minimal-agent 0.13.0 → 0.14.0
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 +6 -5
- package/package.json +60 -4
- package/plugins/computer_use/.claude-plugin/plugin.json +1 -1
- package/plugins/cron-runner/src/commands.js +4 -4
- package/plugins/cron-runner/src/commands.ts +4 -4
- package/plugins/workflow-runner/.claude-plugin/plugin.json +2 -2
- package/plugins/workflow-runner/commands/workflow.md +57 -90
- package/plugins/workflow-runner/commands/workflows.md +4 -4
- package/plugins/workflow-runner/plugin.js +24 -13
- package/plugins/workflow-runner/plugin.ts +28 -13
- package/plugins/workflow-runner/src/discovery.js +160 -0
- package/plugins/workflow-runner/src/discovery.ts +194 -0
- package/plugins/workflow-runner/src/goalAnchor.js +53 -0
- package/plugins/workflow-runner/src/goalAnchor.ts +59 -0
- package/plugins/workflow-runner/src/index.js +197 -206
- package/plugins/workflow-runner/src/index.ts +248 -249
- package/plugins/workflow-runner/src/inputCoercion.js +40 -0
- package/plugins/workflow-runner/src/inputCoercion.ts +43 -0
- package/plugins/workflow-runner/src/journal.js +224 -0
- package/plugins/workflow-runner/src/journal.ts +274 -0
- package/plugins/workflow-runner/src/jsonSchema.js +132 -0
- package/plugins/workflow-runner/src/jsonSchema.ts +162 -0
- package/plugins/workflow-runner/src/moduleResolution.js +64 -0
- package/plugins/workflow-runner/src/moduleResolution.ts +71 -0
- package/plugins/workflow-runner/src/resourceInputs.js +220 -0
- package/plugins/workflow-runner/src/resourceInputs.ts +300 -0
- package/plugins/workflow-runner/src/runner.js +400 -489
- package/plugins/workflow-runner/src/runner.ts +488 -560
- package/plugins/workflow-runner/src/runtime/config.js +46 -0
- package/plugins/workflow-runner/src/runtime/config.ts +69 -0
- package/plugins/workflow-runner/src/runtime/context.js +1304 -0
- package/plugins/workflow-runner/src/runtime/context.ts +1786 -0
- package/plugins/workflow-runner/src/runtime/exec.js +34 -0
- package/plugins/workflow-runner/src/runtime/exec.ts +53 -0
- package/plugins/workflow-runner/src/runtime/flow.js +8 -0
- package/plugins/workflow-runner/src/runtime/flow.ts +12 -0
- package/plugins/workflow-runner/src/runtime/fs.js +27 -0
- package/plugins/workflow-runner/src/runtime/fs.ts +41 -0
- package/plugins/workflow-runner/src/runtime/model.js +6 -0
- package/plugins/workflow-runner/src/runtime/model.ts +8 -0
- package/plugins/workflow-runner/src/runtime/output.js +7 -0
- package/plugins/workflow-runner/src/runtime/output.ts +21 -0
- package/plugins/workflow-runner/src/runtime/scheduler.js +141 -0
- package/plugins/workflow-runner/src/runtime/scheduler.ts +166 -0
- package/plugins/workflow-runner/src/runtime/state.js +16 -0
- package/plugins/workflow-runner/src/runtime/state.ts +19 -0
- package/plugins/workflow-runner/src/runtime/tools.js +3 -0
- package/plugins/workflow-runner/src/runtime/tools.ts +6 -0
- package/plugins/workflow-runner/src/sandbox.js +141 -0
- package/plugins/workflow-runner/src/sandbox.ts +182 -0
- package/plugins/workflow-runner/src/scriptGuard.d.mts +38 -0
- package/plugins/workflow-runner/src/scriptGuard.mjs +449 -0
- package/plugins/workflow-runner/src/taskIdentity.js +151 -307
- package/plugins/workflow-runner/src/taskIdentity.ts +207 -355
- package/plugins/workflow-runner/src/toolRuntime.js +29 -8
- package/plugins/workflow-runner/src/toolRuntime.ts +36 -11
- package/plugins/workflow-runner/src/types.js +7 -211
- package/plugins/workflow-runner/src/types.ts +154 -623
- package/plugins/workflow-runner/src/util.js +47 -0
- package/plugins/workflow-runner/src/util.ts +51 -0
- package/plugins/workflow-runner/src/workflowContext.js +20 -0
- package/plugins/workflow-runner/src/workflowContext.ts +58 -0
- package/plugins/workflow-runner/src/workflowDependencies.js +36 -0
- package/plugins/workflow-runner/src/workflowDependencies.ts +59 -0
- package/plugins/workflow-runner/src/workflowFingerprint.js +118 -0
- package/plugins/workflow-runner/src/workflowFingerprint.ts +122 -0
- package/plugins/workflow-runner/src/workflowState.js +474 -802
- package/plugins/workflow-runner/src/workflowState.ts +575 -970
- package/plugins/workflow-runner/test/api.test.ts +223 -0
- package/plugins/workflow-runner/test/apiCapabilities.test.ts +614 -0
- package/plugins/workflow-runner/test/apiConcurrencyGate.test.ts +104 -0
- package/plugins/workflow-runner/test/apiEcho.test.ts +210 -0
- package/plugins/workflow-runner/test/apiReadText.test.ts +178 -0
- package/plugins/workflow-runner/test/apiResourceLedger.test.ts +184 -0
- package/plugins/workflow-runner/test/childWorkflow.test.ts +304 -0
- package/plugins/workflow-runner/test/discovery.test.ts +234 -0
- package/plugins/workflow-runner/test/goalAnchor.test.ts +106 -0
- package/plugins/workflow-runner/test/goalController.test.ts +317 -0
- package/plugins/workflow-runner/test/journalLanes.test.ts +198 -0
- package/plugins/workflow-runner/test/listResources.test.ts +106 -0
- package/plugins/workflow-runner/test/resourceInputs.test.ts +294 -0
- package/plugins/workflow-runner/test/sandbox.test.ts +126 -0
- package/plugins/workflow-runner/test/scheduler.test.ts +70 -0
- package/plugins/workflow-runner/test/scriptGuard.test.ts +276 -0
- package/plugins/workflow-runner/test/skillDocSync.test.ts +78 -0
- package/plugins/workflow-runner/test/taskIdentity.test.ts +288 -179
- package/plugins/workflow-runner/test/workflowFingerprint.test.ts +106 -0
- package/plugins/workflow-runner/test/workflowState.test.ts +429 -0
- package/skills/pptx/examples/assets/cover-bg.svg +23 -0
- package/skills/pptx/examples/assets/icon-spark.svg +5 -0
- package/skills/ui-design/SKILL.md +5 -2
- package/skills/ui-design/references/layout-composition.md +1 -1
- package/skills/workflow-creator/SKILL.md +344 -0
- package/skills/workflow-creator/assets/template/workflow.mjs +48 -0
- package/skills/workflow-creator/references/api.md +216 -0
- package/skills/workflow-creator/references/migration.md +103 -0
- package/skills/workflow-creator/references/patterns.md +779 -0
- package/skills/workflow-creator/scripts/list-resources.mjs +205 -0
- package/skills/workflow-creator/scripts/scaffold.mjs +148 -0
- package/skills/workflow-creator/scripts/validate.mjs +275 -0
- package/skills/workflow-creator/scripts/workflow-creator.test.mjs +82 -0
- package/src/advisor/advisorCoordinator.js +503 -0
- package/src/advisor/advisorCoordinator.ts +657 -0
- package/src/cli/configMenu.tsx +0 -1
- package/src/cli/print.js +27 -6
- package/src/cli/print.ts +48 -4
- package/src/cli/streamJson.js +5 -6
- package/src/cli/streamJson.ts +8 -4
- package/src/cli/wizards/advisorWizard.js +4 -2
- package/src/cli/wizards/advisorWizard.tsx +8 -2
- package/src/cli/wizards/contextAdvisorWizard.js +5 -4
- package/src/cli/wizards/contextAdvisorWizard.tsx +7 -6
- package/src/collaboration/taskFlow.js +115 -0
- package/src/collaboration/taskFlow.ts +169 -0
- package/src/config/configDefaults.js +41 -1
- package/src/config/configDefaults.ts +43 -1
- package/src/config/configDiagnostics.ts +2 -2
- package/src/config/configFile.js +40 -9
- package/src/config/configFile.ts +60 -11
- package/src/config/configService.js +419 -74
- package/src/config/configService.ts +703 -145
- package/src/config/numericLimits.js +22 -0
- package/src/config/numericLimits.ts +24 -0
- package/src/config.js +1 -0
- package/src/config.ts +3 -2
- package/src/context/budgetPlan.js +1 -1
- package/src/context/budgetPlan.ts +1 -1
- package/src/context/compact.js +7 -2
- package/src/context/compact.ts +9 -2
- package/src/context/microCompactLite.js +212 -13
- package/src/context/microCompactLite.ts +261 -12
- package/src/context/sessionRegistry.js +8 -1
- package/src/context/sessionRegistry.ts +18 -5
- package/src/context/toolResultArchive.js +72 -0
- package/src/context/toolResultArchive.ts +95 -0
- package/src/goal/errors.js +8 -0
- package/src/goal/errors.ts +18 -0
- package/src/goal/goalRuntime.js +344 -0
- package/src/goal/goalRuntime.ts +466 -0
- package/src/goal/goalStore.js +200 -0
- package/src/goal/goalStore.ts +219 -0
- package/src/goal/index.js +4 -0
- package/src/goal/index.ts +4 -0
- package/src/goal/statusMapping.js +71 -0
- package/src/goal/statusMapping.ts +95 -0
- package/src/goal/types.js +7 -0
- package/src/goal/types.ts +179 -0
- package/src/goal/workflowController.js +306 -0
- package/src/goal/workflowController.ts +360 -0
- package/src/harness/evidenceLedger.js +2 -0
- package/src/harness/evidenceLedger.ts +1 -0
- package/src/harness/stallGuard.js +9 -3
- package/src/harness/stallGuard.ts +7 -3
- package/src/images/imageResolver.js +161 -0
- package/src/images/imageResolver.ts +189 -0
- package/src/infra/verification.js +34 -5
- package/src/infra/verification.ts +39 -4
- package/src/llm/client.js +79 -21
- package/src/llm/client.ts +182 -35
- package/src/llm/gatewayClient.js +201 -0
- package/src/llm/gatewayClient.ts +299 -0
- package/src/llm/messageProvenance.js +46 -0
- package/src/llm/messageProvenance.ts +58 -0
- package/src/llm/modelProvider.js +1 -0
- package/src/llm/modelProvider.ts +22 -0
- package/src/llm/providers/anthropicFormat.js +108 -1
- package/src/llm/providers/anthropicFormat.ts +115 -4
- package/src/llm/providers/openaiFormat.js +9 -0
- package/src/llm/providers/openaiFormat.ts +12 -3
- package/src/loop.js +319 -54
- package/src/loop.ts +373 -57
- package/src/main.js +4 -1
- package/src/main.tsx +4 -1
- package/src/managedExecution/index.js +2 -0
- package/src/managedExecution/index.ts +2 -0
- package/src/managedExecution/registry.js +15 -0
- package/src/managedExecution/registry.ts +20 -0
- package/src/managedExecution/types.js +1 -0
- package/src/managedExecution/types.ts +86 -0
- package/src/modelRef.js +9 -0
- package/src/modelRef.ts +31 -0
- package/src/multimodal.js +14 -28
- package/src/multimodal.ts +17 -33
- package/src/net/lanDiscovery.js +65 -0
- package/src/net/lanDiscovery.ts +82 -0
- package/src/plugin-sdk.js +28 -3
- package/src/plugin-sdk.ts +39 -6
- package/src/plugins/pluginApi.ts +16 -1
- package/src/plugins/pluginLoader.js +4 -0
- package/src/plugins/pluginLoader.ts +6 -0
- package/src/plugins/pluginRunner.js +139 -7
- package/src/plugins/pluginRunner.ts +173 -9
- package/src/plugins/slashFallback.js +113 -0
- package/src/plugins/slashFallback.ts +137 -0
- package/src/prompts/policyRegistry.js +12 -32
- package/src/prompts/policyRegistry.ts +15 -32
- package/src/prompts/promptAssembly.js +103 -12
- package/src/prompts/promptAssembly.ts +114 -12
- package/src/prompts/skillList.js +21 -3
- package/src/prompts/skillList.ts +27 -2
- package/src/prompts/stableSystemPrompt.js +18 -0
- package/src/prompts/stableSystemPrompt.ts +33 -0
- package/src/runtime/agentRuntimeContext.js +8 -2
- package/src/runtime/agentRuntimeContext.ts +17 -4
- package/src/session/runAgentTurn.js +78 -12
- package/src/session/runAgentTurn.ts +105 -23
- package/src/session/runTurn.js +10 -9
- package/src/session/runTurn.ts +17 -8
- package/src/session/sessionState.ts +2 -0
- package/src/team/auth.js +186 -0
- package/src/team/auth.ts +298 -0
- package/src/team/cli.js +105 -0
- package/src/team/cli.ts +122 -0
- package/src/team/onlineToolContract.js +64 -0
- package/src/team/onlineToolContract.ts +109 -0
- package/src/team/toolPolicy.js +36 -0
- package/src/team/toolPolicy.ts +75 -0
- package/src/tools/advisor/advisor.js +90 -46
- package/src/tools/advisor/advisor.ts +100 -47
- package/src/tools/advisor/advisorDecision.js +71 -0
- package/src/tools/advisor/advisorDecision.ts +85 -0
- package/src/tools/agent/agent.js +18 -5
- package/src/tools/agent/agent.ts +21 -7
- package/src/tools/agent/agentControl.js +246 -0
- package/src/tools/agent/agentControl.ts +284 -0
- package/src/tools/agent/agentTaskService.js +40 -1
- package/src/tools/agent/agentTaskService.ts +44 -1
- package/src/tools/agent/agentTypes.js +14 -2
- package/src/tools/agent/agentTypes.ts +15 -2
- package/src/tools/applypatch/applyPatch.js +362 -0
- package/src/tools/applypatch/applyPatch.ts +428 -0
- package/src/tools/bash/bash.js +22 -15
- package/src/tools/bash/bash.ts +23 -17
- package/src/tools/capabilities.js +10 -5
- package/src/tools/capabilities.ts +10 -5
- package/src/tools/computeruse/computerIntent.js +32 -0
- package/src/tools/computeruse/computerIntent.ts +46 -0
- package/src/tools/computeruse/telemetryLog.js +12 -0
- package/src/tools/computeruse/telemetryLog.ts +16 -0
- package/src/tools/computeruse/vision/telemetry.js +1 -0
- package/src/tools/computeruse/vision/telemetry.ts +1 -0
- package/src/tools/effectiveToolSet.js +4 -0
- package/src/tools/effectiveToolSet.ts +4 -0
- package/src/tools/index.js +21 -2
- package/src/tools/index.ts +33 -2
- package/src/tools/presentresult/presentresult.js +6 -15
- package/src/tools/presentresult/presentresult.ts +6 -15
- package/src/tools/read/read.js +62 -9
- package/src/tools/read/read.ts +63 -9
- package/src/tools/shared/fileUtils.js +1 -2
- package/src/tools/shared/fileUtils.ts +1 -2
- package/src/tools/shared/toolOutputStore.js +23 -1
- package/src/tools/shared/toolOutputStore.ts +22 -1
- package/src/tools/skill/skill.js +9 -3
- package/src/tools/skill/skill.ts +8 -4
- package/src/tools/toolPromptGuides.js +117 -0
- package/src/tools/toolPromptGuides.ts +137 -0
- package/src/tools/webbrowser/agentBrowserPath.js +19 -9
- package/src/tools/webbrowser/agentBrowserPath.ts +70 -28
- package/src/tools/webbrowser/browser.js +27 -11
- package/src/tools/webbrowser/browser.ts +68 -24
- package/src/tools/webbrowser/browserAutomation.js +329 -0
- package/src/tools/webbrowser/browserAutomation.ts +467 -0
- package/src/tools/webbrowser/browserAxState.js +97 -0
- package/src/tools/webbrowser/browserAxState.ts +142 -0
- package/src/tools/webbrowser/browserBatch.js +1578 -0
- package/src/tools/webbrowser/browserBatch.ts +2382 -0
- package/src/tools/webbrowser/browserCdp.js +208 -0
- package/src/tools/webbrowser/browserCdp.ts +255 -0
- package/src/tools/webbrowser/browserConditions.js +24 -0
- package/src/tools/webbrowser/browserConditions.ts +45 -0
- package/src/tools/webbrowser/browserContent.js +281 -0
- package/src/tools/webbrowser/browserContent.ts +438 -0
- package/src/tools/webbrowser/browserControl.js +131 -34
- package/src/tools/webbrowser/browserControl.ts +210 -56
- package/src/tools/webbrowser/browserDaemon.js +22 -14
- package/src/tools/webbrowser/browserDaemon.ts +54 -21
- package/src/tools/webbrowser/browserFrames.js +47 -0
- package/src/tools/webbrowser/browserFrames.ts +54 -0
- package/src/tools/webbrowser/browserInteractionResolver.js +17 -0
- package/src/tools/webbrowser/browserInteractionResolver.ts +40 -0
- package/src/tools/webbrowser/browserLocator.js +140 -0
- package/src/tools/webbrowser/browserLocator.ts +194 -0
- package/src/tools/webbrowser/browserOutcome.js +10 -17
- package/src/tools/webbrowser/browserOutcome.ts +48 -31
- package/src/tools/webbrowser/browserRecovery.js +58 -0
- package/src/tools/webbrowser/browserRecovery.ts +83 -0
- package/src/tools/webbrowser/browserRunRegistry.js +35 -3
- package/src/tools/webbrowser/browserRunRegistry.ts +91 -15
- package/src/tools/webbrowser/browserService.js +28 -2
- package/src/tools/webbrowser/browserService.ts +49 -9
- package/src/tools/webbrowser/browserStream.js +132 -33
- package/src/tools/webbrowser/browserStream.ts +224 -50
- package/src/tools/webbrowser/browserTabs.js +48 -15
- package/src/tools/webbrowser/browserTabs.ts +178 -96
- package/src/tools/webbrowser/webbrowser.js +679 -144
- package/src/tools/webbrowser/webbrowser.ts +975 -236
- package/src/tools/webfetch/preapproved.js +0 -2
- package/src/tools/webfetch/preapproved.ts +0 -2
- package/src/tools/websearch/websearch.js +129 -8
- package/src/tools/websearch/websearch.ts +138 -6
- package/src/types.ts +440 -38
- package/src/ui/App.js +377 -58
- package/src/ui/App.tsx +576 -111
- package/src/ui/LiveArea.js +14 -8
- package/src/ui/LiveArea.tsx +14 -10
- package/src/ui/MessageList.js +63 -20
- package/src/ui/MessageList.tsx +101 -16
- package/src/ui/ModelPicker.js +7 -4
- package/src/ui/ModelPicker.tsx +55 -8
- package/src/ui/hooks/useChat.js +152 -163
- package/src/ui/hooks/useChat.ts +396 -334
- package/src/ui/loginCommand.js +56 -0
- package/src/ui/loginCommand.ts +63 -0
- package/src/ui/toolProgress.js +41 -0
- package/src/ui/toolProgress.ts +44 -0
- package/src/web/browserBridge.js +20 -2
- package/src/web/browserBridge.ts +20 -2
- package/src/web/protocol.ts +1897 -194
- package/workflows/adaptive-research/workflow.mjs +86 -0
- package/workflows/e2e-capabilities/workflow.mjs +102 -0
- package/workflows/e2e-llm/workflow.mjs +52 -0
- package/workflows/e2e-smoke/workflow.mjs +59 -0
- package/workflows/e2e-vision-schema/README.md +48 -0
- package/workflows/e2e-vision-schema/workflow.mjs +149 -0
- package/workflows/e2e-write-greet/workflow.mjs +37 -0
- package/workflows/file-census/README.md +14 -0
- package/workflows/file-census/scripts/.gitkeep +2 -0
- package/workflows/file-census/workflow.mjs +110 -0
- package/workflows/goal/README.md +166 -0
- package/workflows/goal/prompts/audit.md +57 -0
- package/workflows/goal/prompts/contract.md +56 -0
- package/workflows/goal/prompts/iterate.md +37 -0
- package/workflows/goal/workflow.mjs +501 -0
- package/workflows/illustrate/README.md +90 -0
- package/workflows/illustrate/assets/engines.json +20 -0
- package/workflows/illustrate/assets/extract-baidu.js +21 -0
- package/workflows/illustrate/assets/extract-bing.js +33 -0
- package/workflows/illustrate/assets/extract-google.js +31 -0
- package/workflows/illustrate/workflow.mjs +301 -0
- package/workflows/plan/workflow.mjs +389 -0
- package/workflows/question-crops/README.md +35 -0
- package/workflows/question-crops/prompts/inventory.md +23 -0
- package/workflows/question-crops/scripts/image-ops.mjs +305 -0
- package/workflows/question-crops/scripts/image-ops.test.mjs +95 -0
- package/workflows/question-crops/workflow.mjs +172 -0
- package/workflows/ralph-loop/workflow.mjs +162 -0
- package/workflows/research/README.md +74 -0
- package/workflows/research/prompts/s0-goal.md +43 -0
- package/workflows/research/prompts/s1-early-exit.md +22 -0
- package/workflows/research/prompts/s1-map.md +58 -0
- package/workflows/research/prompts/s2-options.md +53 -0
- package/workflows/research/prompts/s3-deep-dive.md +56 -0
- package/workflows/research/prompts/s4-verify.md +74 -0
- package/workflows/research/prompts/s5-synthesis.md +81 -0
- package/workflows/research/prompts/s5b-fill-gaps.md +32 -0
- package/workflows/research/prompts/s6-package.md +32 -0
- package/workflows/research/scripts/build-report.mjs +98 -0
- package/workflows/research/workflow.mjs +288 -0
- package/workflows/wechat-reading-shots/prompts/collect-params.md +16 -0
- package/workflows/wechat-reading-shots/prompts/evaluate.md +20 -0
- package/workflows/wechat-reading-shots/prompts/open-book.md +28 -0
- package/workflows/wechat-reading-shots/prompts/read-page.md +13 -0
- package/workflows/wechat-reading-shots/prompts/turn-batch.md +24 -0
- package/workflows/wechat-reading-shots/test/control-flow.test.ts +329 -0
- package/workflows/wechat-reading-shots/workflow.mjs +266 -0
- package/workflows/wrong-question-docx/README.md +55 -0
- package/workflows/wrong-question-docx/prompts/figure-qa.md +8 -0
- package/workflows/wrong-question-docx/prompts/transcribe.md +25 -0
- package/workflows/wrong-question-docx/scripts/workbook.mjs +857 -0
- package/workflows/wrong-question-docx/test/control-flow.test.ts +165 -0
- package/workflows/wrong-question-docx/test/workbook.test.mjs +303 -0
- package/workflows/wrong-question-docx/workflow.mjs +485 -0
- package/plugins/workflow-runner/src/agentCommon.js +0 -354
- package/plugins/workflow-runner/src/agentCommon.ts +0 -435
- package/plugins/workflow-runner/src/controlFlow/branch.js +0 -40
- package/plugins/workflow-runner/src/controlFlow/branch.ts +0 -46
- package/plugins/workflow-runner/src/controlFlow/loop.js +0 -211
- package/plugins/workflow-runner/src/controlFlow/loop.ts +0 -248
- package/plugins/workflow-runner/src/controlFlow/pause.js +0 -129
- package/plugins/workflow-runner/src/controlFlow/pause.ts +0 -141
- package/plugins/workflow-runner/src/controlFlow/signal.js +0 -21
- package/plugins/workflow-runner/src/controlFlow/signal.ts +0 -29
- package/plugins/workflow-runner/src/controlFlow/types.js +0 -18
- package/plugins/workflow-runner/src/controlFlow/types.ts +0 -63
- package/plugins/workflow-runner/src/dslManifest.js +0 -72
- package/plugins/workflow-runner/src/dslManifest.ts +0 -127
- package/plugins/workflow-runner/src/expressions.js +0 -639
- package/plugins/workflow-runner/src/expressions.ts +0 -640
- package/plugins/workflow-runner/src/loader.js +0 -573
- package/plugins/workflow-runner/src/loader.ts +0 -721
- package/plugins/workflow-runner/src/multimodal.js +0 -38
- package/plugins/workflow-runner/src/multimodal.ts +0 -64
- package/plugins/workflow-runner/src/recovery.js +0 -126
- package/plugins/workflow-runner/src/recovery.ts +0 -156
- package/plugins/workflow-runner/src/stepExecutors/agent.js +0 -173
- package/plugins/workflow-runner/src/stepExecutors/agent.ts +0 -234
- package/plugins/workflow-runner/src/stepExecutors/assert.js +0 -42
- package/plugins/workflow-runner/src/stepExecutors/assert.ts +0 -52
- package/plugins/workflow-runner/src/stepExecutors/echo.js +0 -31
- package/plugins/workflow-runner/src/stepExecutors/echo.ts +0 -38
- package/plugins/workflow-runner/src/stepExecutors/llm.js +0 -99
- package/plugins/workflow-runner/src/stepExecutors/llm.ts +0 -133
- package/plugins/workflow-runner/src/stepExecutors/set.js +0 -49
- package/plugins/workflow-runner/src/stepExecutors/set.ts +0 -56
- package/plugins/workflow-runner/src/stepExecutors/skill.js +0 -111
- package/plugins/workflow-runner/src/stepExecutors/skill.ts +0 -140
- package/plugins/workflow-runner/src/stepExecutors/tool.js +0 -227
- package/plugins/workflow-runner/src/stepExecutors/tool.ts +0 -274
- package/plugins/workflow-runner/src/stepExecutors/verify.js +0 -61
- package/plugins/workflow-runner/src/stepExecutors/verify.ts +0 -72
- package/plugins/workflow-runner/src/stepPipeline.js +0 -277
- package/plugins/workflow-runner/src/stepPipeline.ts +0 -366
- package/plugins/workflow-runner/test/agent.test.ts +0 -587
- package/plugins/workflow-runner/test/agentLlmError.test.ts +0 -373
- package/plugins/workflow-runner/test/bashBlacklist.sweep.test.ts +0 -65
- package/plugins/workflow-runner/test/captureMisbind.test.ts +0 -99
- package/plugins/workflow-runner/test/cli.e2e.test.ts +0 -167
- package/plugins/workflow-runner/test/context-files.test.ts +0 -320
- package/plugins/workflow-runner/test/e2e.test.ts +0 -370
- package/plugins/workflow-runner/test/e2eNewLayout.test.ts +0 -314
- package/plugins/workflow-runner/test/engineRobustness.test.ts +0 -234
- package/plugins/workflow-runner/test/expressions.test.ts +0 -439
- package/plugins/workflow-runner/test/foundation.test.ts +0 -126
- package/plugins/workflow-runner/test/graceful.test.ts +0 -139
- package/plugins/workflow-runner/test/greedyInputs.load.test.ts +0 -47
- package/plugins/workflow-runner/test/hostCapability.test.ts +0 -108
- package/plugins/workflow-runner/test/illustrate.load.test.ts +0 -183
- package/plugins/workflow-runner/test/jsonPayload.test.ts +0 -83
- package/plugins/workflow-runner/test/loader.test.ts +0 -617
- package/plugins/workflow-runner/test/maxTurnsRetryDelay.test.ts +0 -33
- package/plugins/workflow-runner/test/multimodal.test.ts +0 -342
- package/plugins/workflow-runner/test/output-schema.test.ts +0 -213
- package/plugins/workflow-runner/test/parseWorkflowArgs.test.ts +0 -106
- package/plugins/workflow-runner/test/pluginRunner.isolation.test.ts +0 -269
- package/plugins/workflow-runner/test/ppt.load.test.ts +0 -1220
- package/plugins/workflow-runner/test/ralphLoop.e2e.test.ts +0 -221
- package/plugins/workflow-runner/test/recovery.test.ts +0 -265
- package/plugins/workflow-runner/test/research.load.test.ts +0 -328
- package/plugins/workflow-runner/test/runner.test.ts +0 -911
- package/plugins/workflow-runner/test/runnerPreflight.test.ts +0 -436
- package/plugins/workflow-runner/test/skillTurnDone.test.ts +0 -71
- package/plugins/workflow-runner/test/stateLifecycle.test.ts +0 -198
- package/plugins/workflow-runner/test/stepPipeline.test.ts +0 -114
- package/plugins/workflow-runner/test/taskArbitration.test.ts +0 -354
- package/plugins/workflow-runner/test/taskLifecycle.test.ts +0 -324
- package/plugins/workflow-runner/test/toolArtifactOverwrite.test.ts +0 -221
- package/plugins/workflow-runner/test/toolCapabilityBoundary.test.ts +0 -29
- package/plugins/workflow-runner/test/toolRuntimeIntegration.test.ts +0 -168
- package/plugins/workflow-runner/test/webppt.load.test.ts +0 -250
- package/plugins/workflow-runner/test/word.load.test.ts +0 -208
- package/plugins/workflow-runner/test/workflows.smoke.test.ts +0 -76
- package/plugins/workflow-runner/test/yamlGuardConditions.test.ts +0 -207
- package/src/advisorState.js +0 -22
- package/src/cli/configWizard.js +0 -293
- package/src/tools/toolProgressBridge.js +0 -48
- package/workflows/adaptive-research.yaml +0 -50
- package/workflows/book-review-short.yaml +0 -125
- package/workflows/e2e-write-greet.yaml +0 -31
- package/workflows/illustrate.yaml +0 -227
- package/workflows/plan.yaml +0 -470
- package/workflows/ppt.yaml +0 -1670
- package/workflows/ralph-loop.yaml +0 -201
- package/workflows/research.yaml +0 -873
- package/workflows/schema.json +0 -176
- package/workflows/webppt.yaml +0 -496
- package/workflows/wechat-reading-shots.yaml +0 -315
- package/workflows/word.yaml +0 -720
- package/workflows/youtube-shorts.yaml +0 -202
- /package/workflows/{research.md → research/DESIGN.md} +0 -0
|
@@ -0,0 +1,657 @@
|
|
|
1
|
+
import crypto from 'node:crypto';
|
|
2
|
+
|
|
3
|
+
import type {
|
|
4
|
+
AdvisorCheckpoint,
|
|
5
|
+
AdvisorCheckpointMetadata,
|
|
6
|
+
AdvisorConsultationMetadata,
|
|
7
|
+
AdvisorContextPacket,
|
|
8
|
+
AdvisorDecisionMetadata,
|
|
9
|
+
AdvisorRuntimeBridge,
|
|
10
|
+
AdvisorVersions,
|
|
11
|
+
Message,
|
|
12
|
+
ToolCall,
|
|
13
|
+
ToolResult,
|
|
14
|
+
ToolResultMetadata,
|
|
15
|
+
} from '../types.js';
|
|
16
|
+
|
|
17
|
+
/**
|
|
18
|
+
* 同一个检查点最多连续暂缓几次,超过就降级放行并留痕。
|
|
19
|
+
*
|
|
20
|
+
* 取 3 而不是 1:一次暂缓 + 一次咨询 + 一次记录决策是正常配合的节奏,
|
|
21
|
+
* 留出余量才不会误伤按流程走、只是多绕了一步的模型;取有限值而不是无限,
|
|
22
|
+
* 是因为唯一的替代上界是 maxTurns(200 轮),那不叫闸门叫卡死。
|
|
23
|
+
*/
|
|
24
|
+
const MAX_CONSECUTIVE_DEFERRALS = 3;
|
|
25
|
+
const MAX_FREE_RECONCILIATIONS_PER_CHECKPOINT = 1;
|
|
26
|
+
|
|
27
|
+
const WRITE_TOOLS = new Set(['Edit', 'MultiEdit', 'Write', 'ApplyPatch', 'NotebookEdit']);
|
|
28
|
+
const INVESTIGATION_TOOLS = new Set([
|
|
29
|
+
'Read',
|
|
30
|
+
'Glob',
|
|
31
|
+
'Grep',
|
|
32
|
+
'LSP',
|
|
33
|
+
'WebSearch',
|
|
34
|
+
'WebFetch',
|
|
35
|
+
'WebBrowser',
|
|
36
|
+
'TerminalCapture',
|
|
37
|
+
'RecallMemory',
|
|
38
|
+
'SearchConversations',
|
|
39
|
+
]);
|
|
40
|
+
|
|
41
|
+
function textContent(message: Message | undefined): string {
|
|
42
|
+
if (!message) return '';
|
|
43
|
+
if (typeof message.content === 'string') return message.content;
|
|
44
|
+
if (Array.isArray(message.content)) {
|
|
45
|
+
return message.content
|
|
46
|
+
.filter((part) => part.type === 'text')
|
|
47
|
+
.map((part) => part.text)
|
|
48
|
+
.join('\n');
|
|
49
|
+
}
|
|
50
|
+
return '';
|
|
51
|
+
}
|
|
52
|
+
|
|
53
|
+
function toolNamesByCallId(history: readonly Message[]): Map<string, string> {
|
|
54
|
+
const names = new Map<string, string>();
|
|
55
|
+
for (const message of history) {
|
|
56
|
+
if (message.role !== 'assistant') continue;
|
|
57
|
+
for (const call of message.tool_calls ?? [])
|
|
58
|
+
names.set(call.id, call.function.name);
|
|
59
|
+
}
|
|
60
|
+
return names;
|
|
61
|
+
}
|
|
62
|
+
|
|
63
|
+
function advisorMetadata(
|
|
64
|
+
metadata: ToolResultMetadata | undefined,
|
|
65
|
+
):
|
|
66
|
+
| AdvisorCheckpointMetadata
|
|
67
|
+
| AdvisorConsultationMetadata
|
|
68
|
+
| AdvisorDecisionMetadata
|
|
69
|
+
| undefined {
|
|
70
|
+
if (
|
|
71
|
+
metadata?.kind === 'advisor_checkpoint' ||
|
|
72
|
+
metadata?.kind === 'advisor_consultation' ||
|
|
73
|
+
metadata?.kind === 'advisor_decision'
|
|
74
|
+
)
|
|
75
|
+
return metadata;
|
|
76
|
+
return undefined;
|
|
77
|
+
}
|
|
78
|
+
|
|
79
|
+
function versionsOf(
|
|
80
|
+
metadata:
|
|
81
|
+
| AdvisorCheckpointMetadata
|
|
82
|
+
| AdvisorConsultationMetadata
|
|
83
|
+
| AdvisorDecisionMetadata,
|
|
84
|
+
): AdvisorVersions {
|
|
85
|
+
return {
|
|
86
|
+
taskVersion: metadata.taskVersion,
|
|
87
|
+
workVersion: metadata.workVersion,
|
|
88
|
+
evidenceVersion: metadata.evidenceVersion,
|
|
89
|
+
};
|
|
90
|
+
}
|
|
91
|
+
|
|
92
|
+
function checkpointKey(
|
|
93
|
+
checkpoint: AdvisorCheckpoint,
|
|
94
|
+
versions: AdvisorVersions,
|
|
95
|
+
): string {
|
|
96
|
+
return `${versions.taskVersion}:${checkpoint}:${versions.workVersion}:${versions.evidenceVersion}`;
|
|
97
|
+
}
|
|
98
|
+
|
|
99
|
+
export interface AdvisorToolSemantics {
|
|
100
|
+
mutatesWorkspace: boolean;
|
|
101
|
+
highRisk: boolean;
|
|
102
|
+
riskReason?: string;
|
|
103
|
+
}
|
|
104
|
+
|
|
105
|
+
export interface CheckpointDecision {
|
|
106
|
+
allow: boolean;
|
|
107
|
+
deferred?: ToolResult;
|
|
108
|
+
}
|
|
109
|
+
|
|
110
|
+
/**
|
|
111
|
+
* Run-local controller whose durable source of truth is Advisor metadata already
|
|
112
|
+
* stored on tool messages. It owns checkpoint de-duplication and versioning;
|
|
113
|
+
* provider calls remain in tools/advisor/advisor.ts.
|
|
114
|
+
*/
|
|
115
|
+
export class AdvisorCoordinator implements AdvisorRuntimeBridge {
|
|
116
|
+
readonly #history: readonly Message[];
|
|
117
|
+
readonly #currentRequest: string;
|
|
118
|
+
#advisorAvailable: boolean;
|
|
119
|
+
readonly #completed = new Set<string>();
|
|
120
|
+
readonly #deferredOnce = new Set<string>();
|
|
121
|
+
readonly #deferCounts = new Map<string, number>();
|
|
122
|
+
readonly #successfulConsultationsByKey = new Map<string, number>();
|
|
123
|
+
readonly #softAlerts = new Set<string>();
|
|
124
|
+
#taskVersion: number;
|
|
125
|
+
#taskCommitted = false;
|
|
126
|
+
#workVersion = 0;
|
|
127
|
+
#evidenceVersion = 0;
|
|
128
|
+
#investigationEvidence = 0;
|
|
129
|
+
#mutations = 0;
|
|
130
|
+
#verifications = 0;
|
|
131
|
+
#deliveryProduced = false;
|
|
132
|
+
#required?: { checkpoint: AdvisorCheckpoint; key: string; reason?: string };
|
|
133
|
+
#softCheckpoint?: { checkpoint: AdvisorCheckpoint; key: string };
|
|
134
|
+
#pending?: AdvisorConsultationMetadata;
|
|
135
|
+
#lastDecision?: AdvisorDecisionMetadata;
|
|
136
|
+
#reminders: string[] = [];
|
|
137
|
+
|
|
138
|
+
constructor(input: {
|
|
139
|
+
history: readonly Message[];
|
|
140
|
+
currentRequest: string;
|
|
141
|
+
advisorAvailable: boolean;
|
|
142
|
+
/** 当前用户消息;用于区分同一请求恢复与新请求,避免旧 pending 污染新回合。 */
|
|
143
|
+
currentRequestMessageId?: string;
|
|
144
|
+
}) {
|
|
145
|
+
this.#history = input.history;
|
|
146
|
+
this.#currentRequest = input.currentRequest;
|
|
147
|
+
this.#advisorAvailable = input.advisorAvailable;
|
|
148
|
+
const names = toolNamesByCallId(input.history);
|
|
149
|
+
let maxTaskVersion = 0;
|
|
150
|
+
let maxWorkVersion = 0;
|
|
151
|
+
let maxEvidenceVersion = 0;
|
|
152
|
+
const decisions = new Set<string>();
|
|
153
|
+
const successfulConsultations: Array<{
|
|
154
|
+
index: number;
|
|
155
|
+
metadata: AdvisorConsultationMetadata;
|
|
156
|
+
}> = [];
|
|
157
|
+
const currentRequestIndex = input.currentRequestMessageId
|
|
158
|
+
? input.history.findIndex(
|
|
159
|
+
(message) =>
|
|
160
|
+
message.role === 'user' && message.id === input.currentRequestMessageId,
|
|
161
|
+
)
|
|
162
|
+
: -1;
|
|
163
|
+
|
|
164
|
+
for (let index = 0; index < input.history.length; index += 1) {
|
|
165
|
+
const message = input.history[index]!;
|
|
166
|
+
if (message.role !== 'tool') continue;
|
|
167
|
+
const metadata = advisorMetadata(message.display_metadata);
|
|
168
|
+
if (metadata) {
|
|
169
|
+
maxTaskVersion = Math.max(maxTaskVersion, metadata.taskVersion);
|
|
170
|
+
maxWorkVersion = Math.max(maxWorkVersion, metadata.workVersion);
|
|
171
|
+
maxEvidenceVersion = Math.max(
|
|
172
|
+
maxEvidenceVersion,
|
|
173
|
+
metadata.evidenceVersion,
|
|
174
|
+
);
|
|
175
|
+
if (metadata.kind === 'advisor_decision') {
|
|
176
|
+
decisions.add(metadata.consultationId);
|
|
177
|
+
this.#completed.add(metadata.checkpointKey);
|
|
178
|
+
this.#lastDecision = metadata;
|
|
179
|
+
} else if (metadata.kind === 'advisor_consultation') {
|
|
180
|
+
if (metadata.status === 'failed')
|
|
181
|
+
this.#completed.add(metadata.checkpointKey);
|
|
182
|
+
else {
|
|
183
|
+
this.#successfulConsultationsByKey.set(
|
|
184
|
+
metadata.checkpointKey,
|
|
185
|
+
(this.#successfulConsultationsByKey.get(metadata.checkpointKey) ?? 0) + 1,
|
|
186
|
+
);
|
|
187
|
+
successfulConsultations.push({ index, metadata });
|
|
188
|
+
}
|
|
189
|
+
}
|
|
190
|
+
}
|
|
191
|
+
|
|
192
|
+
const toolName = names.get(message.tool_call_id);
|
|
193
|
+
const display = message.display_metadata;
|
|
194
|
+
if (display?.kind === 'file_activity') {
|
|
195
|
+
const mutationCount = display.files.filter((file) =>
|
|
196
|
+
['created', 'modified', 'generated', 'deleted'].includes(
|
|
197
|
+
file.activity,
|
|
198
|
+
),
|
|
199
|
+
).length;
|
|
200
|
+
this.#workVersion += mutationCount;
|
|
201
|
+
}
|
|
202
|
+
const failed = message.content.startsWith('Error: ');
|
|
203
|
+
if (
|
|
204
|
+
toolName &&
|
|
205
|
+
toolName !== 'Advisor' &&
|
|
206
|
+
toolName !== 'AdvisorDecision' &&
|
|
207
|
+
display?.kind !== 'advisor_checkpoint' &&
|
|
208
|
+
(!failed || display?.kind === 'process_execution')
|
|
209
|
+
) {
|
|
210
|
+
this.#evidenceVersion += 1;
|
|
211
|
+
}
|
|
212
|
+
}
|
|
213
|
+
|
|
214
|
+
this.#workVersion = Math.max(this.#workVersion, maxWorkVersion);
|
|
215
|
+
this.#evidenceVersion = Math.max(this.#evidenceVersion, maxEvidenceVersion);
|
|
216
|
+
this.#taskVersion = maxTaskVersion + 1;
|
|
217
|
+
this.#pending = successfulConsultations
|
|
218
|
+
.filter(({ index }) =>
|
|
219
|
+
input.currentRequestMessageId === undefined
|
|
220
|
+
? true
|
|
221
|
+
: currentRequestIndex >= 0 && index > currentRequestIndex,
|
|
222
|
+
)
|
|
223
|
+
.reverse()
|
|
224
|
+
.map(({ metadata }) => metadata)
|
|
225
|
+
.find((consultation) => !decisions.has(consultation.consultationId));
|
|
226
|
+
}
|
|
227
|
+
|
|
228
|
+
get versions(): AdvisorVersions {
|
|
229
|
+
return {
|
|
230
|
+
taskVersion: this.#taskVersion,
|
|
231
|
+
workVersion: this.#workVersion,
|
|
232
|
+
evidenceVersion: this.#evidenceVersion,
|
|
233
|
+
};
|
|
234
|
+
}
|
|
235
|
+
|
|
236
|
+
get pendingConsultationId(): string | undefined {
|
|
237
|
+
return this.#pending?.consultationId;
|
|
238
|
+
}
|
|
239
|
+
|
|
240
|
+
get canReconcileConsultation(): boolean {
|
|
241
|
+
if (!this.#pending) return false;
|
|
242
|
+
const successful = this.#successfulConsultationsByKey.get(
|
|
243
|
+
this.#pending.checkpointKey,
|
|
244
|
+
) ?? 0;
|
|
245
|
+
return successful <= MAX_FREE_RECONCILIATIONS_PER_CHECKPOINT;
|
|
246
|
+
}
|
|
247
|
+
|
|
248
|
+
setAvailable(available: boolean): void {
|
|
249
|
+
this.#advisorAvailable = available;
|
|
250
|
+
}
|
|
251
|
+
|
|
252
|
+
usageLine(used: number, max: number): string {
|
|
253
|
+
const pending = this.#pending
|
|
254
|
+
? `;顾问意见 #${this.#pending.consultationId.slice(-8)} 等待主 Agent 记录采纳决定`
|
|
255
|
+
: '';
|
|
256
|
+
return used === 0
|
|
257
|
+
? `Advisor 动态:本回合尚未咨询顾问(可用 ${max} 次)${pending}。`
|
|
258
|
+
: `Advisor 动态:本回合已咨询 ${used}/${max} 次${pending}。`;
|
|
259
|
+
}
|
|
260
|
+
|
|
261
|
+
transientReminder(): string {
|
|
262
|
+
return this.#reminders.join('\n\n');
|
|
263
|
+
}
|
|
264
|
+
|
|
265
|
+
markTransientDelivered(): void {
|
|
266
|
+
this.#reminders = [];
|
|
267
|
+
}
|
|
268
|
+
|
|
269
|
+
beforeTool(
|
|
270
|
+
call: ToolCall,
|
|
271
|
+
semantics: AdvisorToolSemantics,
|
|
272
|
+
): CheckpointDecision {
|
|
273
|
+
if (
|
|
274
|
+
!this.#advisorAvailable ||
|
|
275
|
+
call.function.name === 'Advisor' ||
|
|
276
|
+
call.function.name === 'AdvisorDecision'
|
|
277
|
+
) {
|
|
278
|
+
return { allow: true };
|
|
279
|
+
}
|
|
280
|
+
if (this.#pending && (semantics.mutatesWorkspace || semantics.highRisk)) {
|
|
281
|
+
return this.#deferPending(
|
|
282
|
+
'顾问意见已返回,但尚未调用 AdvisorDecision 记录主 Agent 的采纳判断。',
|
|
283
|
+
);
|
|
284
|
+
}
|
|
285
|
+
if (semantics.highRisk) {
|
|
286
|
+
return this.requestCheckpoint(
|
|
287
|
+
'high-risk',
|
|
288
|
+
semantics.riskReason ?? '即将执行不可逆操作',
|
|
289
|
+
);
|
|
290
|
+
}
|
|
291
|
+
if (semantics.mutatesWorkspace) {
|
|
292
|
+
return this.requestCheckpoint(
|
|
293
|
+
'pre-execution',
|
|
294
|
+
'即将进行本任务版本的首次实质写入',
|
|
295
|
+
);
|
|
296
|
+
}
|
|
297
|
+
return { allow: true };
|
|
298
|
+
}
|
|
299
|
+
|
|
300
|
+
requestCheckpoint(
|
|
301
|
+
checkpoint: AdvisorCheckpoint,
|
|
302
|
+
reason: string,
|
|
303
|
+
options: { oneShot?: boolean } = {},
|
|
304
|
+
): CheckpointDecision {
|
|
305
|
+
if (!this.#advisorAvailable) return { allow: true };
|
|
306
|
+
this.#commitTask();
|
|
307
|
+
const key = checkpointKey(checkpoint, this.versions);
|
|
308
|
+
if (this.#completed.has(key)) return { allow: true };
|
|
309
|
+
if (options.oneShot && this.#deferredOnce.has(key)) {
|
|
310
|
+
this.#required = undefined;
|
|
311
|
+
this.#completed.add(key);
|
|
312
|
+
return { allow: true };
|
|
313
|
+
}
|
|
314
|
+
|
|
315
|
+
// ── 死锁上界:模型不配合时也必须让路 ──────────────────────────────
|
|
316
|
+
// 顾问**失败**这条路已有降级(finishConsultation 会 #completed.add),
|
|
317
|
+
// 但模型**不配合**(一直不调 Advisor/AdvisorDecision、只管重试写操作)此前无人管:
|
|
318
|
+
// high-risk 与 pre-execution 都不传 oneShot,同一个 key 会被反复 defer,
|
|
319
|
+
// 唯一的界是 maxTurns —— 而它刚从 50 涨到 200,用户看到的就是「agent 卡住不干活」。
|
|
320
|
+
// 本模块自己的原则是「不会死锁」,这里把那条原则补全到第二条路上。
|
|
321
|
+
const seen = (this.#deferCounts.get(key) ?? 0) + 1;
|
|
322
|
+
this.#deferCounts.set(key, seen);
|
|
323
|
+
if (seen > MAX_CONSECUTIVE_DEFERRALS) {
|
|
324
|
+
this.#required = undefined;
|
|
325
|
+
this.#completed.add(key);
|
|
326
|
+
// 绕过了本该经过的检查点就必须留痕:静默放行会让模型以为自己合规了。
|
|
327
|
+
const notice = `Advisor 检查点 ${checkpoint} 已连续暂缓 ${MAX_CONSECUTIVE_DEFERRALS} 次仍未完成咨询,本次降级放行。这一步没有第二意见背书,请自行加倍核实,并在汇报时如实说明。`;
|
|
328
|
+
if (!this.#reminders.includes(notice)) this.#reminders.push(notice);
|
|
329
|
+
return { allow: true };
|
|
330
|
+
}
|
|
331
|
+
|
|
332
|
+
this.#required = { checkpoint, key, reason };
|
|
333
|
+
this.#deferredOnce.add(key);
|
|
334
|
+
return this.#defer(checkpoint, reason, key);
|
|
335
|
+
}
|
|
336
|
+
|
|
337
|
+
notifyStalled(signature?: string): void {
|
|
338
|
+
if (!this.#advisorAvailable) return;
|
|
339
|
+
this.#commitTask();
|
|
340
|
+
const key = `${checkpointKey('stalled', this.versions)}:${signature ?? 'unknown'}`;
|
|
341
|
+
if (this.#softAlerts.has(key)) return;
|
|
342
|
+
this.#softAlerts.add(key);
|
|
343
|
+
this.#softCheckpoint = {
|
|
344
|
+
checkpoint: 'stalled',
|
|
345
|
+
key: checkpointKey('stalled', this.versions),
|
|
346
|
+
};
|
|
347
|
+
this.#reminders.push(
|
|
348
|
+
'检测到连续同类失败或执行没有进展。继续重试前请优先调用 Advisor 获取第二意见;这是提醒,不阻止只读诊断或切换路径。',
|
|
349
|
+
);
|
|
350
|
+
}
|
|
351
|
+
|
|
352
|
+
shouldCheckDelivery(): boolean {
|
|
353
|
+
return (
|
|
354
|
+
this.#mutations > 0 ||
|
|
355
|
+
this.#verifications > 0 ||
|
|
356
|
+
this.#deliveryProduced ||
|
|
357
|
+
this.#investigationEvidence >= 3
|
|
358
|
+
);
|
|
359
|
+
}
|
|
360
|
+
|
|
361
|
+
beforeDelivery(): CheckpointDecision {
|
|
362
|
+
if (this.#pending) {
|
|
363
|
+
const decision = this.#deferPending(
|
|
364
|
+
'提交最终结果前必须先用 AdvisorDecision 记录对当前顾问意见的判断。',
|
|
365
|
+
);
|
|
366
|
+
if (
|
|
367
|
+
decision.deferred &&
|
|
368
|
+
!decision.deferred.ok &&
|
|
369
|
+
!this.#reminders.includes(decision.deferred.error)
|
|
370
|
+
) {
|
|
371
|
+
this.#reminders.push(decision.deferred.error);
|
|
372
|
+
}
|
|
373
|
+
return decision;
|
|
374
|
+
}
|
|
375
|
+
if (!this.shouldCheckDelivery()) return { allow: true };
|
|
376
|
+
const decision = this.requestCheckpoint(
|
|
377
|
+
'pre-delivery',
|
|
378
|
+
'当前回合已产生实质成果或证据,提交最终结论前进行一次第二意见复核',
|
|
379
|
+
{ oneShot: true },
|
|
380
|
+
);
|
|
381
|
+
if (
|
|
382
|
+
decision.deferred &&
|
|
383
|
+
!decision.deferred.ok &&
|
|
384
|
+
!this.#reminders.includes(decision.deferred.error)
|
|
385
|
+
) {
|
|
386
|
+
this.#reminders.push(decision.deferred.error);
|
|
387
|
+
}
|
|
388
|
+
return decision;
|
|
389
|
+
}
|
|
390
|
+
|
|
391
|
+
beginConsultation(): AdvisorConsultationMetadata {
|
|
392
|
+
this.#commitTask();
|
|
393
|
+
// #pending 存在 = 有一个检查点已经问过、还没被 AdvisorDecision 结清。
|
|
394
|
+
// 这种情况下的任何新咨询(模型主动复核、或 notifyStalled 又冒出一个软
|
|
395
|
+
// 检查点)都必须沿用它的身份,不能落到 #required/#softCheckpoint/默认
|
|
396
|
+
// pre-execution ——否则原始检查点的 checkpointKey 永远进不了 #completed,
|
|
397
|
+
// 重试会被误判成"没咨询过"再次拦下,而模型明明已经照工具说明书的要求
|
|
398
|
+
// 做了复核。这与 beforeTool 顶部"#pending 存在就先拦"的语义保持一致:
|
|
399
|
+
// 既然还有检查点没走完决策,新咨询理应服务于把它结清,而不是凭空开新身份。
|
|
400
|
+
const active = this.#pending
|
|
401
|
+
? { checkpoint: this.#pending.checkpoint, key: this.#pending.checkpointKey }
|
|
402
|
+
: this.#required ??
|
|
403
|
+
this.#softCheckpoint ?? {
|
|
404
|
+
checkpoint: 'pre-execution' as const,
|
|
405
|
+
key: checkpointKey('pre-execution', this.versions),
|
|
406
|
+
};
|
|
407
|
+
return {
|
|
408
|
+
kind: 'advisor_consultation',
|
|
409
|
+
consultationId: crypto.randomUUID(),
|
|
410
|
+
checkpoint: active.checkpoint,
|
|
411
|
+
checkpointKey: active.key,
|
|
412
|
+
status: 'completed',
|
|
413
|
+
...this.versions,
|
|
414
|
+
};
|
|
415
|
+
}
|
|
416
|
+
|
|
417
|
+
finishConsultation(
|
|
418
|
+
metadata: AdvisorConsultationMetadata,
|
|
419
|
+
ok: boolean,
|
|
420
|
+
): AdvisorConsultationMetadata {
|
|
421
|
+
const finished = {
|
|
422
|
+
...metadata,
|
|
423
|
+
status: ok ? ('completed' as const) : ('failed' as const),
|
|
424
|
+
};
|
|
425
|
+
if (ok) {
|
|
426
|
+
this.#pending = finished;
|
|
427
|
+
this.#successfulConsultationsByKey.set(
|
|
428
|
+
finished.checkpointKey,
|
|
429
|
+
(this.#successfulConsultationsByKey.get(finished.checkpointKey) ?? 0) + 1,
|
|
430
|
+
);
|
|
431
|
+
} else {
|
|
432
|
+
this.#completed.add(finished.checkpointKey);
|
|
433
|
+
if (this.#pending?.checkpointKey === finished.checkpointKey)
|
|
434
|
+
this.#pending = undefined;
|
|
435
|
+
}
|
|
436
|
+
this.#required = undefined;
|
|
437
|
+
this.#softCheckpoint = undefined;
|
|
438
|
+
return finished;
|
|
439
|
+
}
|
|
440
|
+
|
|
441
|
+
recordDecision(input: {
|
|
442
|
+
consultationId: string;
|
|
443
|
+
decision: 'accept' | 'partial' | 'reject';
|
|
444
|
+
rationale: string;
|
|
445
|
+
acceptedPoints?: string[];
|
|
446
|
+
rejectedPoints?: string[];
|
|
447
|
+
verificationPlan?: string[];
|
|
448
|
+
}): ToolResult {
|
|
449
|
+
if (
|
|
450
|
+
!this.#pending ||
|
|
451
|
+
this.#pending.consultationId !== input.consultationId
|
|
452
|
+
) {
|
|
453
|
+
const expected = this.#pending?.consultationId;
|
|
454
|
+
return {
|
|
455
|
+
ok: false,
|
|
456
|
+
error: expected
|
|
457
|
+
? `顾问咨询 ID 不匹配:收到 ${input.consultationId},当前待处理的完整 consultationId 是 ${expected}。请原样使用该完整 ID 重试一次 AdvisorDecision。`
|
|
458
|
+
: `没有找到待处理的顾问咨询 ${input.consultationId};不要继续重试 AdvisorDecision。`,
|
|
459
|
+
recovery: expected ? 'change-input' : 'none',
|
|
460
|
+
};
|
|
461
|
+
}
|
|
462
|
+
const metadata: AdvisorDecisionMetadata = {
|
|
463
|
+
kind: 'advisor_decision',
|
|
464
|
+
consultationId: input.consultationId,
|
|
465
|
+
checkpoint: this.#pending.checkpoint,
|
|
466
|
+
checkpointKey: this.#pending.checkpointKey,
|
|
467
|
+
decision: input.decision,
|
|
468
|
+
rationale: input.rationale,
|
|
469
|
+
acceptedPoints: input.acceptedPoints ?? [],
|
|
470
|
+
rejectedPoints: input.rejectedPoints ?? [],
|
|
471
|
+
verificationPlan: input.verificationPlan ?? [],
|
|
472
|
+
...versionsOf(this.#pending),
|
|
473
|
+
};
|
|
474
|
+
this.#completed.add(metadata.checkpointKey);
|
|
475
|
+
this.#lastDecision = metadata;
|
|
476
|
+
this.#pending = undefined;
|
|
477
|
+
return {
|
|
478
|
+
ok: true,
|
|
479
|
+
content: `已记录顾问意见处理决定:${input.decision}。理由:${input.rationale}`,
|
|
480
|
+
metadata,
|
|
481
|
+
};
|
|
482
|
+
}
|
|
483
|
+
|
|
484
|
+
observeToolResult(call: ToolCall, result: ToolResult): void {
|
|
485
|
+
if (!result.ok && result.disposition === 'deferred') return;
|
|
486
|
+
if (result.metadata?.kind === 'advisor_consultation') {
|
|
487
|
+
if (result.metadata.status === 'completed')
|
|
488
|
+
this.#pending = result.metadata;
|
|
489
|
+
else {
|
|
490
|
+
this.#completed.add(result.metadata.checkpointKey);
|
|
491
|
+
if (this.#pending?.checkpointKey === result.metadata.checkpointKey)
|
|
492
|
+
this.#pending = undefined;
|
|
493
|
+
}
|
|
494
|
+
return;
|
|
495
|
+
}
|
|
496
|
+
if (result.metadata?.kind === 'advisor_decision') {
|
|
497
|
+
this.#completed.add(result.metadata.checkpointKey);
|
|
498
|
+
this.#lastDecision = result.metadata;
|
|
499
|
+
this.#pending = undefined;
|
|
500
|
+
return;
|
|
501
|
+
}
|
|
502
|
+
const metadata = result.metadata;
|
|
503
|
+
const verificationResult = Boolean(
|
|
504
|
+
metadata?.kind === 'process_execution' &&
|
|
505
|
+
metadata.commandKind !== 'other' &&
|
|
506
|
+
['completed', 'failed'].includes(metadata.status),
|
|
507
|
+
);
|
|
508
|
+
if (verificationResult) {
|
|
509
|
+
this.#verifications += 1;
|
|
510
|
+
this.#evidenceVersion += 1;
|
|
511
|
+
if (!result.ok) return;
|
|
512
|
+
}
|
|
513
|
+
if (!result.ok) return;
|
|
514
|
+
|
|
515
|
+
if (metadata?.kind === 'file_activity') {
|
|
516
|
+
const mutations = metadata.files.filter((file) =>
|
|
517
|
+
['created', 'modified', 'generated', 'deleted'].includes(file.activity),
|
|
518
|
+
).length;
|
|
519
|
+
this.#mutations += mutations;
|
|
520
|
+
this.#workVersion += mutations;
|
|
521
|
+
}
|
|
522
|
+
if (metadata?.kind === 'result_manifest') this.#deliveryProduced = true;
|
|
523
|
+
if (INVESTIGATION_TOOLS.has(call.function.name))
|
|
524
|
+
this.#investigationEvidence += 1;
|
|
525
|
+
if (
|
|
526
|
+
call.function.name !== 'Advisor' &&
|
|
527
|
+
call.function.name !== 'AdvisorDecision' &&
|
|
528
|
+
!verificationResult
|
|
529
|
+
) {
|
|
530
|
+
this.#evidenceVersion += 1;
|
|
531
|
+
}
|
|
532
|
+
}
|
|
533
|
+
|
|
534
|
+
contextPacket(note?: string): AdvisorContextPacket {
|
|
535
|
+
const names = toolNamesByCallId(this.#history);
|
|
536
|
+
const changedFiles = new Set<string>();
|
|
537
|
+
const evidence: string[] = [];
|
|
538
|
+
let previousAdvice: string | undefined;
|
|
539
|
+
let lastAdvisorIndex = -1;
|
|
540
|
+
|
|
541
|
+
for (let index = 0; index < this.#history.length; index += 1) {
|
|
542
|
+
const message = this.#history[index]!;
|
|
543
|
+
if (message.role !== 'tool') continue;
|
|
544
|
+
const toolName = names.get(message.tool_call_id) ?? 'Tool';
|
|
545
|
+
if (message.display_metadata?.kind === 'file_activity') {
|
|
546
|
+
for (const file of message.display_metadata.files)
|
|
547
|
+
changedFiles.add(file.relativePath);
|
|
548
|
+
}
|
|
549
|
+
if (message.display_metadata?.kind === 'process_execution') {
|
|
550
|
+
evidence.push(
|
|
551
|
+
`${toolName}: ${message.display_metadata.commandKind}/${message.display_metadata.status}`,
|
|
552
|
+
);
|
|
553
|
+
} else if (message.display_metadata?.kind === 'result_manifest') {
|
|
554
|
+
evidence.push(
|
|
555
|
+
`${toolName}: 已生成 ${message.display_metadata.manifest.files.length} 个交付文件`,
|
|
556
|
+
);
|
|
557
|
+
} else if (
|
|
558
|
+
toolName !== 'Advisor' &&
|
|
559
|
+
toolName !== 'AdvisorDecision' &&
|
|
560
|
+
evidence.length < 12
|
|
561
|
+
) {
|
|
562
|
+
evidence.push(`${toolName}: ${message.content.slice(0, 240)}`);
|
|
563
|
+
}
|
|
564
|
+
if (
|
|
565
|
+
toolName === 'Advisor' &&
|
|
566
|
+
message.display_metadata?.kind === 'advisor_consultation' &&
|
|
567
|
+
message.display_metadata.status === 'completed'
|
|
568
|
+
) {
|
|
569
|
+
previousAdvice = message.content.slice(0, 4_000);
|
|
570
|
+
lastAdvisorIndex = index;
|
|
571
|
+
}
|
|
572
|
+
}
|
|
573
|
+
const actionsAfterPreviousAdvice =
|
|
574
|
+
lastAdvisorIndex < 0
|
|
575
|
+
? []
|
|
576
|
+
: this.#history
|
|
577
|
+
.slice(lastAdvisorIndex + 1)
|
|
578
|
+
.filter(
|
|
579
|
+
(message): message is Extract<Message, { role: 'tool' }> =>
|
|
580
|
+
message.role === 'tool',
|
|
581
|
+
)
|
|
582
|
+
.map((message) => names.get(message.tool_call_id) ?? 'Tool')
|
|
583
|
+
.filter((name) => name !== 'AdvisorDecision');
|
|
584
|
+
const firstUser = this.#history.find((message) => message.role === 'user');
|
|
585
|
+
// 与 beginConsultation() 同一条修复:复核调用时顾问看到的 checkpoint 字段
|
|
586
|
+
// 不该被误标成 pre-execution ——它自己就是在评审那个还没结清的检查点。
|
|
587
|
+
const active = this.#pending
|
|
588
|
+
? { checkpoint: this.#pending.checkpoint }
|
|
589
|
+
: this.#required ?? this.#softCheckpoint;
|
|
590
|
+
return {
|
|
591
|
+
objective: (textContent(firstUser) || this.#currentRequest).slice(
|
|
592
|
+
0,
|
|
593
|
+
8_000,
|
|
594
|
+
),
|
|
595
|
+
currentRequest: this.#currentRequest.slice(0, 8_000),
|
|
596
|
+
checkpoint: active?.checkpoint ?? 'pre-execution',
|
|
597
|
+
versions: this.versions,
|
|
598
|
+
changedFiles: [...changedFiles],
|
|
599
|
+
evidence,
|
|
600
|
+
...(previousAdvice ? { previousAdvice } : {}),
|
|
601
|
+
actionsAfterPreviousAdvice,
|
|
602
|
+
openQuestions: note ? [note] : [],
|
|
603
|
+
};
|
|
604
|
+
}
|
|
605
|
+
|
|
606
|
+
#commitTask(): void {
|
|
607
|
+
if (this.#taskCommitted) return;
|
|
608
|
+
this.#taskCommitted = true;
|
|
609
|
+
}
|
|
610
|
+
|
|
611
|
+
#defer(
|
|
612
|
+
checkpoint: AdvisorCheckpoint,
|
|
613
|
+
reason: string,
|
|
614
|
+
key?: string,
|
|
615
|
+
): CheckpointDecision {
|
|
616
|
+
this.#commitTask();
|
|
617
|
+
const actualKey = key ?? checkpointKey(checkpoint, this.versions);
|
|
618
|
+
const metadata: AdvisorCheckpointMetadata = {
|
|
619
|
+
kind: 'advisor_checkpoint',
|
|
620
|
+
checkpoint,
|
|
621
|
+
checkpointKey: actualKey,
|
|
622
|
+
status: 'required',
|
|
623
|
+
...this.versions,
|
|
624
|
+
};
|
|
625
|
+
return {
|
|
626
|
+
allow: false,
|
|
627
|
+
deferred: {
|
|
628
|
+
ok: false,
|
|
629
|
+
disposition: 'deferred',
|
|
630
|
+
recovery: 'change-input',
|
|
631
|
+
sensitivity: 'public',
|
|
632
|
+
metadata,
|
|
633
|
+
error: `Advisor 检查点 ${checkpoint} 暂缓了这一步:${reason}\n请先调用 Advisor 获取第二意见;成功后调用 AdvisorDecision 记录采纳、部分采纳或拒绝及依据,然后重试原操作。若 Advisor 返回不可用/超时/配额不足,本检查点会降级放行,不要循环重试。`,
|
|
634
|
+
},
|
|
635
|
+
};
|
|
636
|
+
}
|
|
637
|
+
|
|
638
|
+
#deferPending(reason: string): CheckpointDecision {
|
|
639
|
+
const pending = this.#pending;
|
|
640
|
+
if (!pending) return { allow: true };
|
|
641
|
+
const key = pending.checkpointKey;
|
|
642
|
+
const seen = (this.#deferCounts.get(key) ?? 0) + 1;
|
|
643
|
+
this.#deferCounts.set(key, seen);
|
|
644
|
+
if (seen > MAX_CONSECUTIVE_DEFERRALS) {
|
|
645
|
+
this.#completed.add(key);
|
|
646
|
+
this.#pending = undefined;
|
|
647
|
+
const notice = `Advisor 检查点 ${pending.checkpoint} 的咨询 ${pending.consultationId} 已连续等待 AdvisorDecision ${MAX_CONSECUTIVE_DEFERRALS} 次,本次降级结清并放行。这一步没有完整的顾问决策记录,请自行核实并在汇报时如实说明。`;
|
|
648
|
+
if (!this.#reminders.includes(notice)) this.#reminders.push(notice);
|
|
649
|
+
return { allow: true };
|
|
650
|
+
}
|
|
651
|
+
return this.#defer(pending.checkpoint, reason, key);
|
|
652
|
+
}
|
|
653
|
+
}
|
|
654
|
+
|
|
655
|
+
export function isWorkspaceMutationTool(name: string): boolean {
|
|
656
|
+
return WRITE_TOOLS.has(name);
|
|
657
|
+
}
|
package/src/cli/configMenu.tsx
CHANGED
|
@@ -96,7 +96,6 @@ export function ConfigMenu({ onDone }: Props): JSX.Element {
|
|
|
96
96
|
const [editingModelId, setEditingModelId] = useState<string | null>(null);
|
|
97
97
|
const [message, setMessage] = useState<string | null>(null);
|
|
98
98
|
const [busy, setBusy] = useState(false);
|
|
99
|
-
|
|
100
99
|
const refresh = useCallback(async () => {
|
|
101
100
|
const snapshot = await configService.reload();
|
|
102
101
|
setConfig(snapshot.config);
|