|
+
replay: AgentTrajectoryReplayReport;
|
|
92
|
+
score: AgentTrajectoryScoreReport;
|
|
93
|
+
inspection?: AgentTrajectoryInspectionReport;
|
|
94
|
+
workspaceManifest?: MaestroScenarioWorkspaceManifest;
|
|
95
|
+
baselineTrajectory?: AgentTrajectoryReport;
|
|
96
|
+
candidateTrajectory?: AgentTrajectoryReport;
|
|
97
|
+
baselineScore?: AgentTrajectoryScoreReport;
|
|
98
|
+
candidateScore?: AgentTrajectoryScoreReport;
|
|
99
|
+
}
|
|
100
|
+
export declare function parseAgentTrajectoryScenario(value: unknown, label: string): MaestroScenario;
|
|
101
|
+
export declare function loadAgentTrajectoryScenario(path: string): MaestroScenario;
|
|
102
|
+
export declare function loadAgentTrajectoryScenarioInputs(scenario: MaestroScenario, baseDir: string): ScenarioInputs;
|
|
103
|
+
export declare function validateAgentTrajectoryScenario(scenario: MaestroScenario, label?: string): void;
|
|
104
|
+
export declare function evaluateAgentTrajectoryScenario(scenario: MaestroScenario, inputs: ScenarioInputs): AgentTrajectoryScenarioResult;
|
|
105
|
+
export declare function scenarioResultToJunit(result: AgentTrajectoryScenarioResult): string;
|
|
106
|
+
export declare function runAgentTrajectoryScenarioFile(path: string, options?: {
|
|
107
|
+
baseDir?: string;
|
|
108
|
+
}): AgentTrajectoryScenarioResult;
|
|
109
|
+
//# sourceMappingURL=agent-trajectory-scenarios.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agent-trajectory-scenarios.d.ts","sourceRoot":"","sources":["../../src/server/agent-trajectory-scenarios.ts"],"names":[],"mappings":"AAEA,OAAO,EACN,uBAAuB,EAEvB,KAAK,eAAe,EACpB,KAAK,wBAAwB,EAC7B,KAAK,2BAA2B,EAEhC,KAAK,sBAAsB,EAC3B,KAAK,2BAA2B,EAChC,KAAK,0BAA0B,EAC/B,KAAK,+BAA+B,EACpC,KAAK,0BAA0B,EAC/B,KAAK,uBAAuB,EAE5B,KAAK,qCAAqC,EAC1C,KAAK,gCAAgC,EAErC,MAAM,oBAAoB,CAAC;AAC5B,OAAO,KAAK,EAAE,+BAA+B,EAAE,MAAM,kCAAkC,CAAC;AACxF,OAAO,KAAK,EAAE,2BAA2B,EAAE,MAAM,8BAA8B,CAAC;AAChF,OAAO,KAAK,EAAE,0BAA0B,EAAE,MAAM,+BAA+B,CAAC;AAChF,OAAO,KAAK,EACX,oBAAoB,EACpB,qBAAqB,EACrB,MAAM,uBAAuB,CAAC;AAG/B,eAAO,MAAM,uCAAuC,wDACE,CAAC;AACvD,OAAO,EACN,0CAA0C,EAC1C,uBAAuB,EACvB,KAAK,eAAe,EACpB,KAAK,wBAAwB,EAC7B,KAAK,2BAA2B,EAChC,KAAK,uBAAuB,EAC5B,KAAK,sBAAsB,EAC3B,KAAK,2BAA2B,EAChC,KAAK,0BAA0B,EAC/B,KAAK,0BAA0B,EAC/B,KAAK,+BAA+B,EACpC,KAAK,uBAAuB,EAC5B,KAAK,gCAAgC,EACrC,KAAK,qCAAqC,GAC1C,MAAM,oBAAoB,CAAC;AAE5B,MAAM,MAAM,6BAA6B,GAAG,MAAM,GAAG,MAAM,GAAG,MAAM,CAAC;AAErE,MAAM,WAAW,+BAA+B;IAC/C,IAAI,EAAE,MAAM,CAAC;IACb,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,YAAY,GAAG,QAAQ,GAAG,OAAO,GAAG,YAAY,GAAG,UAAU,CAAC;IACtE,KAAK,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,sCAAsC;IACtD,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,wBAAwB,CAAC,MAAM,CAAC,CAAC;IACvC,MAAM,EAAE,6BAA6B,CAAC;IACtC,QAAQ,EAAE,uBAAuB,CAAC;IAClC,OAAO,EAAE,MAAM,CAAC;IAChB,QAAQ,EAAE,+BAA+B,EAAE,CAAC;CAC5C;AAED,MAAM,WAAW,qCAAqC;IACrD,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,oBAAoB,CAAC,OAAO,CAAC,CAAC;IACrC,KAAK,EAAE,oBAAoB,CAAC,OAAO,CAAC,CAAC;IACrC,QAAQ,EAAE,+BAA+B,EAAE,CAAC;CAC5C;AAED,MAAM,WAAW,2BAA2B;IAC3C,aAAa,EAAE,MAAM,CAAC;IACtB,cAAc,EAAE,MAAM,CAAC;IACvB,WAAW,EAAE,MAAM,CAAC;IACpB,cAAc,EAAE,MAAM,CAAC;IACvB,kBAAkB,EAAE,MAAM,CAAC;CAC3B;AAED,MAAM,WAAW,uCAAuC;IACvD,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,gCAAgC,CAAC,QAAQ,CAAC,CAAC;IACnD,UAAU,EAAE,MAAM,CAAC;IACnB,aAAa,EAAE,qCAAqC,CAAC;IACrD,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,yCAChB,SAAQ,0BAA0B;IAClC,SAAS,EAAE,OAAO,CAAC;IACnB,gBAAgB,EAAE,+BAA+B,EAAE,CAAC;IACpD,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAC3B,gBAAgB,EAAE,MAAM,EAAE,CAAC;CAC3B;AAED,MAAM,WAAW,6BAA6B;IAC7C,aAAa,EAAE,OAAO,uCAAuC,CAAC;IAC9D,qBAAqB,EAAE,OAAO,uBAAuB,CAAC;IACtD,QAAQ,EAAE;QACT,EAAE,EAAE,MAAM,CAAC;QACX,KAAK,EAAE,MAAM,CAAC;QACd,eAAe,EAAE,sBAAsB,CAAC;QACxC,eAAe,EAAE,sBAAsB,CAAC;QACxC,YAAY,EAAE,0BAA0B,EAAE,CAAC;KAC3C,CAAC;IACF,YAAY,CAAC,EAAE,2BAA2B,CAAC;IAC3C,GAAG,EAAE,qBAAqB,CAAC,KAAK,CAAC,GAAG;QACnC,UAAU,EAAE,MAAM,CAAC;QACnB,MAAM,EAAE,IAAI,CAAC;KACb,CAAC;IACF,MAAM,EAAE;QACP,UAAU,EAAE,MAAM,CAAC;QACnB,MAAM,EAAE,MAAM,CAAC;QACf,MAAM,EAAE,MAAM,CAAC;QACf,QAAQ,EAAE,MAAM,CAAC;QACjB,MAAM,EAAE,MAAM,CAAC;QACf,SAAS,EAAE,MAAM,CAAC;QAClB,YAAY,EAAE,MAAM,CAAC;QACrB,aAAa,EAAE,MAAM,CAAC;QACtB,aAAa,EAAE,MAAM,CAAC;QACtB,cAAc,EAAE,MAAM,CAAC;QACvB,YAAY,EAAE,MAAM,CAAC;KACrB,CAAC;IACF,QAAQ,EAAE,2BAA2B,GAAG;QACvC,iBAAiB,EAAE,4BAA4B,CAAC;KAChD,CAAC;IACF,WAAW,CAAC,EAAE,yCAAyC,CAAC;IACxD,SAAS,CAAC,EAAE,uCAAuC,CAAC;IACpD,WAAW,EAAE,eAAe,CAAC,aAAa,CAAC,CAAC;IAC5C,UAAU,EAAE,sCAAsC,EAAE,CAAC;IACrD,UAAU,EAAE,qCAAqC,EAAE,CAAC;IACpD,IAAI,CAAC,EAAE,2BAA2B,CAAC;CACnC;AAED,UAAU,cAAc;IACvB,UAAU,EAAE,qBAAqB,CAAC;IAClC,MAAM,EAAE,2BAA2B,CAAC;IACpC,KAAK,EAAE,0BAA0B,CAAC;IAClC,UAAU,CAAC,EAAE,+BAA+B,CAAC;IAC7C,iBAAiB,CAAC,EAAE,gCAAgC,CAAC;IACrD,kBAAkB,CAAC,EAAE,qBAAqB,CAAC;IAC3C,mBAAmB,CAAC,EAAE,qBAAqB,CAAC;IAC5C,aAAa,CAAC,EAAE,0BAA0B,CAAC;IAC3C,cAAc,CAAC,EAAE,0BAA0B,CAAC;CAC5C;AAoMD,wBAAgB,4BAA4B,CAC3C,KAAK,EAAE,OAAO,EACd,KAAK,EAAE,MAAM,GACX,eAAe,CAajB;AAED,wBAAgB,2BAA2B,CAAC,IAAI,EAAE,MAAM,GAAG,eAAe,CAEzE;AAED,wBAAgB,iCAAiC,CAChD,QAAQ,EAAE,eAAe,EACzB,OAAO,EAAE,MAAM,GACb,cAAc,CAqEhB;AAED,wBAAgB,+BAA+B,CAC9C,QAAQ,EAAE,eAAe,EACzB,KAAK,SAAc,GACjB,IAAI,CAmSN;AAqjBD,wBAAgB,+BAA+B,CAC9C,QAAQ,EAAE,eAAe,EACzB,MAAM,EAAE,cAAc,GACpB,6BAA6B,CAsD/B;AAED,wBAAgB,qBAAqB,CACpC,MAAM,EAAE,6BAA6B,GACnC,MAAM,CAqCR;AAED,wBAAgB,8BAA8B,CAC7C,IAAI,EAAE,MAAM,EACZ,OAAO,GAAE;IAAE,OAAO,CAAC,EAAE,MAAM,CAAA;CAAO,GAChC,6BAA6B,CAO/B"}
|