stageflow 0.40.0 → 0.44.0
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 +15 -4
- package/dist/agent/activity.d.ts +18 -0
- package/dist/agent/claudeAdapter.js +7 -0
- package/dist/agent/claudeTools.d.ts +27 -18
- package/dist/agent/claudeTools.js +54 -15
- package/dist/agent/compactionEvents.d.ts +34 -0
- package/dist/agent/compactionEvents.js +37 -0
- package/dist/agent/openRouterRouting.d.ts +7 -0
- package/dist/agent/openRouterRouting.js +23 -0
- package/dist/agent/piAdapter.d.ts +6 -0
- package/dist/agent/piAdapter.js +128 -27
- package/dist/agent/piSettings.d.ts +7 -0
- package/dist/agent/piSettings.js +9 -0
- package/dist/agent/sandboxPiTools.d.ts +3 -1
- package/dist/agent/sandboxPiTools.js +8 -2
- package/dist/agent/stageTurnDriver.d.ts +11 -2
- package/dist/agent/stageTurnDriver.js +45 -4
- package/dist/agent/toolOutputBudget.d.ts +15 -1
- package/dist/agent/toolOutputBudget.js +18 -0
- package/dist/agent/toolOutputBudgetExtension.d.ts +2 -0
- package/dist/agent/toolOutputBudgetExtension.js +40 -3
- package/dist/agents/authorChat.d.ts +5 -0
- package/dist/agents/authorChat.js +120 -12
- package/dist/agents/authorSessionStore.d.ts +38 -0
- package/dist/agents/authorSessionStore.js +167 -0
- package/dist/agents/chatTurn.d.ts +2 -0
- package/dist/agents/chatTurn.js +17 -1
- package/dist/agents/errors.d.ts +1 -1
- package/dist/agents/errors.js +1 -0
- package/dist/agents/evalGate.d.ts +61 -0
- package/dist/agents/evalGate.js +103 -0
- package/dist/agents/evalSession.d.ts +26 -0
- package/dist/agents/evalSession.js +29 -0
- package/dist/agents/hostContext.d.ts +4 -1
- package/dist/agents/hostContext.js +4 -1
- package/dist/agents/runtime/piRuntime.d.ts +6 -2
- package/dist/agents/runtime/piRuntime.js +50 -16
- package/dist/agents/runtime/types.d.ts +3 -1
- package/dist/agents/service.d.ts +27 -2
- package/dist/agents/service.js +91 -15
- package/dist/agents/sessions.d.ts +8 -0
- package/dist/agents/sessions.js +53 -11
- package/dist/agents/store.d.ts +14 -3
- package/dist/agents/store.js +45 -7
- package/dist/agents/toolRisk.d.ts +21 -0
- package/dist/agents/toolRisk.js +50 -0
- package/dist/cli/doctorCommand.js +26 -0
- package/dist/cli/evalCommand.d.ts +20 -0
- package/dist/cli/evalCommand.js +720 -0
- package/dist/cli/exportAllCommand.js +1 -1
- package/dist/cli/hostClient.d.ts +60 -0
- package/dist/cli/hostClient.js +71 -0
- package/dist/cli/validateOutput.d.ts +2 -2
- package/dist/cli/validateOutput.js +2 -2
- package/dist/cli.js +10 -0
- package/dist/compaction/directives.d.ts +18 -0
- package/dist/compaction/directives.js +119 -0
- package/dist/compaction/extension.d.ts +20 -0
- package/dist/compaction/extension.js +78 -0
- package/dist/compaction/mask.d.ts +18 -0
- package/dist/compaction/mask.js +155 -0
- package/dist/compaction/pin.d.ts +21 -0
- package/dist/compaction/pin.js +270 -0
- package/dist/compaction/scrub.d.ts +7 -0
- package/dist/compaction/scrub.js +26 -0
- package/dist/compaction/sessionWiring.d.ts +22 -0
- package/dist/compaction/sessionWiring.js +26 -0
- package/dist/compaction/summarize.d.ts +79 -0
- package/dist/compaction/summarize.js +171 -0
- package/dist/compaction/toolPolicy.d.ts +3 -0
- package/dist/compaction/toolPolicy.js +33 -0
- package/dist/compaction/types.d.ts +45 -0
- package/dist/compaction/types.js +7 -0
- package/dist/compaction/workingState.d.ts +56 -0
- package/dist/compaction/workingState.js +198 -0
- package/dist/config/atomicWrite.d.ts +15 -0
- package/dist/config/atomicWrite.js +43 -0
- package/dist/config/browseCatalog.d.ts +2 -0
- package/dist/config/browseCatalog.js +4 -2
- package/dist/config/createPipeline.js +2 -0
- package/dist/config/createTrigger.js +7 -1
- package/dist/config/csv.d.ts +9 -0
- package/dist/config/csv.js +81 -0
- package/dist/config/datasetAgentCheck.d.ts +18 -0
- package/dist/config/datasetAgentCheck.js +109 -0
- package/dist/config/datasetCandidates.d.ts +22 -0
- package/dist/config/datasetCandidates.js +86 -0
- package/dist/config/datasetFromSessions.d.ts +34 -0
- package/dist/config/datasetFromSessions.js +105 -0
- package/dist/config/datasetHash.d.ts +1 -0
- package/dist/config/datasetHash.js +4 -0
- package/dist/config/datasetImport.d.ts +21 -0
- package/dist/config/datasetImport.js +120 -0
- package/dist/config/datasetStageCheck.d.ts +9 -0
- package/dist/config/datasetStageCheck.js +17 -0
- package/dist/config/datasetTypes.d.ts +27 -0
- package/dist/config/datasetTypes.js +1 -0
- package/dist/config/ensureCatalogScanRoot.d.ts +4 -1
- package/dist/config/ensureCatalogScanRoot.js +8 -6
- package/dist/config/evalCatalog.d.ts +22 -0
- package/dist/config/evalCatalog.js +95 -0
- package/dist/config/evalWriteRoots.d.ts +17 -0
- package/dist/config/evalWriteRoots.js +109 -0
- package/dist/config/githubTriggerAction.d.ts +8 -0
- package/dist/config/githubTriggerAction.js +29 -0
- package/dist/config/hostConfig.js +3 -0
- package/dist/config/installEvalSample.d.ts +34 -0
- package/dist/config/installEvalSample.js +182 -0
- package/dist/config/loadDataset.d.ts +16 -0
- package/dist/config/loadDataset.js +318 -0
- package/dist/config/loadEval.d.ts +32 -0
- package/dist/config/loadEval.js +1103 -0
- package/dist/config/loadOutcome.d.ts +5 -0
- package/dist/config/loadStageflowManifest.js +64 -0
- package/dist/config/pipelineNeeds.d.ts +2 -0
- package/dist/config/pipelineNeeds.js +16 -3
- package/dist/config/pipelineRoute.js +5 -4
- package/dist/config/resolveAgentTarget.d.ts +32 -0
- package/dist/config/resolveAgentTarget.js +118 -0
- package/dist/config/resolveEvalTarget.d.ts +21 -0
- package/dist/config/resolveEvalTarget.js +89 -0
- package/dist/config/scanCatalogPaths.d.ts +1 -1
- package/dist/config/scanCatalogPaths.js +17 -5
- package/dist/config/validateCatalog.d.ts +2 -2
- package/dist/config/validateCatalog.js +21 -1
- package/dist/config/validateEval.d.ts +36 -0
- package/dist/config/validateEval.js +446 -0
- package/dist/config/writeDataset.d.ts +54 -0
- package/dist/config/writeDataset.js +186 -0
- package/dist/config/writeEval.d.ts +25 -0
- package/dist/config/writeEval.js +398 -0
- package/dist/diagnostics/checks.js +1 -1
- package/dist/envelope/assertEmittedArtifacts.d.ts +12 -0
- package/dist/envelope/assertEmittedArtifacts.js +43 -0
- package/dist/envelope/preEmitChecks.d.ts +3 -0
- package/dist/evals/agentDiff.d.ts +13 -0
- package/dist/evals/agentDiff.js +11 -0
- package/dist/evals/canonical.d.ts +3 -0
- package/dist/evals/canonical.js +24 -0
- package/dist/evals/caseTask.d.ts +3 -0
- package/dist/evals/caseTask.js +9 -0
- package/dist/evals/compare.d.ts +94 -0
- package/dist/evals/compare.js +201 -0
- package/dist/evals/constants.d.ts +62 -0
- package/dist/evals/constants.js +97 -0
- package/dist/evals/engine/FakeEngine.d.ts +33 -0
- package/dist/evals/engine/FakeEngine.js +177 -0
- package/dist/evals/engine/index.d.ts +9 -0
- package/dist/evals/engine/index.js +19 -0
- package/dist/evals/engine/promptfoo/PromptfooEngine.d.ts +52 -0
- package/dist/evals/engine/promptfoo/PromptfooEngine.js +312 -0
- package/dist/evals/engine/promptfoo/errors.d.ts +5 -0
- package/dist/evals/engine/promptfoo/errors.js +8 -0
- package/dist/evals/engine/promptfoo/generate.d.ts +30 -0
- package/dist/evals/engine/promptfoo/generate.js +345 -0
- package/dist/evals/engine/promptfoo/guard.d.ts +10 -0
- package/dist/evals/engine/promptfoo/guard.js +116 -0
- package/dist/evals/engine/promptfoo/install.d.ts +17 -0
- package/dist/evals/engine/promptfoo/install.js +73 -0
- package/dist/evals/engine/promptfoo/judgeProviders.d.ts +2 -0
- package/dist/evals/engine/promptfoo/judgeProviders.js +9 -0
- package/dist/evals/engine/promptfoo/judgeServices.d.ts +6 -0
- package/dist/evals/engine/promptfoo/judgeServices.js +41 -0
- package/dist/evals/engine/promptfoo/locate.d.ts +21 -0
- package/dist/evals/engine/promptfoo/locate.js +68 -0
- package/dist/evals/engine/promptfoo/nodeVersion.d.ts +7 -0
- package/dist/evals/engine/promptfoo/nodeVersion.js +14 -0
- package/dist/evals/engine/promptfoo/normalize.d.ts +66 -0
- package/dist/evals/engine/promptfoo/normalize.js +261 -0
- package/dist/evals/engine/promptfoo/pinned.d.ts +5 -0
- package/dist/evals/engine/promptfoo/pinned.js +5 -0
- package/dist/evals/engine/promptfoo/rubricPrompt.d.ts +14 -0
- package/dist/evals/engine/promptfoo/rubricPrompt.js +105 -0
- package/dist/evals/engine/promptfoo/runTokens.d.ts +18 -0
- package/dist/evals/engine/promptfoo/runTokens.js +31 -0
- package/dist/evals/engine/promptfoo/status.d.ts +13 -0
- package/dist/evals/engine/promptfoo/status.js +16 -0
- package/dist/evals/estimate.d.ts +56 -0
- package/dist/evals/estimate.js +66 -0
- package/dist/evals/graders/codeGrader.d.ts +6 -0
- package/dist/evals/graders/codeGrader.js +85 -0
- package/dist/evals/graders/conversationCheckGrader.d.ts +6 -0
- package/dist/evals/graders/conversationCheckGrader.js +62 -0
- package/dist/evals/graders/fieldGrader.d.ts +11 -0
- package/dist/evals/graders/fieldGrader.js +117 -0
- package/dist/evals/graders/goal.d.ts +3 -0
- package/dist/evals/graders/goal.js +14 -0
- package/dist/evals/graders/index.d.ts +18 -0
- package/dist/evals/graders/index.js +34 -0
- package/dist/evals/graders/notApplicable.d.ts +5 -0
- package/dist/evals/graders/notApplicable.js +9 -0
- package/dist/evals/graders/path.d.ts +26 -0
- package/dist/evals/graders/path.js +75 -0
- package/dist/evals/graders/referenceGrader.d.ts +5 -0
- package/dist/evals/graders/referenceGrader.js +59 -0
- package/dist/evals/graders/schemaGrader.d.ts +5 -0
- package/dist/evals/graders/schemaGrader.js +29 -0
- package/dist/evals/graders/subject.d.ts +6 -0
- package/dist/evals/graders/subject.js +33 -0
- package/dist/evals/graders/toolCallsGrader.d.ts +6 -0
- package/dist/evals/graders/toolCallsGrader.js +100 -0
- package/dist/evals/graders/transcript.d.ts +12 -0
- package/dist/evals/graders/transcript.js +57 -0
- package/dist/evals/graders/types.d.ts +20 -0
- package/dist/evals/graders/types.js +1 -0
- package/dist/evals/judgeProviders.d.ts +16 -0
- package/dist/evals/judgeProviders.js +27 -0
- package/dist/evals/lineDiff.d.ts +5 -0
- package/dist/evals/lineDiff.js +34 -0
- package/dist/evals/listStatus.d.ts +31 -0
- package/dist/evals/listStatus.js +48 -0
- package/dist/evals/manager.d.ts +187 -0
- package/dist/evals/manager.js +795 -0
- package/dist/evals/plan.d.ts +65 -0
- package/dist/evals/plan.js +621 -0
- package/dist/evals/planInternalAgent.d.ts +19 -0
- package/dist/evals/planInternalAgent.js +205 -0
- package/dist/evals/providerKeys.d.ts +1 -0
- package/dist/evals/providerKeys.js +21 -0
- package/dist/evals/read/agentTargets.d.ts +37 -0
- package/dist/evals/read/agentTargets.js +101 -0
- package/dist/evals/read/api.d.ts +208 -0
- package/dist/evals/read/api.js +363 -0
- package/dist/evals/read/compare.d.ts +4 -0
- package/dist/evals/read/compare.js +39 -0
- package/dist/evals/read/datasets.d.ts +105 -0
- package/dist/evals/read/datasets.js +318 -0
- package/dist/evals/read/list.d.ts +66 -0
- package/dist/evals/read/list.js +182 -0
- package/dist/evals/read/results.d.ts +46 -0
- package/dist/evals/read/results.js +107 -0
- package/dist/evals/read/runs.d.ts +154 -0
- package/dist/evals/read/runs.js +295 -0
- package/dist/evals/read/suggestions.d.ts +43 -0
- package/dist/evals/read/suggestions.js +73 -0
- package/dist/evals/runner/FakeAgentBackend.d.ts +40 -0
- package/dist/evals/runner/FakeAgentBackend.js +82 -0
- package/dist/evals/runner/agentCaseRunner.d.ts +24 -0
- package/dist/evals/runner/agentCaseRunner.js +388 -0
- package/dist/evals/runner/caseRunner.d.ts +8 -0
- package/dist/evals/runner/caseRunner.js +28 -0
- package/dist/evals/runner/conversationSubject.d.ts +18 -0
- package/dist/evals/runner/conversationSubject.js +123 -0
- package/dist/evals/runner/internalAgentCaseRunner.d.ts +6 -0
- package/dist/evals/runner/internalAgentCaseRunner.js +250 -0
- package/dist/evals/runner/simulatorModel.d.ts +51 -0
- package/dist/evals/runner/simulatorModel.js +58 -0
- package/dist/evals/runner/simulatorPrompt.d.ts +29 -0
- package/dist/evals/runner/simulatorPrompt.js +59 -0
- package/dist/evals/runner/stageCaseRunner.d.ts +30 -0
- package/dist/evals/runner/stageCaseRunner.js +279 -0
- package/dist/evals/runner/userDriver.d.ts +71 -0
- package/dist/evals/runner/userDriver.js +104 -0
- package/dist/evals/scoring.d.ts +34 -0
- package/dist/evals/scoring.js +67 -0
- package/dist/evals/snapshot.d.ts +191 -0
- package/dist/evals/snapshot.js +70 -0
- package/dist/evals/stageDiff.d.ts +8 -0
- package/dist/evals/stageDiff.js +48 -0
- package/dist/evals/stats.d.ts +95 -0
- package/dist/evals/stats.js +220 -0
- package/dist/evals/statsAdapter.d.ts +4 -0
- package/dist/evals/statsAdapter.js +30 -0
- package/dist/evals/statsTypes.d.ts +51 -0
- package/dist/evals/statsTypes.js +18 -0
- package/dist/evals/storeError.d.ts +5 -0
- package/dist/evals/storeError.js +8 -0
- package/dist/evals/storeTypes.d.ts +207 -0
- package/dist/evals/storeTypes.js +1 -0
- package/dist/evals/tTable.d.ts +3 -0
- package/dist/evals/tTable.js +15 -0
- package/dist/evals/triage/constants.d.ts +26 -0
- package/dist/evals/triage/constants.js +33 -0
- package/dist/evals/triage/credentials.d.ts +13 -0
- package/dist/evals/triage/credentials.js +20 -0
- package/dist/evals/triage/jevClassifier.d.ts +14 -0
- package/dist/evals/triage/jevClassifier.js +124 -0
- package/dist/evals/triage/judgeClassifier.d.ts +10 -0
- package/dist/evals/triage/judgeClassifier.js +76 -0
- package/dist/evals/triage/prompts.d.ts +16 -0
- package/dist/evals/triage/prompts.js +64 -0
- package/dist/evals/triage/service.d.ts +79 -0
- package/dist/evals/triage/service.js +217 -0
- package/dist/evals/triage/stateBuilder.d.ts +19 -0
- package/dist/evals/triage/stateBuilder.js +125 -0
- package/dist/evals/triage/types.d.ts +60 -0
- package/dist/evals/triage/types.js +8 -0
- package/dist/evals/types.d.ts +357 -0
- package/dist/evals/types.js +4 -0
- package/dist/guide/cardContract.d.ts +133 -0
- package/dist/guide/cardContract.js +156 -0
- package/dist/guide/chat.d.ts +45 -0
- package/dist/guide/chat.js +378 -0
- package/dist/guide/knowledge/index.d.ts +14 -0
- package/dist/guide/knowledge/index.js +109 -0
- package/dist/guide/knowledge/topics/agents.d.ts +2 -0
- package/dist/guide/knowledge/topics/agents.js +58 -0
- package/dist/guide/knowledge/topics/bestPractices.d.ts +2 -0
- package/dist/guide/knowledge/topics/bestPractices.js +67 -0
- package/dist/guide/knowledge/topics/connections.d.ts +2 -0
- package/dist/guide/knowledge/topics/connections.js +66 -0
- package/dist/guide/knowledge/topics/consoleMap.d.ts +2 -0
- package/dist/guide/knowledge/topics/consoleMap.js +55 -0
- package/dist/guide/knowledge/topics/evals.d.ts +2 -0
- package/dist/guide/knowledge/topics/evals.js +72 -0
- package/dist/guide/knowledge/topics/limitations.d.ts +2 -0
- package/dist/guide/knowledge/topics/limitations.js +84 -0
- package/dist/guide/knowledge/topics/mcpSkillsBrowser.d.ts +2 -0
- package/dist/guide/knowledge/topics/mcpSkillsBrowser.js +57 -0
- package/dist/guide/knowledge/topics/memories.d.ts +2 -0
- package/dist/guide/knowledge/topics/memories.js +63 -0
- package/dist/guide/knowledge/topics/modelsAndProviders.d.ts +2 -0
- package/dist/guide/knowledge/topics/modelsAndProviders.js +56 -0
- package/dist/guide/knowledge/topics/overview.d.ts +2 -0
- package/dist/guide/knowledge/topics/overview.js +56 -0
- package/dist/guide/knowledge/topics/pipelines.d.ts +2 -0
- package/dist/guide/knowledge/topics/pipelines.js +93 -0
- package/dist/guide/knowledge/topics/stages.d.ts +2 -0
- package/dist/guide/knowledge/topics/stages.js +54 -0
- package/dist/guide/knowledge/topics/tasks.d.ts +2 -0
- package/dist/guide/knowledge/topics/tasks.js +52 -0
- package/dist/guide/knowledge/topics/triggers.d.ts +2 -0
- package/dist/guide/knowledge/topics/triggers.js +68 -0
- package/dist/guide/knowledge/topics/troubleshooting.d.ts +2 -0
- package/dist/guide/knowledge/topics/troubleshooting.js +85 -0
- package/dist/guide/knowledgeTools.d.ts +2 -0
- package/dist/guide/knowledgeTools.js +56 -0
- package/dist/guide/showUi.d.ts +6 -0
- package/dist/guide/showUi.js +546 -0
- package/dist/guide/threadStore.d.ts +48 -0
- package/dist/guide/threadStore.js +189 -0
- package/dist/guide/writeRoot.d.ts +11 -0
- package/dist/guide/writeRoot.js +70 -0
- package/dist/hosted/translate.js +2 -0
- package/dist/mcp/agentTools.js +12 -0
- package/dist/mcp/catalogTools.js +18 -20
- package/dist/mcp/catalogWriteTools.d.ts +8 -0
- package/dist/mcp/catalogWriteTools.js +225 -0
- package/dist/mcp/connectionTools.d.ts +3 -0
- package/dist/mcp/connectionTools.js +85 -0
- package/dist/mcp/controlTools.js +34 -1
- package/dist/mcp/deps.d.ts +16 -0
- package/dist/mcp/evalTools.d.ts +3 -0
- package/dist/mcp/evalTools.js +943 -0
- package/dist/mcp/memoryHostTools.d.ts +3 -0
- package/dist/mcp/memoryHostTools.js +72 -0
- package/dist/mcp/pipelineAuthorTools.js +14 -1
- package/dist/mcp/server.js +6 -0
- package/dist/mcp/tools.js +8 -0
- package/dist/mcp/triggerTools.js +77 -31
- package/dist/models/response.d.ts +7 -2
- package/dist/models/response.js +6 -1
- package/dist/operatorAgent/agentAuthorContext.d.ts +1 -1
- package/dist/operatorAgent/agentAuthorContext.js +1 -1
- package/dist/operatorAgent/evalGate.d.ts +22 -0
- package/dist/operatorAgent/evalGate.js +77 -0
- package/dist/operatorAgent/fakeHost.js +10 -2
- package/dist/operatorAgent/host.js +18 -2
- package/dist/operatorAgent/index.d.ts +1 -0
- package/dist/operatorAgent/index.js +1 -0
- package/dist/operatorAgent/internalAgents.d.ts +11 -0
- package/dist/operatorAgent/internalAgents.js +13 -0
- package/dist/operatorAgent/piModel.d.ts +5 -1
- package/dist/operatorAgent/piModel.js +32 -14
- package/dist/operatorAgent/profiles/agentAuthor.d.ts +3 -1
- package/dist/operatorAgent/profiles/agentAuthor.js +201 -32
- package/dist/operatorAgent/profiles/agentAuthorPlaybook.d.ts +1 -1
- package/dist/operatorAgent/profiles/agentAuthorPlaybook.js +48 -15
- package/dist/operatorAgent/profiles/authorReadable.d.ts +2 -0
- package/dist/operatorAgent/profiles/authorReadable.js +26 -0
- package/dist/operatorAgent/profiles/guideConfirm.d.ts +21 -0
- package/dist/operatorAgent/profiles/guideConfirm.js +100 -0
- package/dist/operatorAgent/profiles/guideTools.d.ts +7 -0
- package/dist/operatorAgent/profiles/guideTools.js +123 -0
- package/dist/operatorAgent/profiles/stageflowGuide.d.ts +7 -0
- package/dist/operatorAgent/profiles/stageflowGuide.js +132 -0
- package/dist/operatorAgent/profiles/workshopAuthor.js +23 -0
- package/dist/operatorAgent/profiles/workshopAuthorPlaybook.d.ts +1 -1
- package/dist/operatorAgent/profiles/workshopAuthorPlaybook.js +9 -1
- package/dist/operatorAgent/profiles/workshopAuthorTools.d.ts +2 -0
- package/dist/operatorAgent/profiles/workshopAuthorTools.js +197 -9
- package/dist/operatorAgent/types.d.ts +23 -0
- package/dist/package-meta.d.ts +1 -1
- package/dist/package-meta.js +1 -1
- package/dist/runstore/port.d.ts +43 -0
- package/dist/runstore/retention.d.ts +2 -1
- package/dist/runstore/retention.js +3 -7
- package/dist/runstore/runStatusClock.d.ts +17 -0
- package/dist/runstore/runStatusClock.js +19 -0
- package/dist/runstore/sqlite/SqliteRunStore.d.ts +29 -0
- package/dist/runstore/sqlite/SqliteRunStore.js +144 -50
- package/dist/runstore/sqlite/evalStore.d.ts +20 -0
- package/dist/runstore/sqlite/evalStore.js +541 -0
- package/dist/runstore/sqlite/migrations/015-evals.d.ts +12 -0
- package/dist/runstore/sqlite/migrations/015-evals.js +157 -0
- package/dist/runstore/sqlite/migrations/016-eval-subject-ref.d.ts +12 -0
- package/dist/runstore/sqlite/migrations/016-eval-subject-ref.js +20 -0
- package/dist/runstore/sqlite/migrations/017-eval-result-suggestions.d.ts +11 -0
- package/dist/runstore/sqlite/migrations/017-eval-result-suggestions.js +36 -0
- package/dist/runstore/sqlite/migrations/index.js +6 -0
- package/dist/runstore/sqlite/suggestionStore.d.ts +14 -0
- package/dist/runstore/sqlite/suggestionStore.js +132 -0
- package/dist/runstore/workspaceLayout.js +6 -1
- package/dist/runtime/artifactDiskInspect.d.ts +14 -0
- package/dist/runtime/artifactDiskInspect.js +52 -0
- package/dist/runtime/artifactPath.d.ts +24 -0
- package/dist/runtime/artifactPath.js +84 -0
- package/dist/runtime/completionCheckRunner.js +15 -43
- package/dist/runtime/githubPollSource.d.ts +22 -4
- package/dist/runtime/githubPollSource.js +93 -8
- package/dist/runtime/pipelineRunner.d.ts +2 -1
- package/dist/runtime/pipelineRunner.js +2 -0
- package/dist/runtime/priorArtifactGrants.d.ts +8 -0
- package/dist/runtime/priorArtifactGrants.js +105 -0
- package/dist/runtime/runActionEligibility.d.ts +46 -0
- package/dist/runtime/runActionEligibility.js +123 -0
- package/dist/runtime/runManager.d.ts +6 -2
- package/dist/runtime/runManager.js +60 -16
- package/dist/runtime/runRetentionSweep.js +2 -2
- package/dist/runtime/runRetryCoordinator.d.ts +2 -14
- package/dist/runtime/runRetryCoordinator.js +3 -67
- package/dist/runtime/sandboxContainer.d.ts +7 -0
- package/dist/runtime/sandboxContainer.js +48 -6
- package/dist/runtime/sandboxDesk.d.ts +2 -1
- package/dist/runtime/sandboxDesk.js +11 -12
- package/dist/runtime/stageEnvironment.js +1 -0
- package/dist/runtime/stageRoots.d.ts +12 -0
- package/dist/runtime/stageSandbox.d.ts +2 -1
- package/dist/runtime/stageSandbox.js +18 -2
- package/dist/runtime/stageWorker.js +20 -0
- package/dist/server/agentAuthorRoutes.js +59 -1
- package/dist/server/bootstrap.d.ts +13 -0
- package/dist/server/bootstrap.js +69 -0
- package/dist/server/controlToken.js +3 -0
- package/dist/server/evalInternalRoutes.d.ts +14 -0
- package/dist/server/evalInternalRoutes.js +87 -0
- package/dist/server/evalRouteContext.d.ts +46 -0
- package/dist/server/evalRouteContext.js +43 -0
- package/dist/server/evalRoutes.d.ts +5 -0
- package/dist/server/evalRoutes.js +1456 -0
- package/dist/server/evalTriageRoutes.d.ts +2 -0
- package/dist/server/evalTriageRoutes.js +95 -0
- package/dist/server/guideRoutes.d.ts +5 -0
- package/dist/server/guideRoutes.js +166 -0
- package/dist/server/http.d.ts +15 -0
- package/dist/server/http.js +67 -3
- package/dist/server/internalAgentRoutes.d.ts +6 -0
- package/dist/server/internalAgentRoutes.js +12 -0
- package/dist/server/mcpHost.js +5 -0
- package/dist/toolkit/meta.d.ts +1 -1
- package/dist/toolkit/meta.js +53 -0
- package/dist/toolkit/profiles.d.ts +5 -0
- package/dist/toolkit/profiles.js +13 -1
- package/dist/tools/askOperator.js +2 -1
- package/dist/tools/emitStageEnvelope.d.ts +17 -16
- package/dist/tools/emitStageEnvelope.js +46 -32
- package/dist/tools/stageToolNames.d.ts +4 -0
- package/dist/tools/stageToolNames.js +4 -0
- package/dist/tools/workingState.d.ts +40 -0
- package/dist/tools/workingState.js +108 -0
- package/dist/tools/writeStageArtifact.d.ts +0 -9
- package/dist/tools/writeStageArtifact.js +2 -1
- package/dist/types/stageflowManifest.d.ts +6 -0
- package/dist/ui/assets/{MarkdownText-D7tYaIFO.js → MarkdownText-S5C_vemc.js} +2 -2
- package/dist/ui/assets/adapters-C0l4ruiR.js +17 -0
- package/dist/ui/assets/chat-CjnsmvqR.js +2 -0
- package/dist/ui/assets/chat-DFDd4UjS.css +2 -0
- package/dist/ui/assets/{main-BFo79krC.css → main-Del1GeLf.css} +1 -1
- package/dist/ui/assets/main-_yXVqcW9.js +277 -0
- package/dist/ui/chat.html +3 -4
- package/dist/ui/index.html +3 -4
- package/dist/workshop/chatTurn.js +1 -1
- package/dist/workshop/workshopPin.d.ts +3 -3
- package/dist/workshop/workshopPin.js +16 -15
- package/examples/README.md +2 -0
- package/examples/compaction-stress/README.md +30 -0
- package/examples/compaction-stress/check-run.mjs +53 -0
- package/examples/compaction-stress/collect-evidence.yaml +60 -0
- package/examples/compaction-stress/compaction-stress.pipeline.yaml +141 -0
- package/examples/compaction-stress/compaction-stress.task.yaml +15 -0
- package/examples/compaction-stress/generate-evidence.mjs +110 -0
- package/examples/compaction-stress/root-cause.yaml +26 -0
- package/examples/compaction-stress/stageflow.yaml +9 -0
- package/examples/compaction-stress/write-postmortem.yaml +40 -0
- package/examples/document-policy-review/README.md +35 -0
- package/examples/document-policy-review/agents/compliance.agent.yaml +42 -0
- package/examples/document-policy-review/documents/acme-vendor-sow.md +15 -0
- package/examples/document-policy-review/documents/support-runbook.md +13 -0
- package/examples/document-policy-review/policies/acceptable-use.md +13 -0
- package/examples/document-policy-review/policies/data-handling.md +17 -0
- package/examples/document-policy-review/policies/vendor-security.md +17 -0
- package/examples/evals/README.md +141 -0
- package/examples/evals/critique-sample.dataset.jsonl +8 -0
- package/examples/evals/critique-sample.eval.yaml +34 -0
- package/examples/evals/critique-sample.task.yaml +4 -0
- package/examples/evals/eval-sample.pipeline.yaml +35 -0
- package/examples/evals/guide-behavior.dataset.jsonl +15 -0
- package/examples/evals/guide-behavior.eval.yaml +35 -0
- package/examples/evals/guide-depth.dataset.jsonl +35 -0
- package/examples/evals/guide-depth.eval.yaml +169 -0
- package/examples/evals/support-sample.agent.yaml +20 -0
- package/examples/evals/support-sample.dataset.jsonl +6 -0
- package/examples/evals/support-sample.eval.yaml +38 -0
- package/examples/memory/evals/guide-behavior.dataset.jsonl +15 -0
- package/examples/memory/evals/guide-behavior.eval.yaml +35 -0
- package/examples/memory/evals/guide-depth.dataset.jsonl +35 -0
- package/examples/memory/evals/guide-depth.eval.yaml +169 -0
- package/examples/stageflow.yaml +6 -0
- package/package.json +1 -1
- package/skills/stageflow/SKILL.md +5 -2
- package/skills/stageflow/references/control-surface.md +4 -2
- package/skills/stageflow/references/evals.md +68 -0
- package/skills/stageflow-author/SKILL.md +4 -0
- package/skills/stageflow-run/SKILL.md +4 -0
- package/dist/ui/assets/AssistantRuntimeProvider-CsiPFhCT.js +0 -9
- package/dist/ui/assets/ReadonlyStore-BfXqHykm.js +0 -6
- package/dist/ui/assets/chat-BbLukq9W.js +0 -5
- package/dist/ui/assets/chat-zIZU_bP1.css +0 -2
- package/dist/ui/assets/main-CijYMeM0.js +0 -259
|
@@ -9,10 +9,13 @@ import { parsePipelineDagSnapshot } from "../pipelineDagSnapshot.js";
|
|
|
9
9
|
import { newRunId, runWorkspaceDir } from "../paths.js";
|
|
10
10
|
import { attemptAgentDir, attemptArtifactsDir, stageDir, } from "../workspaceLayout.js";
|
|
11
11
|
import { importDiskRunsIfEmpty } from "./migrateFromDisk.js";
|
|
12
|
+
import * as evalStore from "./evalStore.js";
|
|
13
|
+
import * as suggestionStore from "./suggestionStore.js";
|
|
12
14
|
import { applyPendingMigrations, assertSchemaVersion, } from "./migrations/index.js";
|
|
13
15
|
import { StoreSchemaError } from "./storeSchemaError.js";
|
|
14
16
|
import { applyStorePragmas } from "./applyStorePragmas.js";
|
|
15
17
|
import { RunSubmissionExistsError } from "../submission.js";
|
|
18
|
+
import { nextFinishedAt } from "../runStatusClock.js";
|
|
16
19
|
const EXECUTION_SELECT_COLS = `run_id, stage_id, attempt, status, verification_outcome, started_at, finished_at, envelope_json, cost_usd, usage_json, auto_resume_count`;
|
|
17
20
|
const ensuredStageDirs = new Set();
|
|
18
21
|
function executionFromRow(row) {
|
|
@@ -220,9 +223,9 @@ export class SqliteRunStore {
|
|
|
220
223
|
}
|
|
221
224
|
const inserted = this.db
|
|
222
225
|
.prepare(`INSERT INTO runs
|
|
223
|
-
(run_id, pipeline_id, task_id, task_yaml, status, created_at, updated_at, checkout_root, pipeline_dag_json, git_sha, ci_pr_url, ci_job_url, pipeline_path, task_path, project_root, repository, ref, resolved_sha, run_branch, git_author_name, git_author_email, pipeline_source, pipeline_body, caller_id, run_manifest, skip_gates, dispatch_key)
|
|
226
|
+
(run_id, pipeline_id, task_id, task_yaml, status, created_at, updated_at, checkout_root, pipeline_dag_json, git_sha, ci_pr_url, ci_job_url, pipeline_path, task_path, project_root, repository, ref, resolved_sha, run_branch, git_author_name, git_author_email, pipeline_source, pipeline_body, caller_id, origin, run_manifest, skip_gates, dispatch_key)
|
|
224
227
|
VALUES
|
|
225
|
-
(@run_id, @pipeline_id, @task_id, @task_yaml, @status, @created_at, @updated_at, @checkout_root, @pipeline_dag_json, @git_sha, @ci_pr_url, @ci_job_url, @pipeline_path, @task_path, @project_root, @repository, @ref, @resolved_sha, @run_branch, @git_author_name, @git_author_email, @pipeline_source, @pipeline_body, @caller_id, @run_manifest, @skip_gates, @dispatch_key)
|
|
228
|
+
(@run_id, @pipeline_id, @task_id, @task_yaml, @status, @created_at, @updated_at, @checkout_root, @pipeline_dag_json, @git_sha, @ci_pr_url, @ci_job_url, @pipeline_path, @task_path, @project_root, @repository, @ref, @resolved_sha, @run_branch, @git_author_name, @git_author_email, @pipeline_source, @pipeline_body, @caller_id, @origin, @run_manifest, @skip_gates, @dispatch_key)
|
|
226
229
|
ON CONFLICT(dispatch_key) DO NOTHING`)
|
|
227
230
|
.run({
|
|
228
231
|
run_id: runId,
|
|
@@ -252,6 +255,7 @@ export class SqliteRunStore {
|
|
|
252
255
|
pipeline_source: pipelineSource,
|
|
253
256
|
pipeline_body: pipelineBody,
|
|
254
257
|
caller_id: input.callerId ?? null,
|
|
258
|
+
origin: input.origin ?? null,
|
|
255
259
|
run_manifest: input.runManifest !== undefined
|
|
256
260
|
? JSON.stringify(input.runManifest)
|
|
257
261
|
: null,
|
|
@@ -286,58 +290,57 @@ export class SqliteRunStore {
|
|
|
286
290
|
await this.ready();
|
|
287
291
|
return this.readSubmission(key);
|
|
288
292
|
}
|
|
293
|
+
applyRunStatusUpdate(runId, status, expectedStatus) {
|
|
294
|
+
const now = new Date().toISOString();
|
|
295
|
+
const current = this.db
|
|
296
|
+
.prepare(`SELECT finished_at FROM runs WHERE run_id = ?`)
|
|
297
|
+
.get(runId);
|
|
298
|
+
if (current === undefined) {
|
|
299
|
+
throw new Error(`Run not found: ${runId}`);
|
|
300
|
+
}
|
|
301
|
+
const decision = nextFinishedAt({
|
|
302
|
+
nextStatus: status,
|
|
303
|
+
currentFinishedAt: current.finished_at,
|
|
304
|
+
now,
|
|
305
|
+
});
|
|
306
|
+
const params = {
|
|
307
|
+
run_id: runId,
|
|
308
|
+
status,
|
|
309
|
+
updated_at: now,
|
|
310
|
+
};
|
|
311
|
+
let finishedSql;
|
|
312
|
+
if (decision.action === "stamp") {
|
|
313
|
+
finishedSql = "finished_at = @finished_at";
|
|
314
|
+
params.finished_at = decision.finishedAt;
|
|
315
|
+
}
|
|
316
|
+
else if (decision.action === "clear") {
|
|
317
|
+
finishedSql = "finished_at = NULL";
|
|
318
|
+
}
|
|
319
|
+
else {
|
|
320
|
+
finishedSql = "finished_at = finished_at";
|
|
321
|
+
}
|
|
322
|
+
const expectedClause = expectedStatus !== undefined
|
|
323
|
+
? " AND status = @expected_status"
|
|
324
|
+
: "";
|
|
325
|
+
if (expectedStatus !== undefined) {
|
|
326
|
+
params.expected_status = expectedStatus;
|
|
327
|
+
}
|
|
328
|
+
const result = this.db
|
|
329
|
+
.prepare(`UPDATE runs SET status = @status, updated_at = @updated_at, ${finishedSql}
|
|
330
|
+
WHERE run_id = @run_id${expectedClause}`)
|
|
331
|
+
.run(params);
|
|
332
|
+
return result.changes;
|
|
333
|
+
}
|
|
289
334
|
async updateRunStatus(runId, status) {
|
|
290
335
|
await this.ready();
|
|
291
|
-
|
|
292
|
-
const isTerminal = status === "succeeded" || status === "failed" || status === "cancelled";
|
|
293
|
-
const result = isTerminal
|
|
294
|
-
? this.db
|
|
295
|
-
.prepare(`UPDATE runs SET status = @status, updated_at = @updated_at,
|
|
296
|
-
finished_at = COALESCE(finished_at, @finished_at)
|
|
297
|
-
WHERE run_id = @run_id`)
|
|
298
|
-
.run({
|
|
299
|
-
run_id: runId,
|
|
300
|
-
status,
|
|
301
|
-
updated_at: now,
|
|
302
|
-
finished_at: now,
|
|
303
|
-
})
|
|
304
|
-
: this.db
|
|
305
|
-
.prepare(`UPDATE runs SET status = @status, updated_at = @updated_at WHERE run_id = @run_id`)
|
|
306
|
-
.run({
|
|
307
|
-
run_id: runId,
|
|
308
|
-
status,
|
|
309
|
-
updated_at: now,
|
|
310
|
-
});
|
|
311
|
-
if (result.changes === 0) {
|
|
336
|
+
if (this.applyRunStatusUpdate(runId, status, undefined) === 0) {
|
|
312
337
|
throw new Error(`Run not found: ${runId}`);
|
|
313
338
|
}
|
|
314
339
|
}
|
|
315
340
|
async tryUpdateRunStatus(runId, status, expectedStatus) {
|
|
316
341
|
await this.ready();
|
|
317
|
-
const
|
|
318
|
-
|
|
319
|
-
const result = isTerminal
|
|
320
|
-
? this.db
|
|
321
|
-
.prepare(`UPDATE runs SET status = @status, updated_at = @updated_at,
|
|
322
|
-
finished_at = COALESCE(finished_at, @finished_at)
|
|
323
|
-
WHERE run_id = @run_id AND status = @expected_status`)
|
|
324
|
-
.run({
|
|
325
|
-
run_id: runId,
|
|
326
|
-
status,
|
|
327
|
-
expected_status: expectedStatus,
|
|
328
|
-
updated_at: now,
|
|
329
|
-
finished_at: now,
|
|
330
|
-
})
|
|
331
|
-
: this.db
|
|
332
|
-
.prepare(`UPDATE runs SET status = @status, updated_at = @updated_at
|
|
333
|
-
WHERE run_id = @run_id AND status = @expected_status`)
|
|
334
|
-
.run({
|
|
335
|
-
run_id: runId,
|
|
336
|
-
status,
|
|
337
|
-
expected_status: expectedStatus,
|
|
338
|
-
updated_at: now,
|
|
339
|
-
});
|
|
340
|
-
if (result.changes > 0)
|
|
342
|
+
const changes = this.applyRunStatusUpdate(runId, status, expectedStatus);
|
|
343
|
+
if (changes > 0)
|
|
341
344
|
return true;
|
|
342
345
|
const exists = this.db
|
|
343
346
|
.prepare(`SELECT 1 AS ok FROM runs WHERE run_id = ?`)
|
|
@@ -854,9 +857,15 @@ export class SqliteRunStore {
|
|
|
854
857
|
clauses.push("caller_id = ?");
|
|
855
858
|
params.push(filter.caller_id);
|
|
856
859
|
}
|
|
860
|
+
if (filter?.origin === "eval") {
|
|
861
|
+
clauses.push("origin = 'eval'");
|
|
862
|
+
}
|
|
863
|
+
else if (filter?.origin !== "any") {
|
|
864
|
+
clauses.push("origin IS NOT 'eval'");
|
|
865
|
+
}
|
|
857
866
|
const where = clauses.length > 0 ? `WHERE ${clauses.join(" AND ")}` : "";
|
|
858
867
|
const rows = this.db
|
|
859
|
-
.prepare(`SELECT run_id, pipeline_id, task_id, task_yaml, status, created_at, updated_at, checkout_root, pipeline_dag_json, git_sha, ci_pr_url, ci_job_url, pipeline_path, task_path, project_root, repository, ref, resolved_sha, run_branch, git_author_name, git_author_email, cancel_reason, finished_at, slimmed_at, disk_bytes, disk_measured_at, config_origins_json, pipeline_source, pipeline_body, caller_id, run_manifest, skip_gates
|
|
868
|
+
.prepare(`SELECT run_id, pipeline_id, task_id, task_yaml, status, created_at, updated_at, checkout_root, pipeline_dag_json, git_sha, ci_pr_url, ci_job_url, pipeline_path, task_path, project_root, repository, ref, resolved_sha, run_branch, git_author_name, git_author_email, cancel_reason, finished_at, slimmed_at, disk_bytes, disk_measured_at, config_origins_json, pipeline_source, pipeline_body, caller_id, origin, run_manifest, skip_gates
|
|
860
869
|
FROM runs ${where} ORDER BY created_at DESC`)
|
|
861
870
|
.all(...params);
|
|
862
871
|
const summaries = [];
|
|
@@ -876,7 +885,7 @@ export class SqliteRunStore {
|
|
|
876
885
|
.prepare(`SELECT r.run_id, r.pipeline_id, r.status, r.created_at, r.finished_at,
|
|
877
886
|
(SELECT SUM(e.cost_usd) FROM stage_executions e WHERE e.run_id = r.run_id) AS cost_usd
|
|
878
887
|
FROM runs r
|
|
879
|
-
WHERE r.created_at >= ?
|
|
888
|
+
WHERE r.created_at >= ? AND r.origin IS NOT 'eval'
|
|
880
889
|
ORDER BY r.created_at DESC`)
|
|
881
890
|
.all(since);
|
|
882
891
|
return rows.map((row) => ({
|
|
@@ -888,6 +897,90 @@ export class SqliteRunStore {
|
|
|
888
897
|
...(row.cost_usd != null ? { cost_usd: row.cost_usd } : {}),
|
|
889
898
|
}));
|
|
890
899
|
}
|
|
900
|
+
async createEvalRun(input) {
|
|
901
|
+
await this.ready();
|
|
902
|
+
return evalStore.createEvalRun(this.db, input);
|
|
903
|
+
}
|
|
904
|
+
async insertEvalCases(evalRunId, cases) {
|
|
905
|
+
await this.ready();
|
|
906
|
+
evalStore.insertEvalCases(this.db, evalRunId, cases);
|
|
907
|
+
}
|
|
908
|
+
async upsertEvalResult(result) {
|
|
909
|
+
await this.ready();
|
|
910
|
+
evalStore.upsertEvalResult(this.db, result);
|
|
911
|
+
}
|
|
912
|
+
async updateEvalRunProgress(evalRunId, progress) {
|
|
913
|
+
await this.ready();
|
|
914
|
+
evalStore.updateEvalRunProgress(this.db, evalRunId, progress);
|
|
915
|
+
}
|
|
916
|
+
async finishEvalRun(evalRunId, input) {
|
|
917
|
+
await this.ready();
|
|
918
|
+
evalStore.finishEvalRun(this.db, evalRunId, input);
|
|
919
|
+
}
|
|
920
|
+
async getEvalRun(evalRunId) {
|
|
921
|
+
await this.ready();
|
|
922
|
+
return evalStore.getEvalRun(this.db, evalRunId);
|
|
923
|
+
}
|
|
924
|
+
async getEvalRunByNumber(projectRoot, evalId, runNumber) {
|
|
925
|
+
await this.ready();
|
|
926
|
+
return evalStore.getEvalRunByNumber(this.db, projectRoot, evalId, runNumber);
|
|
927
|
+
}
|
|
928
|
+
async listEvalRuns(query) {
|
|
929
|
+
await this.ready();
|
|
930
|
+
return evalStore.listEvalRuns(this.db, query);
|
|
931
|
+
}
|
|
932
|
+
async listEvalResults(evalRunId) {
|
|
933
|
+
await this.ready();
|
|
934
|
+
return evalStore.listEvalResults(this.db, evalRunId);
|
|
935
|
+
}
|
|
936
|
+
async getEvalResultsForCase(evalRunId, caseId) {
|
|
937
|
+
await this.ready();
|
|
938
|
+
return evalStore.getEvalResultsForCase(this.db, evalRunId, caseId);
|
|
939
|
+
}
|
|
940
|
+
async getEvalCases(evalRunId) {
|
|
941
|
+
await this.ready();
|
|
942
|
+
return evalStore.getEvalCases(this.db, evalRunId);
|
|
943
|
+
}
|
|
944
|
+
async putEvalReview(evalRunId, caseId, repeatIndex, patch) {
|
|
945
|
+
await this.ready();
|
|
946
|
+
return evalStore.putEvalReview(this.db, evalRunId, caseId, repeatIndex, patch);
|
|
947
|
+
}
|
|
948
|
+
async getEvalReviews(evalRunId) {
|
|
949
|
+
await this.ready();
|
|
950
|
+
return evalStore.getEvalReviews(this.db, evalRunId);
|
|
951
|
+
}
|
|
952
|
+
async listFailureTypes(projectRoot, evalId) {
|
|
953
|
+
await this.ready();
|
|
954
|
+
return evalStore.listFailureTypes(this.db, projectRoot, evalId);
|
|
955
|
+
}
|
|
956
|
+
async recordGraderVersions(projectRoot, evalId, graders) {
|
|
957
|
+
await this.ready();
|
|
958
|
+
return evalStore.recordGraderVersions(this.db, projectRoot, evalId, graders);
|
|
959
|
+
}
|
|
960
|
+
async listEvalSubjectRefs(evalRunId, targetKind) {
|
|
961
|
+
await this.ready();
|
|
962
|
+
return evalStore.listEvalSubjectRefs(this.db, evalRunId, targetKind);
|
|
963
|
+
}
|
|
964
|
+
async upsertEvalSuggestions(evalRunId, inputs) {
|
|
965
|
+
await this.ready();
|
|
966
|
+
return suggestionStore.upsertEvalSuggestions(this.db, evalRunId, inputs);
|
|
967
|
+
}
|
|
968
|
+
async listEvalSuggestions(evalRunId, status) {
|
|
969
|
+
await this.ready();
|
|
970
|
+
return suggestionStore.listEvalSuggestions(this.db, evalRunId, status);
|
|
971
|
+
}
|
|
972
|
+
async dismissEvalSuggestions(evalRunId, ids) {
|
|
973
|
+
await this.ready();
|
|
974
|
+
return suggestionStore.dismissEvalSuggestions(this.db, evalRunId, ids);
|
|
975
|
+
}
|
|
976
|
+
async acceptEvalSuggestions(evalRunId, ids) {
|
|
977
|
+
await this.ready();
|
|
978
|
+
return suggestionStore.acceptEvalSuggestions(this.db, evalRunId, ids);
|
|
979
|
+
}
|
|
980
|
+
async deleteEvalRun(evalRunId) {
|
|
981
|
+
await this.ready();
|
|
982
|
+
return evalStore.deleteEvalRun(this.db, evalRunId);
|
|
983
|
+
}
|
|
891
984
|
async ensureProject(absPath) {
|
|
892
985
|
await this.ready();
|
|
893
986
|
const resolved = path.resolve(absPath);
|
|
@@ -1450,6 +1543,7 @@ export class SqliteRunStore {
|
|
|
1450
1543
|
? { pipeline_source: row.pipeline_source }
|
|
1451
1544
|
: {}),
|
|
1452
1545
|
...(row.caller_id != null ? { caller_id: row.caller_id } : {}),
|
|
1546
|
+
...(row.origin === "eval" ? { origin: "eval" } : {}),
|
|
1453
1547
|
...(row.run_manifest
|
|
1454
1548
|
? { run_manifest: JSON.parse(row.run_manifest) }
|
|
1455
1549
|
: {}),
|
|
@@ -1464,7 +1558,7 @@ export class SqliteRunStore {
|
|
|
1464
1558
|
}
|
|
1465
1559
|
getRunRow(runId) {
|
|
1466
1560
|
const row = this.db
|
|
1467
|
-
.prepare(`SELECT run_id, pipeline_id, task_id, task_yaml, status, created_at, updated_at, checkout_root, pipeline_dag_json, git_sha, ci_pr_url, ci_job_url, pipeline_path, task_path, project_root, repository, ref, resolved_sha, run_branch, git_author_name, git_author_email, cancel_reason, finished_at, slimmed_at, disk_bytes, disk_measured_at, config_origins_json, pipeline_source, pipeline_body, caller_id, run_manifest, skip_gates
|
|
1561
|
+
.prepare(`SELECT run_id, pipeline_id, task_id, task_yaml, status, created_at, updated_at, checkout_root, pipeline_dag_json, git_sha, ci_pr_url, ci_job_url, pipeline_path, task_path, project_root, repository, ref, resolved_sha, run_branch, git_author_name, git_author_email, cancel_reason, finished_at, slimmed_at, disk_bytes, disk_measured_at, config_origins_json, pipeline_source, pipeline_body, caller_id, origin, run_manifest, skip_gates
|
|
1468
1562
|
FROM runs WHERE run_id = ?`)
|
|
1469
1563
|
.get(runId);
|
|
1470
1564
|
if (!row)
|
|
@@ -0,0 +1,20 @@
|
|
|
1
|
+
import type Database from "better-sqlite3";
|
|
2
|
+
import { type EvalTargetKind } from "../../evals/constants.js";
|
|
3
|
+
import type { CreateEvalRunInput, EvalCaseInput, EvalCaseRow, EvalGraderVersionInput, EvalResultInput, EvalResultWithGrades, EvalReviewPatch, EvalReviewRow, EvalRunPage, EvalRunProgress, EvalRunRow, FinishEvalRunInput, ListEvalRunsQuery } from "../../evals/storeTypes.js";
|
|
4
|
+
export declare function createEvalRun(db: Database.Database, input: CreateEvalRunInput): EvalRunRow;
|
|
5
|
+
export declare function insertEvalCases(db: Database.Database, evalRunId: string, cases: EvalCaseInput[]): void;
|
|
6
|
+
export declare function upsertEvalResult(db: Database.Database, input: EvalResultInput): void;
|
|
7
|
+
export declare function updateEvalRunProgress(db: Database.Database, evalRunId: string, progress: EvalRunProgress): void;
|
|
8
|
+
export declare function finishEvalRun(db: Database.Database, evalRunId: string, input: FinishEvalRunInput): void;
|
|
9
|
+
export declare function getEvalRun(db: Database.Database, evalRunId: string): EvalRunRow | null;
|
|
10
|
+
export declare function getEvalRunByNumber(db: Database.Database, projectRoot: string, evalId: string, runNumber: number): EvalRunRow | null;
|
|
11
|
+
export declare function listEvalRuns(db: Database.Database, query: ListEvalRunsQuery): EvalRunPage;
|
|
12
|
+
export declare function listEvalResults(db: Database.Database, evalRunId: string): EvalResultWithGrades[];
|
|
13
|
+
export declare function getEvalResultsForCase(db: Database.Database, evalRunId: string, caseId: string): EvalResultWithGrades[];
|
|
14
|
+
export declare function getEvalCases(db: Database.Database, evalRunId: string): EvalCaseRow[];
|
|
15
|
+
export declare function putEvalReview(db: Database.Database, evalRunId: string, caseId: string, repeatIndex: number, patch: EvalReviewPatch): EvalReviewRow;
|
|
16
|
+
export declare function getEvalReviews(db: Database.Database, evalRunId: string): EvalReviewRow[];
|
|
17
|
+
export declare function listFailureTypes(db: Database.Database, projectRoot: string, evalId: string): string[];
|
|
18
|
+
export declare function recordGraderVersions(db: Database.Database, projectRoot: string, evalId: string, graders: EvalGraderVersionInput[]): Record<string, number>;
|
|
19
|
+
export declare function listEvalSubjectRefs(db: Database.Database, evalRunId: string, targetKind: EvalTargetKind): string[];
|
|
20
|
+
export declare function deleteEvalRun(db: Database.Database, evalRunId: string): string[];
|