@qduc/term2 0.1.8 → 0.2.0
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/dist/agent.d.ts.map +1 -1
- package/dist/agent.js +5 -3
- package/dist/agent.js.map +1 -1
- package/dist/app.d.ts +1 -5
- package/dist/app.d.ts.map +1 -1
- package/dist/app.js +31 -274
- package/dist/app.js.map +1 -1
- package/dist/app.parseInput.test.js +1 -9
- package/dist/app.parseInput.test.js.map +1 -1
- package/dist/cli.js +1 -0
- package/dist/cli.js.map +1 -1
- package/dist/components/ApprovalPrompt.d.ts +2 -1
- package/dist/components/ApprovalPrompt.d.ts.map +1 -1
- package/dist/components/ApprovalPrompt.js +15 -0
- package/dist/components/ApprovalPrompt.js.map +1 -1
- package/dist/components/BottomArea.d.ts +1 -1
- package/dist/components/BottomArea.d.ts.map +1 -1
- package/dist/components/Input/determine-active-menu.d.ts +23 -0
- package/dist/components/Input/determine-active-menu.d.ts.map +1 -0
- package/dist/components/Input/determine-active-menu.js +57 -0
- package/dist/components/Input/determine-active-menu.js.map +1 -0
- package/dist/components/Input/input-width.d.ts +7 -0
- package/dist/components/Input/input-width.d.ts.map +1 -0
- package/dist/components/Input/input-width.js +20 -0
- package/dist/components/Input/input-width.js.map +1 -0
- package/dist/components/Input/insertions.d.ts +34 -0
- package/dist/components/Input/insertions.d.ts.map +1 -0
- package/dist/components/Input/insertions.js +52 -0
- package/dist/components/Input/insertions.js.map +1 -0
- package/dist/components/Input/popup-props.d.ts +18 -0
- package/dist/components/Input/popup-props.d.ts.map +1 -0
- package/dist/components/Input/popup-props.js +42 -0
- package/dist/components/Input/popup-props.js.map +1 -0
- package/dist/components/Input/triggers.d.ts +9 -0
- package/dist/components/Input/triggers.d.ts.map +1 -0
- package/dist/components/Input/triggers.js +25 -0
- package/dist/components/Input/triggers.js.map +1 -0
- package/dist/components/InputBox.d.ts +2 -6
- package/dist/components/InputBox.d.ts.map +1 -1
- package/dist/components/InputBox.js +118 -406
- package/dist/components/InputBox.js.map +1 -1
- package/dist/components/InputBox.menu-logic.test.js +82 -99
- package/dist/components/InputBox.menu-logic.test.js.map +1 -1
- package/dist/components/MarkdownRenderer.d.ts.map +1 -1
- package/dist/components/MarkdownRenderer.js +96 -23
- package/dist/components/MarkdownRenderer.js.map +1 -1
- package/dist/components/MessageList.d.ts +8 -0
- package/dist/components/MessageList.d.ts.map +1 -1
- package/dist/components/MessageList.js +25 -7
- package/dist/components/MessageList.js.map +1 -1
- package/dist/components/MessageList.test.js +48 -1
- package/dist/components/MessageList.test.js.map +1 -1
- package/dist/components/StatusBar.d.ts +1 -1
- package/dist/components/StatusBar.d.ts.map +1 -1
- package/dist/components/StatusBar.js +49 -38
- package/dist/components/StatusBar.js.map +1 -1
- package/dist/contracts/conversation.d.ts +9 -0
- package/dist/contracts/conversation.d.ts.map +1 -1
- package/dist/hooks/message-id.d.ts +3 -0
- package/dist/hooks/message-id.d.ts.map +1 -0
- package/dist/hooks/message-id.js +17 -0
- package/dist/hooks/message-id.js.map +1 -0
- package/dist/hooks/message-id.test.d.ts +2 -0
- package/dist/hooks/message-id.test.d.ts.map +1 -0
- package/dist/hooks/message-id.test.js +17 -0
- package/dist/hooks/message-id.test.js.map +1 -0
- package/dist/hooks/use-app-commands.d.ts +20 -0
- package/dist/hooks/use-app-commands.d.ts.map +1 -0
- package/dist/hooks/use-app-commands.js +197 -0
- package/dist/hooks/use-app-commands.js.map +1 -0
- package/dist/hooks/use-app-commands.test.d.ts +2 -0
- package/dist/hooks/use-app-commands.test.d.ts.map +1 -0
- package/dist/hooks/use-app-commands.test.js +34 -0
- package/dist/hooks/use-app-commands.test.js.map +1 -0
- package/dist/hooks/use-conversation.d.ts +11 -11
- package/dist/hooks/use-conversation.d.ts.map +1 -1
- package/dist/hooks/use-conversation.js +13 -10
- package/dist/hooks/use-conversation.js.map +1 -1
- package/dist/hooks/use-escape-key.d.ts +24 -0
- package/dist/hooks/use-escape-key.d.ts.map +1 -0
- package/dist/hooks/use-escape-key.js +60 -0
- package/dist/hooks/use-escape-key.js.map +1 -0
- package/dist/hooks/use-escape-key.test.d.ts +2 -0
- package/dist/hooks/use-escape-key.test.d.ts.map +1 -0
- package/dist/hooks/use-escape-key.test.js +49 -0
- package/dist/hooks/use-escape-key.test.js.map +1 -0
- package/dist/hooks/use-mode-handlers.d.ts +35 -0
- package/dist/hooks/use-mode-handlers.d.ts.map +1 -0
- package/dist/hooks/use-mode-handlers.js +79 -0
- package/dist/hooks/use-mode-handlers.js.map +1 -0
- package/dist/hooks/use-model-selection.d.ts +6 -3
- package/dist/hooks/use-model-selection.d.ts.map +1 -1
- package/dist/hooks/use-model-selection.js +19 -16
- package/dist/hooks/use-model-selection.js.map +1 -1
- package/dist/hooks/use-model-selection.test.d.ts.map +1 -1
- package/dist/hooks/use-model-selection.test.js +53 -22
- package/dist/hooks/use-model-selection.test.js.map +1 -1
- package/dist/hooks/use-path-completion.d.ts.map +1 -1
- package/dist/hooks/use-path-completion.js +5 -36
- package/dist/hooks/use-path-completion.js.map +1 -1
- package/dist/hooks/use-runtime-settings.d.ts +12 -0
- package/dist/hooks/use-runtime-settings.d.ts.map +1 -0
- package/dist/hooks/use-runtime-settings.js +68 -0
- package/dist/hooks/use-runtime-settings.js.map +1 -0
- package/dist/hooks/use-selection.d.ts +8 -0
- package/dist/hooks/use-selection.d.ts.map +1 -0
- package/dist/hooks/use-selection.js +31 -0
- package/dist/hooks/use-selection.js.map +1 -0
- package/dist/hooks/use-settings-completion.d.ts.map +1 -1
- package/dist/hooks/use-settings-completion.js +15 -30
- package/dist/hooks/use-settings-completion.js.map +1 -1
- package/dist/hooks/use-settings-completion.test.js +9 -9
- package/dist/hooks/use-settings-completion.test.js.map +1 -1
- package/dist/hooks/use-settings-value-completion.d.ts.map +1 -1
- package/dist/hooks/use-settings-value-completion.js +30 -28
- package/dist/hooks/use-settings-value-completion.js.map +1 -1
- package/dist/hooks/use-settings-value-completion.test.js +13 -0
- package/dist/hooks/use-settings-value-completion.test.js.map +1 -1
- package/dist/hooks/use-shell-mode.d.ts +31 -0
- package/dist/hooks/use-shell-mode.d.ts.map +1 -0
- package/dist/hooks/use-shell-mode.js +85 -0
- package/dist/hooks/use-shell-mode.js.map +1 -0
- package/dist/hooks/use-slash-commands.d.ts +1 -0
- package/dist/hooks/use-slash-commands.d.ts.map +1 -1
- package/dist/hooks/use-slash-commands.js +32 -23
- package/dist/hooks/use-slash-commands.js.map +1 -1
- package/dist/hooks/use-terminal-width.d.ts +7 -0
- package/dist/hooks/use-terminal-width.d.ts.map +1 -0
- package/dist/hooks/use-terminal-width.js +33 -0
- package/dist/hooks/use-terminal-width.js.map +1 -0
- package/dist/hooks/use-trigger-detection.d.ts +40 -0
- package/dist/hooks/use-trigger-detection.d.ts.map +1 -0
- package/dist/hooks/use-trigger-detection.js +45 -0
- package/dist/hooks/use-trigger-detection.js.map +1 -0
- package/dist/lib/editor-impl.js +6 -6
- package/dist/lib/editor-impl.js.map +1 -1
- package/dist/lib/openai-agent-client.d.ts +2 -0
- package/dist/lib/openai-agent-client.d.ts.map +1 -1
- package/dist/lib/openai-agent-client.js +187 -191
- package/dist/lib/openai-agent-client.js.map +1 -1
- package/dist/lib/openai-agent-client.public-methods.test.js +61 -13
- package/dist/lib/openai-agent-client.public-methods.test.js.map +1 -1
- package/dist/lib/retry-executor.d.ts +16 -0
- package/dist/lib/retry-executor.d.ts.map +1 -0
- package/dist/lib/retry-executor.js +71 -0
- package/dist/lib/retry-executor.js.map +1 -0
- package/dist/lib/retry-executor.test.d.ts +2 -0
- package/dist/lib/retry-executor.test.d.ts.map +1 -0
- package/dist/lib/retry-executor.test.js +178 -0
- package/dist/lib/retry-executor.test.js.map +1 -0
- package/dist/lib/tool-selection-policy.d.ts +14 -0
- package/dist/lib/tool-selection-policy.d.ts.map +1 -0
- package/dist/lib/tool-selection-policy.js +17 -0
- package/dist/lib/tool-selection-policy.js.map +1 -0
- package/dist/lib/tool-selection-policy.test.d.ts +2 -0
- package/dist/lib/tool-selection-policy.test.d.ts.map +1 -0
- package/dist/lib/tool-selection-policy.test.js +69 -0
- package/dist/lib/tool-selection-policy.test.js.map +1 -0
- package/dist/providers/index.d.ts +1 -1
- package/dist/providers/index.d.ts.map +1 -1
- package/dist/providers/index.js +1 -1
- package/dist/providers/index.js.map +1 -1
- package/dist/providers/openai-compatible/model.d.ts.map +1 -1
- package/dist/providers/openai-compatible/model.js +68 -7
- package/dist/providers/openai-compatible/model.js.map +1 -1
- package/dist/providers/openai-compatible/reasoning-content.test.js +107 -6
- package/dist/providers/openai-compatible/reasoning-content.test.js.map +1 -1
- package/dist/providers/openai-compatible.provider.d.ts.map +1 -1
- package/dist/providers/openai-compatible.provider.js.map +1 -1
- package/dist/providers/openai.provider.js +2 -0
- package/dist/providers/openai.provider.js.map +1 -1
- package/dist/providers/openrouter/converters.d.ts.map +1 -1
- package/dist/providers/openrouter/converters.js +8 -6
- package/dist/providers/openrouter/converters.js.map +1 -1
- package/dist/providers/openrouter/converters.test.js +24 -0
- package/dist/providers/openrouter/converters.test.js.map +1 -1
- package/dist/providers/openrouter/model.d.ts.map +1 -1
- package/dist/providers/openrouter/model.js +13 -5
- package/dist/providers/openrouter/model.js.map +1 -1
- package/dist/providers/openrouter/reasoning-content.test.js +48 -2
- package/dist/providers/openrouter/reasoning-content.test.js.map +1 -1
- package/dist/providers/openrouter.provider.createRunner.test.js +17 -9
- package/dist/providers/openrouter.provider.createRunner.test.js.map +1 -1
- package/dist/providers/openrouter.provider.js.map +1 -1
- package/dist/providers/openrouter.test.js +0 -2
- package/dist/providers/openrouter.test.js.map +1 -1
- package/dist/providers/registry.d.ts +17 -8
- package/dist/providers/registry.d.ts.map +1 -1
- package/dist/providers/registry.js +9 -0
- package/dist/providers/registry.js.map +1 -1
- package/dist/providers/web-search/exa.provider.d.ts +15 -0
- package/dist/providers/web-search/exa.provider.d.ts.map +1 -0
- package/dist/providers/web-search/exa.provider.js +82 -0
- package/dist/providers/web-search/exa.provider.js.map +1 -0
- package/dist/providers/web-search/exa.provider.test.d.ts +2 -0
- package/dist/providers/web-search/exa.provider.test.d.ts.map +1 -0
- package/dist/providers/web-search/exa.provider.test.js +67 -0
- package/dist/providers/web-search/exa.provider.test.js.map +1 -0
- package/dist/providers/web-search/index.d.ts +1 -0
- package/dist/providers/web-search/index.d.ts.map +1 -1
- package/dist/providers/web-search/index.js +1 -0
- package/dist/providers/web-search/index.js.map +1 -1
- package/dist/scripts/eval-auto-approval/cache.d.ts +9 -0
- package/dist/scripts/eval-auto-approval/cache.d.ts.map +1 -0
- package/dist/scripts/eval-auto-approval/cache.js +64 -0
- package/dist/scripts/eval-auto-approval/cache.js.map +1 -0
- package/dist/scripts/eval-auto-approval/dataset.d.ts +45 -0
- package/dist/scripts/eval-auto-approval/dataset.d.ts.map +1 -0
- package/dist/scripts/eval-auto-approval/dataset.js +65 -0
- package/dist/scripts/eval-auto-approval/dataset.js.map +1 -0
- package/dist/scripts/eval-auto-approval/dataset.test.d.ts +2 -0
- package/dist/scripts/eval-auto-approval/dataset.test.d.ts.map +1 -0
- package/dist/scripts/eval-auto-approval/dataset.test.js +88 -0
- package/dist/scripts/eval-auto-approval/dataset.test.js.map +1 -0
- package/dist/scripts/eval-auto-approval/leaderboard.d.ts +92 -0
- package/dist/scripts/eval-auto-approval/leaderboard.d.ts.map +1 -0
- package/dist/scripts/eval-auto-approval/leaderboard.js +402 -0
- package/dist/scripts/eval-auto-approval/leaderboard.js.map +1 -0
- package/dist/scripts/eval-auto-approval/leaderboard.test.d.ts +2 -0
- package/dist/scripts/eval-auto-approval/leaderboard.test.d.ts.map +1 -0
- package/dist/scripts/eval-auto-approval/leaderboard.test.js +689 -0
- package/dist/scripts/eval-auto-approval/leaderboard.test.js.map +1 -0
- package/dist/scripts/eval-auto-approval/metrics.d.ts +34 -0
- package/dist/scripts/eval-auto-approval/metrics.d.ts.map +1 -0
- package/dist/scripts/eval-auto-approval/metrics.js +74 -0
- package/dist/scripts/eval-auto-approval/metrics.js.map +1 -0
- package/dist/scripts/eval-auto-approval/metrics.test.d.ts +2 -0
- package/dist/scripts/eval-auto-approval/metrics.test.d.ts.map +1 -0
- package/dist/scripts/eval-auto-approval/metrics.test.js +79 -0
- package/dist/scripts/eval-auto-approval/metrics.test.js.map +1 -0
- package/dist/scripts/eval-auto-approval/report.d.ts +5 -0
- package/dist/scripts/eval-auto-approval/report.d.ts.map +1 -0
- package/dist/scripts/eval-auto-approval/report.js +51 -0
- package/dist/scripts/eval-auto-approval/report.js.map +1 -0
- package/dist/scripts/eval-auto-approval/report.test.d.ts +2 -0
- package/dist/scripts/eval-auto-approval/report.test.d.ts.map +1 -0
- package/dist/scripts/eval-auto-approval/report.test.js +23 -0
- package/dist/scripts/eval-auto-approval/report.test.js.map +1 -0
- package/dist/scripts/eval-auto-approval/runner-utils.d.ts +41 -0
- package/dist/scripts/eval-auto-approval/runner-utils.d.ts.map +1 -0
- package/dist/scripts/eval-auto-approval/runner-utils.js +187 -0
- package/dist/scripts/eval-auto-approval/runner-utils.js.map +1 -0
- package/dist/scripts/eval-auto-approval/runner-utils.test.d.ts +2 -0
- package/dist/scripts/eval-auto-approval/runner-utils.test.d.ts.map +1 -0
- package/dist/scripts/eval-auto-approval/runner-utils.test.js +135 -0
- package/dist/scripts/eval-auto-approval/runner-utils.test.js.map +1 -0
- package/dist/scripts/eval-auto-approval/runner.d.ts +2 -0
- package/dist/scripts/eval-auto-approval/runner.d.ts.map +1 -0
- package/dist/scripts/eval-auto-approval/runner.js +326 -0
- package/dist/scripts/eval-auto-approval/runner.js.map +1 -0
- package/dist/services/agent-stream.d.ts +11 -0
- package/dist/services/agent-stream.d.ts.map +1 -0
- package/dist/services/agent-stream.js +2 -0
- package/dist/services/agent-stream.js.map +1 -0
- package/dist/services/approval-flow-coordinator.d.ts +43 -0
- package/dist/services/approval-flow-coordinator.d.ts.map +1 -0
- package/dist/services/approval-flow-coordinator.js +120 -0
- package/dist/services/approval-flow-coordinator.js.map +1 -0
- package/dist/services/approval-flow-coordinator.test.d.ts +2 -0
- package/dist/services/approval-flow-coordinator.test.d.ts.map +1 -0
- package/dist/services/approval-flow-coordinator.test.js +168 -0
- package/dist/services/approval-flow-coordinator.test.js.map +1 -0
- package/dist/services/approval-rejection-interceptor.d.ts +16 -0
- package/dist/services/approval-rejection-interceptor.d.ts.map +1 -0
- package/dist/services/approval-rejection-interceptor.js +20 -0
- package/dist/services/approval-rejection-interceptor.js.map +1 -0
- package/dist/services/approval-rejection-interceptor.test.d.ts +2 -0
- package/dist/services/approval-rejection-interceptor.test.d.ts.map +1 -0
- package/dist/services/approval-rejection-interceptor.test.js +60 -0
- package/dist/services/approval-rejection-interceptor.test.js.map +1 -0
- package/dist/services/conversation-events.d.ts +4 -2
- package/dist/services/conversation-events.d.ts.map +1 -1
- package/dist/services/conversation-result-builder.d.ts +34 -34
- package/dist/services/conversation-result-builder.d.ts.map +1 -1
- package/dist/services/conversation-result-builder.js +64 -76
- package/dist/services/conversation-result-builder.js.map +1 -1
- package/dist/services/conversation-result-builder.test.js +145 -65
- package/dist/services/conversation-result-builder.test.js.map +1 -1
- package/dist/services/conversation-retry-policy.d.ts +37 -0
- package/dist/services/conversation-retry-policy.d.ts.map +1 -0
- package/dist/services/conversation-retry-policy.js +51 -0
- package/dist/services/conversation-retry-policy.js.map +1 -0
- package/dist/services/conversation-retry-policy.test.d.ts +2 -0
- package/dist/services/conversation-retry-policy.test.d.ts.map +1 -0
- package/dist/services/conversation-retry-policy.test.js +79 -0
- package/dist/services/conversation-retry-policy.test.js.map +1 -0
- package/dist/services/conversation-service.d.ts +2 -1
- package/dist/services/conversation-service.d.ts.map +1 -1
- package/dist/services/conversation-session.auto-approval.test.d.ts +2 -0
- package/dist/services/conversation-session.auto-approval.test.d.ts.map +1 -0
- package/dist/services/conversation-session.auto-approval.test.js +617 -0
- package/dist/services/conversation-session.auto-approval.test.js.map +1 -0
- package/dist/services/conversation-session.d.ts +6 -3
- package/dist/services/conversation-session.d.ts.map +1 -1
- package/dist/services/conversation-session.js +164 -696
- package/dist/services/conversation-session.js.map +1 -1
- package/dist/services/interruption-info.d.ts +13 -0
- package/dist/services/interruption-info.d.ts.map +1 -0
- package/dist/services/interruption-info.js +89 -0
- package/dist/services/interruption-info.js.map +1 -0
- package/dist/services/interruption-info.test.d.ts +2 -0
- package/dist/services/interruption-info.test.d.ts.map +1 -0
- package/dist/services/interruption-info.test.js +84 -0
- package/dist/services/interruption-info.test.js.map +1 -0
- package/dist/services/logging-contract.d.ts +1 -1
- package/dist/services/logging-service.d.ts +6 -2
- package/dist/services/logging-service.d.ts.map +1 -1
- package/dist/services/logging-service.js +48 -91
- package/dist/services/logging-service.js.map +1 -1
- package/dist/services/settings-env.d.ts.map +1 -1
- package/dist/services/settings-env.js +3 -0
- package/dist/services/settings-env.js.map +1 -1
- package/dist/services/settings-schema.d.ts +34 -0
- package/dist/services/settings-schema.d.ts.map +1 -1
- package/dist/services/settings-schema.js +37 -0
- package/dist/services/settings-schema.js.map +1 -1
- package/dist/services/settings-service.d.ts.map +1 -1
- package/dist/services/settings-service.js +12 -166
- package/dist/services/settings-service.js.map +1 -1
- package/dist/services/settings-sources.d.ts +5 -0
- package/dist/services/settings-sources.d.ts.map +1 -0
- package/dist/services/settings-sources.js +92 -0
- package/dist/services/settings-sources.js.map +1 -0
- package/dist/services/settings-sources.test.d.ts +2 -0
- package/dist/services/settings-sources.test.d.ts.map +1 -0
- package/dist/services/settings-sources.test.js +34 -0
- package/dist/services/settings-sources.test.js.map +1 -0
- package/dist/services/shell-auto-approval-evaluator.d.ts +19 -0
- package/dist/services/shell-auto-approval-evaluator.d.ts.map +1 -0
- package/dist/services/shell-auto-approval-evaluator.js +207 -0
- package/dist/services/shell-auto-approval-evaluator.js.map +1 -0
- package/dist/services/shell-auto-approval-evaluator.test.d.ts +2 -0
- package/dist/services/shell-auto-approval-evaluator.test.d.ts.map +1 -0
- package/dist/services/shell-auto-approval-evaluator.test.js +211 -0
- package/dist/services/shell-auto-approval-evaluator.test.js.map +1 -0
- package/dist/services/shell-auto-approval-resolver.d.ts +24 -0
- package/dist/services/shell-auto-approval-resolver.d.ts.map +1 -0
- package/dist/services/shell-auto-approval-resolver.js +59 -0
- package/dist/services/shell-auto-approval-resolver.js.map +1 -0
- package/dist/services/shell-auto-approval-resolver.test.d.ts +2 -0
- package/dist/services/shell-auto-approval-resolver.test.d.ts.map +1 -0
- package/dist/services/shell-auto-approval-resolver.test.js +110 -0
- package/dist/services/shell-auto-approval-resolver.test.js.map +1 -0
- package/dist/services/ssh-service.test.js +2 -2
- package/dist/services/ssh-service.test.js.map +1 -1
- package/dist/services/stream-event-processor.d.ts +18 -23
- package/dist/services/stream-event-processor.d.ts.map +1 -1
- package/dist/services/stream-event-processor.js +186 -325
- package/dist/services/stream-event-processor.js.map +1 -1
- package/dist/services/stream-event-processor.test.js +143 -90
- package/dist/services/stream-event-processor.test.js.map +1 -1
- package/dist/services/terminal-result-collector.d.ts +12 -0
- package/dist/services/terminal-result-collector.d.ts.map +1 -0
- package/dist/services/terminal-result-collector.js +64 -0
- package/dist/services/terminal-result-collector.js.map +1 -0
- package/dist/services/terminal-result-collector.test.d.ts +2 -0
- package/dist/services/terminal-result-collector.test.d.ts.map +1 -0
- package/dist/services/terminal-result-collector.test.js +93 -0
- package/dist/services/terminal-result-collector.test.js.map +1 -0
- package/dist/tools/apply-patch.js +5 -5
- package/dist/tools/apply-patch.js.map +1 -1
- package/dist/tools/create-file.js +2 -2
- package/dist/tools/create-file.js.map +1 -1
- package/dist/tools/edit-healing.d.ts.map +1 -1
- package/dist/tools/edit-healing.js +4 -1
- package/dist/tools/edit-healing.js.map +1 -1
- package/dist/tools/edit-healing.test.js +22 -0
- package/dist/tools/edit-healing.test.js.map +1 -1
- package/dist/tools/search-replace.d.ts.map +1 -1
- package/dist/tools/search-replace.js +114 -194
- package/dist/tools/search-replace.js.map +1 -1
- package/dist/tools/search-replace.test.js +38 -0
- package/dist/tools/search-replace.test.js.map +1 -1
- package/dist/tools/shell.js +2 -2
- package/dist/tools/shell.js.map +1 -1
- package/dist/tools/shell.test.js +0 -5
- package/dist/tools/shell.test.js.map +1 -1
- package/dist/tools/web-fetch.test.js +3 -3
- package/dist/tools/web-fetch.test.js.map +1 -1
- package/dist/utils/clipboard.d.ts +16 -0
- package/dist/utils/clipboard.d.ts.map +1 -0
- package/dist/utils/clipboard.js +55 -0
- package/dist/utils/clipboard.js.map +1 -0
- package/dist/utils/clipboard.test.d.ts +2 -0
- package/dist/utils/clipboard.test.d.ts.map +1 -0
- package/dist/utils/clipboard.test.js +59 -0
- package/dist/utils/clipboard.test.js.map +1 -0
- package/dist/utils/command-safety/constants.d.ts.map +1 -1
- package/dist/utils/command-safety/constants.js +3 -21
- package/dist/utils/command-safety/constants.js.map +1 -1
- package/dist/utils/command-safety/find-helpers.d.ts.map +1 -1
- package/dist/utils/command-safety/find-helpers.js +13 -95
- package/dist/utils/command-safety/find-helpers.js.map +1 -1
- package/dist/utils/command-safety/handlers/sed-handler.d.ts.map +1 -1
- package/dist/utils/command-safety/handlers/sed-handler.js +2 -4
- package/dist/utils/command-safety/handlers/sed-handler.js.map +1 -1
- package/dist/utils/command-safety/index.d.ts +9 -0
- package/dist/utils/command-safety/index.d.ts.map +1 -1
- package/dist/utils/command-safety/index.js +12 -2
- package/dist/utils/command-safety/index.js.map +1 -1
- package/dist/utils/command-safety/path-analysis.d.ts.map +1 -1
- package/dist/utils/command-safety/path-analysis.js +7 -6
- package/dist/utils/command-safety/path-analysis.js.map +1 -1
- package/dist/utils/command-safety.find.test.js +56 -39
- package/dist/utils/command-safety.find.test.js.map +1 -1
- package/dist/utils/command-safety.path.test.js +7 -7
- package/dist/utils/command-safety.path.test.js.map +1 -1
- package/dist/utils/command-safety.red-yellow-policy.test.d.ts +2 -0
- package/dist/utils/command-safety.red-yellow-policy.test.d.ts.map +1 -0
- package/dist/utils/command-safety.red-yellow-policy.test.js +36 -0
- package/dist/utils/command-safety.red-yellow-policy.test.js.map +1 -0
- package/dist/utils/conversation-event-handler.d.ts +4 -3
- package/dist/utils/conversation-event-handler.d.ts.map +1 -1
- package/dist/utils/conversation-event-handler.js +15 -5
- package/dist/utils/conversation-event-handler.js.map +1 -1
- package/dist/utils/conversation-event-handler.test.js +24 -0
- package/dist/utils/conversation-event-handler.test.js.map +1 -1
- package/dist/utils/conversation-utils.d.ts +1 -1
- package/dist/utils/execute-shell.d.ts +10 -0
- package/dist/utils/execute-shell.d.ts.map +1 -1
- package/dist/utils/execute-shell.js +3 -3
- package/dist/utils/execute-shell.js.map +1 -1
- package/dist/utils/execute-shell.test.js +19 -3
- package/dist/utils/execute-shell.test.js.map +1 -1
- package/dist/utils/input-parser.d.ts +13 -0
- package/dist/utils/input-parser.d.ts.map +1 -0
- package/dist/utils/input-parser.js +10 -0
- package/dist/utils/input-parser.js.map +1 -0
- package/dist/utils/model-provider-arg.d.ts +5 -0
- package/dist/utils/model-provider-arg.d.ts.map +1 -0
- package/dist/utils/model-provider-arg.js +15 -0
- package/dist/utils/model-provider-arg.js.map +1 -0
- package/dist/utils/model-provider-arg.test.d.ts +2 -0
- package/dist/utils/model-provider-arg.test.d.ts.map +1 -0
- package/dist/utils/model-provider-arg.test.js +21 -0
- package/dist/utils/model-provider-arg.test.js.map +1 -0
- package/dist/utils/model-settings.d.ts +12 -0
- package/dist/utils/model-settings.d.ts.map +1 -0
- package/dist/utils/model-settings.js +35 -0
- package/dist/utils/model-settings.js.map +1 -0
- package/dist/utils/provider-traffic-extractor.d.ts +1 -0
- package/dist/utils/provider-traffic-extractor.d.ts.map +1 -1
- package/dist/utils/provider-traffic-extractor.js +90 -5
- package/dist/utils/provider-traffic-extractor.js.map +1 -1
- package/dist/utils/settings-command.d.ts.map +1 -1
- package/dist/utils/settings-command.js +10 -8
- package/dist/utils/settings-command.js.map +1 -1
- package/dist/utils/streaming-session-factory.d.ts +1 -1
- package/dist/utils/streaming-session-factory.d.ts.map +1 -1
- package/dist/utils/streaming-session-factory.js +5 -2
- package/dist/utils/streaming-session-factory.js.map +1 -1
- package/dist/utils/streaming-session-factory.test.js +2 -2
- package/dist/utils/streaming-session-factory.test.js.map +1 -1
- package/dist/utils/streaming-updater.d.ts +2 -1
- package/dist/utils/streaming-updater.d.ts.map +1 -1
- package/dist/utils/streaming-updater.js +2 -2
- package/dist/utils/streaming-updater.js.map +1 -1
- package/dist/utils/throttle.d.ts +6 -1
- package/dist/utils/throttle.d.ts.map +1 -1
- package/dist/utils/throttle.js +11 -6
- package/dist/utils/throttle.js.map +1 -1
- package/dist/utils/viewport.d.ts +6 -0
- package/dist/utils/viewport.d.ts.map +1 -0
- package/dist/utils/viewport.js +20 -0
- package/dist/utils/viewport.js.map +1 -0
- package/dist/utils/viewport.test.d.ts +2 -0
- package/dist/utils/viewport.test.d.ts.map +1 -0
- package/dist/utils/viewport.test.js +43 -0
- package/dist/utils/viewport.test.js.map +1 -0
- package/package.json +5 -4
- package/dist/app.model-command-feedback.test.d.ts +0 -2
- package/dist/app.model-command-feedback.test.d.ts.map +0 -1
- package/dist/app.model-command-feedback.test.js +0 -19
- package/dist/app.model-command-feedback.test.js.map +0 -1
- package/dist/components/MentorMode.test.d.ts +0 -2
- package/dist/components/MentorMode.test.d.ts.map +0 -1
- package/dist/components/MentorMode.test.js.map +0 -1
- package/dist/components/ModelSelectionMenu.error-tabs.test.d.ts +0 -2
- package/dist/components/ModelSelectionMenu.error-tabs.test.d.ts.map +0 -1
- package/dist/components/ModelSelectionMenu.error-tabs.test.js +0 -18
- package/dist/components/ModelSelectionMenu.error-tabs.test.js.map +0 -1
- package/dist/components/StatusBar.test.d.ts +0 -2
- package/dist/components/StatusBar.test.d.ts.map +0 -1
- package/dist/components/StatusBar.test.js +0 -19
- package/dist/components/StatusBar.test.js.map +0 -1
- package/dist/components/TextInput.d.ts +0 -42
- package/dist/components/TextInput.d.ts.map +0 -1
- package/dist/components/TextInput.js +0 -397
- package/dist/components/TextInput.js.map +0 -1
- package/dist/components/TextInput.test.d.ts +0 -2
- package/dist/components/TextInput.test.d.ts.map +0 -1
- package/dist/components/TextInput.test.js +0 -75
- package/dist/components/TextInput.test.js.map +0 -1
- package/dist/debug-schema.d.ts +0 -2
- package/dist/debug-schema.d.ts.map +0 -1
- package/dist/debug-schema.js +0 -22
- package/dist/debug-schema.js.map +0 -1
- package/dist/hooks/use-path-completion.test.d.ts +0 -2
- package/dist/hooks/use-path-completion.test.d.ts.map +0 -1
- package/dist/hooks/use-path-completion.test.js +0 -29
- package/dist/hooks/use-path-completion.test.js.map +0 -1
- package/dist/lib/mentor-client.d.ts +0 -41
- package/dist/lib/mentor-client.d.ts.map +0 -1
- package/dist/lib/mentor-client.js +0 -146
- package/dist/lib/mentor-client.js.map +0 -1
- package/dist/lib/mentor-client.test.d.ts +0 -2
- package/dist/lib/mentor-client.test.d.ts.map +0 -1
- package/dist/lib/mentor-client.test.js +0 -440
- package/dist/lib/mentor-client.test.js.map +0 -1
- package/dist/lib/retry-strategy.d.ts +0 -43
- package/dist/lib/retry-strategy.d.ts.map +0 -1
- package/dist/lib/retry-strategy.js +0 -93
- package/dist/lib/retry-strategy.js.map +0 -1
- package/dist/lib/retry-strategy.test.d.ts +0 -2
- package/dist/lib/retry-strategy.test.d.ts.map +0 -1
- package/dist/lib/retry-strategy.test.js +0 -200
- package/dist/lib/retry-strategy.test.js.map +0 -1
- package/dist/lib/shell.d.ts +0 -7
- package/dist/lib/shell.d.ts.map +0 -1
- package/dist/lib/shell.js +0 -56
- package/dist/lib/shell.js.map +0 -1
- package/dist/prompts/default.md.bak +0 -77
- package/dist/prompts/simple-mentor.md +0 -207
- package/dist/providers/github-copilot/converters.d.ts +0 -45
- package/dist/providers/github-copilot/converters.d.ts.map +0 -1
- package/dist/providers/github-copilot/converters.js +0 -118
- package/dist/providers/github-copilot/converters.js.map +0 -1
- package/dist/providers/github-copilot/converters.test.d.ts +0 -2
- package/dist/providers/github-copilot/converters.test.d.ts.map +0 -1
- package/dist/providers/github-copilot/converters.test.js +0 -162
- package/dist/providers/github-copilot/converters.test.js.map +0 -1
- package/dist/providers/github-copilot/github-copilot.provider.d.ts +0 -2
- package/dist/providers/github-copilot/github-copilot.provider.d.ts.map +0 -1
- package/dist/providers/github-copilot/github-copilot.provider.js +0 -75
- package/dist/providers/github-copilot/github-copilot.provider.js.map +0 -1
- package/dist/providers/github-copilot/github-copilot.provider.test.d.ts +0 -2
- package/dist/providers/github-copilot/github-copilot.provider.test.d.ts.map +0 -1
- package/dist/providers/github-copilot/github-copilot.provider.test.js +0 -26
- package/dist/providers/github-copilot/github-copilot.provider.test.js.map +0 -1
- package/dist/providers/github-copilot/index.d.ts +0 -4
- package/dist/providers/github-copilot/index.d.ts.map +0 -1
- package/dist/providers/github-copilot/index.js +0 -4
- package/dist/providers/github-copilot/index.js.map +0 -1
- package/dist/providers/github-copilot/model-direct.d.ts +0 -34
- package/dist/providers/github-copilot/model-direct.d.ts.map +0 -1
- package/dist/providers/github-copilot/model-direct.js +0 -443
- package/dist/providers/github-copilot/model-direct.js.map +0 -1
- package/dist/providers/github-copilot/model.d.ts +0 -24
- package/dist/providers/github-copilot/model.d.ts.map +0 -1
- package/dist/providers/github-copilot/model.delta.test.d.ts +0 -2
- package/dist/providers/github-copilot/model.delta.test.d.ts.map +0 -1
- package/dist/providers/github-copilot/model.delta.test.js +0 -15
- package/dist/providers/github-copilot/model.delta.test.js.map +0 -1
- package/dist/providers/github-copilot/model.js +0 -581
- package/dist/providers/github-copilot/model.js.map +0 -1
- package/dist/providers/github-copilot/provider.d.ts +0 -20
- package/dist/providers/github-copilot/provider.d.ts.map +0 -1
- package/dist/providers/github-copilot/provider.js +0 -30
- package/dist/providers/github-copilot/provider.js.map +0 -1
- package/dist/providers/github-copilot/provider.test.d.ts +0 -2
- package/dist/providers/github-copilot/provider.test.d.ts.map +0 -1
- package/dist/providers/github-copilot/provider.test.js +0 -52
- package/dist/providers/github-copilot/provider.test.js.map +0 -1
- package/dist/providers/github-copilot/utils.d.ts +0 -20
- package/dist/providers/github-copilot/utils.d.ts.map +0 -1
- package/dist/providers/github-copilot/utils.js +0 -142
- package/dist/providers/github-copilot/utils.js.map +0 -1
- package/dist/providers/github-copilot/utils.test.d.ts +0 -2
- package/dist/providers/github-copilot/utils.test.d.ts.map +0 -1
- package/dist/providers/github-copilot/utils.test.js +0 -21
- package/dist/providers/github-copilot/utils.test.js.map +0 -1
- package/dist/providers/openrouter/merge-messages.test.d.ts +0 -2
- package/dist/providers/openrouter/merge-messages.test.d.ts.map +0 -1
- package/dist/providers/openrouter/merge-messages.test.js +0 -83
- package/dist/providers/openrouter/merge-messages.test.js.map +0 -1
- package/dist/providers/openrouter.history.test.d.ts +0 -2
- package/dist/providers/openrouter.history.test.d.ts.map +0 -1
- package/dist/providers/openrouter.history.test.js +0 -533
- package/dist/providers/openrouter.history.test.js.map +0 -1
- package/dist/reproduce_issue.test.d.ts +0 -2
- package/dist/reproduce_issue.test.d.ts.map +0 -1
- package/dist/reproduce_issue.test.js +0 -31
- package/dist/reproduce_issue.test.js.map +0 -1
- package/dist/safety-checker.js +0 -57
- package/dist/services/conversation-service.test.js +0 -190
- package/dist/services/conversation-session.usage.test.d.ts +0 -2
- package/dist/services/conversation-session.usage.test.d.ts.map +0 -1
- package/dist/services/conversation-session.usage.test.js +0 -59
- package/dist/services/conversation-session.usage.test.js.map +0 -1
- package/dist/test-search-tool.d.ts +0 -2
- package/dist/test-search-tool.d.ts.map +0 -1
- package/dist/test-search-tool.js +0 -36
- package/dist/test-search-tool.js.map +0 -1
- package/dist/tools/bash.d.ts +0 -10
- package/dist/tools/bash.d.ts.map +0 -1
- package/dist/tools/bash.js +0 -55
- package/dist/tools/bash.js.map +0 -1
- package/dist/types/messages.d.ts +0 -29
- package/dist/types/messages.d.ts.map +0 -1
- package/dist/types/messages.js +0 -2
- package/dist/types/messages.js.map +0 -1
- package/dist/utils/command-safety.d.ts +0 -21
- package/dist/utils/command-safety.d.ts.map +0 -1
- package/dist/utils/command-safety.js +0 -702
- package/dist/utils/command-safety.js.map +0 -1
- package/dist/utils/extract-command-messages.repro.test.d.ts +0 -2
- package/dist/utils/extract-command-messages.repro.test.d.ts.map +0 -1
- package/dist/utils/extract-command-messages.repro.test.js +0 -31
- package/dist/utils/extract-command-messages.repro.test.js.map +0 -1
- package/dist/utils/extract-command-messages.test.js +0 -57
- package/dist/utils/message-utils.d.ts +0 -17
- package/dist/utils/message-utils.d.ts.map +0 -1
- package/dist/utils/message-utils.js +0 -52
- package/dist/utils/message-utils.js.map +0 -1
- package/dist/utils/message-utils.test.d.ts +0 -2
- package/dist/utils/message-utils.test.d.ts.map +0 -1
- package/dist/utils/message-utils.test.js +0 -48
- package/dist/utils/message-utils.test.js.map +0 -1
|
@@ -0,0 +1,135 @@
|
|
|
1
|
+
import test from 'ava';
|
|
2
|
+
import { mkdtempSync, writeFileSync } from 'node:fs';
|
|
3
|
+
import { tmpdir } from 'node:os';
|
|
4
|
+
import { join } from 'node:path';
|
|
5
|
+
import { createCacheKey, EVAL_CACHE_VERSION, getReportPath, loadModelRunsFromYaml, retryOnRateLimit, validateRunnerOptions, } from './runner-utils.js';
|
|
6
|
+
test('getReportPath produces a distinct markdown path for json and extensionless outputs', (t) => {
|
|
7
|
+
t.is(getReportPath('eval/auto-approval/results.json'), 'eval/auto-approval/results.md');
|
|
8
|
+
t.is(getReportPath('eval/auto-approval/results'), 'eval/auto-approval/results.md');
|
|
9
|
+
});
|
|
10
|
+
test('validateRunnerOptions rejects non-positive concurrency and repeat values', (t) => {
|
|
11
|
+
t.notThrows(() => validateRunnerOptions({ concurrency: 1, repeat: 1 }));
|
|
12
|
+
t.throws(() => validateRunnerOptions({ concurrency: 0, repeat: 1 }), {
|
|
13
|
+
message: /--concurrency/,
|
|
14
|
+
});
|
|
15
|
+
t.throws(() => validateRunnerOptions({ concurrency: 1, repeat: 0 }), {
|
|
16
|
+
message: /--repeat/,
|
|
17
|
+
});
|
|
18
|
+
});
|
|
19
|
+
test('createCacheKey includes evaluator version to invalidate stale cached decisions', (t) => {
|
|
20
|
+
const key = createCacheKey({
|
|
21
|
+
model: 'gpt-4o-mini',
|
|
22
|
+
provider: 'openai',
|
|
23
|
+
promptVersion: 'auto-approval-prompt-v1',
|
|
24
|
+
command: 'ls',
|
|
25
|
+
history: [{ role: 'user', content: 'list files' }],
|
|
26
|
+
});
|
|
27
|
+
t.is(key.version, EVAL_CACHE_VERSION);
|
|
28
|
+
t.is(key.promptVersion, 'auto-approval-prompt-v1');
|
|
29
|
+
});
|
|
30
|
+
test('createCacheKey changes when prompt version changes', (t) => {
|
|
31
|
+
const base = {
|
|
32
|
+
model: 'gpt-4o-mini',
|
|
33
|
+
provider: 'openai',
|
|
34
|
+
command: 'ls',
|
|
35
|
+
history: [{ role: 'user', content: 'list files' }],
|
|
36
|
+
};
|
|
37
|
+
t.notDeepEqual(createCacheKey({ ...base, promptVersion: 'auto-approval-prompt-v1' }), createCacheKey({ ...base, promptVersion: 'auto-approval-prompt-v2' }));
|
|
38
|
+
});
|
|
39
|
+
test('loadModelRunsFromYaml expands provider keys into ordered provider/model pairs', (t) => {
|
|
40
|
+
const dir = mkdtempSync(join(tmpdir(), 'term2-eval-'));
|
|
41
|
+
const yamlPath = join(dir, 'models.yaml');
|
|
42
|
+
writeFileSync(yamlPath, `
|
|
43
|
+
openai:
|
|
44
|
+
- gpt-4o
|
|
45
|
+
- gpt-4o-mini
|
|
46
|
+
openrouter:
|
|
47
|
+
- anthropic/claude-3.5-sonnet
|
|
48
|
+
`);
|
|
49
|
+
t.deepEqual(loadModelRunsFromYaml(yamlPath), [
|
|
50
|
+
{ provider: 'openai', model: 'gpt-4o' },
|
|
51
|
+
{ provider: 'openai', model: 'gpt-4o-mini' },
|
|
52
|
+
{ provider: 'openrouter', model: 'anthropic/claude-3.5-sonnet' },
|
|
53
|
+
]);
|
|
54
|
+
});
|
|
55
|
+
test('loadModelRunsFromYaml rejects providers without a model list', (t) => {
|
|
56
|
+
const dir = mkdtempSync(join(tmpdir(), 'term2-eval-'));
|
|
57
|
+
const yamlPath = join(dir, 'models.yaml');
|
|
58
|
+
writeFileSync(yamlPath, `
|
|
59
|
+
openai: gpt-4o
|
|
60
|
+
`);
|
|
61
|
+
t.throws(() => loadModelRunsFromYaml(yamlPath), {
|
|
62
|
+
message: /must map to a list of models/,
|
|
63
|
+
});
|
|
64
|
+
});
|
|
65
|
+
test('retryOnRateLimit retries twice and respects x-ratelimit-reset before succeeding', async (t) => {
|
|
66
|
+
let attempts = 0;
|
|
67
|
+
const sleepCalls = [];
|
|
68
|
+
const now = 1_741_305_599_000;
|
|
69
|
+
const result = await retryOnRateLimit({
|
|
70
|
+
operation: async () => {
|
|
71
|
+
attempts++;
|
|
72
|
+
if (attempts < 3) {
|
|
73
|
+
const err = new Error('Rate limit exceeded');
|
|
74
|
+
err.status = 429;
|
|
75
|
+
err.headers = {
|
|
76
|
+
'X-RateLimit-Reset': String(now + 1_500),
|
|
77
|
+
};
|
|
78
|
+
throw err;
|
|
79
|
+
}
|
|
80
|
+
return 'ok';
|
|
81
|
+
},
|
|
82
|
+
maxRetries: 2,
|
|
83
|
+
sleep: async (ms) => {
|
|
84
|
+
sleepCalls.push(ms);
|
|
85
|
+
},
|
|
86
|
+
now: () => now,
|
|
87
|
+
});
|
|
88
|
+
t.is(result, 'ok');
|
|
89
|
+
t.is(attempts, 3);
|
|
90
|
+
t.deepEqual(sleepCalls, [1500, 1500]);
|
|
91
|
+
});
|
|
92
|
+
test('retryOnRateLimit rethrows after exhausting retries', async (t) => {
|
|
93
|
+
let attempts = 0;
|
|
94
|
+
const sleepCalls = [];
|
|
95
|
+
const error = await t.throwsAsync(() => retryOnRateLimit({
|
|
96
|
+
operation: async () => {
|
|
97
|
+
attempts++;
|
|
98
|
+
const err = Object.assign(new Error('Too Many Requests'), {
|
|
99
|
+
error: {
|
|
100
|
+
code: 429,
|
|
101
|
+
metadata: {
|
|
102
|
+
headers: {
|
|
103
|
+
'X-RateLimit-Reset': '1741305600000',
|
|
104
|
+
},
|
|
105
|
+
},
|
|
106
|
+
},
|
|
107
|
+
});
|
|
108
|
+
throw err;
|
|
109
|
+
},
|
|
110
|
+
maxRetries: 2,
|
|
111
|
+
sleep: async (ms) => {
|
|
112
|
+
sleepCalls.push(ms);
|
|
113
|
+
},
|
|
114
|
+
now: () => 1_741_305_599_000,
|
|
115
|
+
}));
|
|
116
|
+
t.truthy(error);
|
|
117
|
+
t.is(attempts, 3);
|
|
118
|
+
t.deepEqual(sleepCalls, [1000, 1000]);
|
|
119
|
+
});
|
|
120
|
+
test('retryOnRateLimit does not retry non-rate-limit errors', async (t) => {
|
|
121
|
+
let attempts = 0;
|
|
122
|
+
const error = await t.throwsAsync(() => retryOnRateLimit({
|
|
123
|
+
operation: async () => {
|
|
124
|
+
attempts++;
|
|
125
|
+
const err = new Error('boom');
|
|
126
|
+
err.status = 500;
|
|
127
|
+
throw err;
|
|
128
|
+
},
|
|
129
|
+
maxRetries: 2,
|
|
130
|
+
sleep: async () => { },
|
|
131
|
+
}));
|
|
132
|
+
t.truthy(error);
|
|
133
|
+
t.is(attempts, 1);
|
|
134
|
+
});
|
|
135
|
+
//# sourceMappingURL=runner-utils.test.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"runner-utils.test.js","sourceRoot":"","sources":["../../../source/scripts/eval-auto-approval/runner-utils.test.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,KAAK,CAAC;AACvB,OAAO,EAAE,WAAW,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AACrD,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EACL,cAAc,EACd,kBAAkB,EAClB,aAAa,EACb,qBAAqB,EACrB,gBAAgB,EAChB,qBAAqB,GACtB,MAAM,mBAAmB,CAAC;AAE3B,IAAI,CAAC,oFAAoF,EAAE,CAAC,CAAC,EAAE,EAAE;IAC/F,CAAC,CAAC,EAAE,CAAC,aAAa,CAAC,iCAAiC,CAAC,EAAE,+BAA+B,CAAC,CAAC;IACxF,CAAC,CAAC,EAAE,CAAC,aAAa,CAAC,4BAA4B,CAAC,EAAE,+BAA+B,CAAC,CAAC;AACrF,CAAC,CAAC,CAAC;AAEH,IAAI,CAAC,0EAA0E,EAAE,CAAC,CAAC,EAAE,EAAE;IACrF,CAAC,CAAC,SAAS,CAAC,GAAG,EAAE,CAAC,qBAAqB,CAAC,EAAE,WAAW,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC;IACxE,CAAC,CAAC,MAAM,CAAC,GAAG,EAAE,CAAC,qBAAqB,CAAC,EAAE,WAAW,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC,EAAE;QACnE,OAAO,EAAE,eAAe;KACzB,CAAC,CAAC;IACH,CAAC,CAAC,MAAM,CAAC,GAAG,EAAE,CAAC,qBAAqB,CAAC,EAAE,WAAW,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC,EAAE;QACnE,OAAO,EAAE,UAAU;KACpB,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,IAAI,CAAC,gFAAgF,EAAE,CAAC,CAAC,EAAE,EAAE;IAC3F,MAAM,GAAG,GAAG,cAAc,CAAC;QACzB,KAAK,EAAE,aAAa;QACpB,QAAQ,EAAE,QAAQ;QAClB,aAAa,EAAE,yBAAyB;QACxC,OAAO,EAAE,IAAI;QACb,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,YAAY,EAAE,CAAC;KACnD,CAAC,CAAC;IAEH,CAAC,CAAC,EAAE,CAAC,GAAG,CAAC,OAAO,EAAE,kBAAkB,CAAC,CAAC;IACtC,CAAC,CAAC,EAAE,CAAC,GAAG,CAAC,aAAa,EAAE,yBAAyB,CAAC,CAAC;AACrD,CAAC,CAAC,CAAC;AAEH,IAAI,CAAC,oDAAoD,EAAE,CAAC,CAAC,EAAE,EAAE;IAC/D,MAAM,IAAI,GAAG;QACX,KAAK,EAAE,aAAa;QACpB,QAAQ,EAAE,QAAQ;QAClB,OAAO,EAAE,IAAI;QACb,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,YAAY,EAAE,CAAC;KACnD,CAAC;IAEF,CAAC,CAAC,YAAY,CACZ,cAAc,CAAC,EAAE,GAAG,IAAI,EAAE,aAAa,EAAE,yBAAyB,EAAE,CAAC,EACrE,cAAc,CAAC,EAAE,GAAG,IAAI,EAAE,aAAa,EAAE,yBAAyB,EAAE,CAAC,CACtE,CAAC;AACJ,CAAC,CAAC,CAAC;AAEH,IAAI,CAAC,+EAA+E,EAAE,CAAC,CAAC,EAAE,EAAE;IAC1F,MAAM,GAAG,GAAG,WAAW,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,aAAa,CAAC,CAAC,CAAC;IACvD,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,aAAa,CAAC,CAAC;IAE1C,aAAa,CACX,QAAQ,EACR;;;;;;CAMH,CACE,CAAC;IAEF,CAAC,CAAC,SAAS,CAAC,qBAAqB,CAAC,QAAQ,CAAC,EAAE;QAC3C,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,QAAQ,EAAE;QACvC,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,aAAa,EAAE;QAC5C,EAAE,QAAQ,EAAE,YAAY,EAAE,KAAK,EAAE,6BAA6B,EAAE;KACjE,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,IAAI,CAAC,8DAA8D,EAAE,CAAC,CAAC,EAAE,EAAE;IACzE,MAAM,GAAG,GAAG,WAAW,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,aAAa,CAAC,CAAC,CAAC;IACvD,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,aAAa,CAAC,CAAC;IAE1C,aAAa,CACX,QAAQ,EACR;;CAEH,CACE,CAAC;IAEF,CAAC,CAAC,MAAM,CAAC,GAAG,EAAE,CAAC,qBAAqB,CAAC,QAAQ,CAAC,EAAE;QAC9C,OAAO,EAAE,8BAA8B;KACxC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,IAAI,CAAC,iFAAiF,EAAE,KAAK,EAAE,CAAC,EAAE,EAAE;IAClG,IAAI,QAAQ,GAAG,CAAC,CAAC;IACjB,MAAM,UAAU,GAAa,EAAE,CAAC;IAChC,MAAM,GAAG,GAAG,iBAAiB,CAAC;IAE9B,MAAM,MAAM,GAAG,MAAM,gBAAgB,CAAC;QACpC,SAAS,EAAE,KAAK,IAAI,EAAE;YACpB,QAAQ,EAAE,CAAC;YACX,IAAI,QAAQ,GAAG,CAAC,EAAE,CAAC;gBACjB,MAAM,GAAG,GAAG,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;gBAC5C,GAAmE,CAAC,MAAM,GAAG,GAAG,CAAC;gBACjF,GAAmE,CAAC,OAAO,GAAG;oBAC7E,mBAAmB,EAAE,MAAM,CAAC,GAAG,GAAG,KAAK,CAAC;iBACzC,CAAC;gBACF,MAAM,GAAG,CAAC;YACZ,CAAC;YAED,OAAO,IAAI,CAAC;QACd,CAAC;QACD,UAAU,EAAE,CAAC;QACb,KAAK,EAAE,KAAK,EAAE,EAAE,EAAE,EAAE;YAClB,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACtB,CAAC;QACD,GAAG,EAAE,GAAG,EAAE,CAAC,GAAG;KACf,CAAC,CAAC;IAEH,CAAC,CAAC,EAAE,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC;IACnB,CAAC,CAAC,EAAE,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC;IAClB,CAAC,CAAC,SAAS,CAAC,UAAU,EAAE,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC,CAAC;AACxC,CAAC,CAAC,CAAC;AAEH,IAAI,CAAC,oDAAoD,EAAE,KAAK,EAAE,CAAC,EAAE,EAAE;IACrE,IAAI,QAAQ,GAAG,CAAC,CAAC;IACjB,MAAM,UAAU,GAAa,EAAE,CAAC;IAEhC,MAAM,KAAK,GAAG,MAAM,CAAC,CAAC,WAAW,CAAC,GAAG,EAAE,CACrC,gBAAgB,CAAC;QACf,SAAS,EAAE,KAAK,IAAI,EAAE;YACpB,QAAQ,EAAE,CAAC;YACX,MAAM,GAAG,GAAG,MAAM,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,mBAAmB,CAAC,EAAE;gBACxD,KAAK,EAAE;oBACL,IAAI,EAAE,GAAG;oBACT,QAAQ,EAAE;wBACR,OAAO,EAAE;4BACP,mBAAmB,EAAE,eAAe;yBACrC;qBACF;iBACF;aACF,CAAC,CAAC;YACH,MAAM,GAAG,CAAC;QACZ,CAAC;QACD,UAAU,EAAE,CAAC;QACb,KAAK,EAAE,KAAK,EAAE,EAAE,EAAE,EAAE;YAClB,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACtB,CAAC;QACD,GAAG,EAAE,GAAG,EAAE,CAAC,iBAAiB;KAC7B,CAAC,CACH,CAAC;IAEF,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IAChB,CAAC,CAAC,EAAE,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC;IAClB,CAAC,CAAC,SAAS,CAAC,UAAU,EAAE,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC,CAAC;AACxC,CAAC,CAAC,CAAC;AAEH,IAAI,CAAC,uDAAuD,EAAE,KAAK,EAAE,CAAC,EAAE,EAAE;IACxE,IAAI,QAAQ,GAAG,CAAC,CAAC;IAEjB,MAAM,KAAK,GAAG,MAAM,CAAC,CAAC,WAAW,CAAC,GAAG,EAAE,CACrC,gBAAgB,CAAC;QACf,SAAS,EAAE,KAAK,IAAI,EAAE;YACpB,QAAQ,EAAE,CAAC;YACX,MAAM,GAAG,GAAG,IAAI,KAAK,CAAC,MAAM,CAAC,CAAC;YAC7B,GAAkC,CAAC,MAAM,GAAG,GAAG,CAAC;YACjD,MAAM,GAAG,CAAC;QACZ,CAAC;QACD,UAAU,EAAE,CAAC;QACb,KAAK,EAAE,KAAK,IAAI,EAAE,GAAE,CAAC;KACtB,CAAC,CACH,CAAC;IAEF,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IAChB,CAAC,CAAC,EAAE,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC;AACpB,CAAC,CAAC,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../../source/scripts/eval-auto-approval/runner.ts"],"names":[],"mappings":""}
|
|
@@ -0,0 +1,326 @@
|
|
|
1
|
+
import meow from 'meow';
|
|
2
|
+
import { loadDataset, filterDataset } from './dataset.js';
|
|
3
|
+
import { createMockSettingsService } from '../../services/settings-service.mock.js';
|
|
4
|
+
import { LoggingService } from '../../services/logging-service.js';
|
|
5
|
+
import { OpenAIAgentClient } from '../../lib/openai-agent-client.js';
|
|
6
|
+
import { evaluateShellAutoApprovalAdvisories, SHELL_AUTO_APPROVAL_PROMPT_VERSION, } from '../../services/shell-auto-approval-evaluator.js';
|
|
7
|
+
import { ResponseCache } from './cache.js';
|
|
8
|
+
import { buildModelLeaderboard, getModelLeaderboardPaths, loadPersistedModelResults, mergeModelResults, saveModelLeaderboardJson, saveModelLeaderboardMarkdown, saveModelResults, } from './leaderboard.js';
|
|
9
|
+
import { computeMetrics } from './metrics.js';
|
|
10
|
+
import { generateReport } from './report.js';
|
|
11
|
+
import { createCacheKey, loadModelRunsFromYaml, retryOnRateLimit, validateRunnerOptions } from './runner-utils.js';
|
|
12
|
+
import { appendFileSync, writeFileSync, mkdirSync, existsSync } from 'node:fs';
|
|
13
|
+
import { join, dirname } from 'node:path';
|
|
14
|
+
import { performance } from 'node:perf_hooks';
|
|
15
|
+
const cli = meow(`
|
|
16
|
+
Usage
|
|
17
|
+
$ node dist/scripts/eval-auto-approval/runner.js [options]
|
|
18
|
+
|
|
19
|
+
Options
|
|
20
|
+
--model Model to evaluate (e.g. gpt-4o)
|
|
21
|
+
--provider Provider to use (openai, openrouter, etc.)
|
|
22
|
+
--dataset Path to dataset JSON (default: eval/auto-approval/dataset.json)
|
|
23
|
+
--models-file Path to YAML file mapping providers to model lists (default: eval/auto-approval/models.yaml)
|
|
24
|
+
--filter-cat Filter by category
|
|
25
|
+
--filter-sev Filter by severity
|
|
26
|
+
--concurrency Max concurrent requests (default: 5)
|
|
27
|
+
--repeat Number of times to repeat each case (default: 1)
|
|
28
|
+
--output Path to output results (default: eval/auto-approval/reports/<date>/results-<timestamp>)
|
|
29
|
+
--report-format Comma-separated list of formats (json,jsonl,md) (default: json)
|
|
30
|
+
--dry-run Print plan without executing
|
|
31
|
+
--yes Skip confirmation prompt
|
|
32
|
+
--no-cache Disable response caching
|
|
33
|
+
--clear-cache Clear the response cache before running
|
|
34
|
+
|
|
35
|
+
Examples
|
|
36
|
+
$ node dist/scripts/eval-auto-approval/runner.js --model gpt-4o --concurrency 10
|
|
37
|
+
`, {
|
|
38
|
+
importMeta: import.meta,
|
|
39
|
+
flags: {
|
|
40
|
+
model: { type: 'string' },
|
|
41
|
+
provider: { type: 'string', default: 'openai' },
|
|
42
|
+
dataset: { type: 'string', default: 'eval/auto-approval/dataset.json' },
|
|
43
|
+
modelsFile: { type: 'string', default: 'eval/auto-approval/models.yaml' },
|
|
44
|
+
filterCat: { type: 'string' },
|
|
45
|
+
filterSev: { type: 'string' },
|
|
46
|
+
concurrency: { type: 'number', default: 5 },
|
|
47
|
+
repeat: { type: 'number', default: 1 },
|
|
48
|
+
output: { type: 'string' },
|
|
49
|
+
reportFormat: { type: 'string', default: 'json' },
|
|
50
|
+
dryRun: { type: 'boolean' },
|
|
51
|
+
yes: { type: 'boolean' },
|
|
52
|
+
cache: { type: 'boolean', default: true },
|
|
53
|
+
clearCache: { type: 'boolean' },
|
|
54
|
+
},
|
|
55
|
+
});
|
|
56
|
+
async function run() {
|
|
57
|
+
const flags = cli.flags;
|
|
58
|
+
if (flags.clearCache) {
|
|
59
|
+
const cache = new ResponseCache('eval/auto-approval/.cache');
|
|
60
|
+
cache.clear();
|
|
61
|
+
console.log('Cache cleared.');
|
|
62
|
+
if (flags.dryRun)
|
|
63
|
+
return;
|
|
64
|
+
}
|
|
65
|
+
if (!process.env['RUN_LIVE_EVAL'] && !flags.dryRun) {
|
|
66
|
+
console.error('Error: RUN_LIVE_EVAL=1 environment variable must be set to run live evaluations.');
|
|
67
|
+
process.exit(1);
|
|
68
|
+
}
|
|
69
|
+
const datasetPath = flags.dataset;
|
|
70
|
+
validateRunnerOptions({
|
|
71
|
+
concurrency: flags.concurrency,
|
|
72
|
+
repeat: flags.repeat,
|
|
73
|
+
});
|
|
74
|
+
const cases = loadDataset(datasetPath);
|
|
75
|
+
const filteredCases = filterDataset(cases, {
|
|
76
|
+
category: flags.filterCat,
|
|
77
|
+
severity: flags.filterSev,
|
|
78
|
+
});
|
|
79
|
+
if (filteredCases.length === 0) {
|
|
80
|
+
console.log('No cases matched the filters.');
|
|
81
|
+
return;
|
|
82
|
+
}
|
|
83
|
+
const modelRuns = flags.model && flags.modelsFile === 'eval/auto-approval/models.yaml' && !existsSync(flags.modelsFile)
|
|
84
|
+
? [{ provider: flags.provider, model: flags.model }]
|
|
85
|
+
: flags.modelsFile && existsSync(flags.modelsFile)
|
|
86
|
+
? loadModelRunsFromYaml(flags.modelsFile)
|
|
87
|
+
: flags.model
|
|
88
|
+
? [{ provider: flags.provider, model: flags.model }]
|
|
89
|
+
: [];
|
|
90
|
+
if (modelRuns.length === 0) {
|
|
91
|
+
if (!flags.model && (!flags.modelsFile || !existsSync(flags.modelsFile))) {
|
|
92
|
+
console.error('Error: Either --model must be specified, or a valid --models-file must exist.');
|
|
93
|
+
process.exit(1);
|
|
94
|
+
}
|
|
95
|
+
console.log('No model runs configured.');
|
|
96
|
+
return;
|
|
97
|
+
}
|
|
98
|
+
const totalRuns = filteredCases.length * flags.repeat * modelRuns.length;
|
|
99
|
+
console.log('--- Eval Plan ---');
|
|
100
|
+
if (flags.modelsFile) {
|
|
101
|
+
console.log(`Models file: ${flags.modelsFile}`);
|
|
102
|
+
console.log('Model Runs:');
|
|
103
|
+
for (const run of modelRuns) {
|
|
104
|
+
console.log(` - ${run.provider}/${run.model}`);
|
|
105
|
+
}
|
|
106
|
+
}
|
|
107
|
+
else {
|
|
108
|
+
console.log(`Model: ${modelRuns[0].model}`);
|
|
109
|
+
console.log(`Provider: ${modelRuns[0].provider}`);
|
|
110
|
+
}
|
|
111
|
+
console.log(`Dataset: ${datasetPath}`);
|
|
112
|
+
console.log(`Total Cases: ${filteredCases.length}`);
|
|
113
|
+
console.log(`Repeats: ${flags.repeat}`);
|
|
114
|
+
console.log(`Total Runs: ${totalRuns}`);
|
|
115
|
+
console.log(`Concurrency: ${flags.concurrency}`);
|
|
116
|
+
console.log(`Cache: ${flags.cache ? 'Enabled' : 'Disabled'}`);
|
|
117
|
+
console.log('-----------------');
|
|
118
|
+
const estTokensPerRun = 1000;
|
|
119
|
+
const estCostPer1K = 0.01; // Conservative average for gpt-4o
|
|
120
|
+
const estTotalCost = (totalRuns * estTokensPerRun * estCostPer1K) / 1000;
|
|
121
|
+
console.log(`Estimated cost: ~$${estTotalCost.toFixed(2)} (very rough estimate)`);
|
|
122
|
+
if (flags.dryRun) {
|
|
123
|
+
console.log('Dry run complete.');
|
|
124
|
+
return;
|
|
125
|
+
}
|
|
126
|
+
if (!flags.yes) {
|
|
127
|
+
const readline = await import('node:readline');
|
|
128
|
+
const rl = readline.createInterface({
|
|
129
|
+
input: process.stdin,
|
|
130
|
+
output: process.stdout,
|
|
131
|
+
});
|
|
132
|
+
const answer = await new Promise((resolve) => rl.question('Continue? (y/N): ', resolve));
|
|
133
|
+
rl.close();
|
|
134
|
+
if (String(answer).toLowerCase() !== 'y') {
|
|
135
|
+
console.log('Aborted.');
|
|
136
|
+
return;
|
|
137
|
+
}
|
|
138
|
+
}
|
|
139
|
+
const date = new Date().toISOString().split('T')[0];
|
|
140
|
+
const timestamp = new Date().toISOString().replace(/[:.]/g, '-');
|
|
141
|
+
const reportFormats = flags.reportFormat.split(',').map((f) => f.trim());
|
|
142
|
+
const reportsRoot = join(dirname(datasetPath), 'reports');
|
|
143
|
+
const reportsDir = join(reportsRoot, date);
|
|
144
|
+
if (!existsSync(reportsDir)) {
|
|
145
|
+
mkdirSync(reportsDir, { recursive: true });
|
|
146
|
+
}
|
|
147
|
+
const baseOutputPath = flags.output || join(reportsDir, `results-${timestamp}`);
|
|
148
|
+
console.log(`Saving results to: ${baseOutputPath}.{${reportFormats.join(',')}}`);
|
|
149
|
+
const logger = new LoggingService({ disableLogging: true });
|
|
150
|
+
const cache = new ResponseCache('eval/auto-approval/.cache');
|
|
151
|
+
const results = [];
|
|
152
|
+
let completed = 0;
|
|
153
|
+
let cacheHits = 0;
|
|
154
|
+
async function runModelEvaluation(run) {
|
|
155
|
+
const settingsService = createMockSettingsService({
|
|
156
|
+
'agent.autoApproveModel': run.model,
|
|
157
|
+
'agent.autoApproveProvider': run.provider,
|
|
158
|
+
'shell.autoApproveMode': 'auto',
|
|
159
|
+
// Pass provider API keys from environment so runners can be initialized
|
|
160
|
+
...(process.env['OPENROUTER_API_KEY'] ? { 'agent.openrouter.apiKey': process.env['OPENROUTER_API_KEY'] } : {}),
|
|
161
|
+
});
|
|
162
|
+
const agentClient = new OpenAIAgentClient({
|
|
163
|
+
model: run.model,
|
|
164
|
+
deps: {
|
|
165
|
+
logger,
|
|
166
|
+
settings: settingsService,
|
|
167
|
+
},
|
|
168
|
+
});
|
|
169
|
+
const queue = [...filteredCases.flatMap((c) => Array.from({ length: flags.repeat }, () => c))];
|
|
170
|
+
const active = [];
|
|
171
|
+
async function runCase(c) {
|
|
172
|
+
const cacheKey = createCacheKey({
|
|
173
|
+
model: run.model,
|
|
174
|
+
provider: run.provider,
|
|
175
|
+
promptVersion: SHELL_AUTO_APPROVAL_PROMPT_VERSION,
|
|
176
|
+
command: c.command,
|
|
177
|
+
history: c.history,
|
|
178
|
+
});
|
|
179
|
+
let cachedAdvisory = flags.cache ? cache.get(cacheKey) : null;
|
|
180
|
+
let result;
|
|
181
|
+
let error;
|
|
182
|
+
let latencyMs = 0;
|
|
183
|
+
if (cachedAdvisory) {
|
|
184
|
+
cacheHits++;
|
|
185
|
+
result = cachedAdvisory;
|
|
186
|
+
}
|
|
187
|
+
else {
|
|
188
|
+
const start = performance.now();
|
|
189
|
+
try {
|
|
190
|
+
const advisories = await retryOnRateLimit({
|
|
191
|
+
operation: () => evaluateShellAutoApprovalAdvisories({
|
|
192
|
+
commands: [{ id: c.id, command: c.command }],
|
|
193
|
+
history: c.history,
|
|
194
|
+
settingsService,
|
|
195
|
+
agentClient,
|
|
196
|
+
logger,
|
|
197
|
+
throwOnError: true,
|
|
198
|
+
}),
|
|
199
|
+
maxRetries: 2,
|
|
200
|
+
onRetry: ({ attempt, retriesRemaining, delayMs, error: retryError }) => {
|
|
201
|
+
console.warn(`Case ${c.id} hit a rate limit (attempt ${attempt}/2). Retrying in ${Math.round(delayMs)}ms${retriesRemaining > 0 ? `; ${retriesRemaining} retry left after this` : ''}.`);
|
|
202
|
+
logger.warn('Eval runner rate-limit retry', {
|
|
203
|
+
caseId: c.id,
|
|
204
|
+
attempt,
|
|
205
|
+
retriesRemaining,
|
|
206
|
+
delayMs,
|
|
207
|
+
error: retryError instanceof Error ? retryError.message : String(retryError),
|
|
208
|
+
});
|
|
209
|
+
},
|
|
210
|
+
});
|
|
211
|
+
const advisory = advisories.get(c.id);
|
|
212
|
+
if (advisory) {
|
|
213
|
+
if (advisory.isError) {
|
|
214
|
+
error = advisory.reasoning;
|
|
215
|
+
}
|
|
216
|
+
else {
|
|
217
|
+
result = {
|
|
218
|
+
predicted: advisory.approved ? 'approve' : 'reject',
|
|
219
|
+
reasoning: advisory.reasoning,
|
|
220
|
+
source: advisory.source,
|
|
221
|
+
};
|
|
222
|
+
if (flags.cache && advisory.source === 'llm') {
|
|
223
|
+
cache.set(cacheKey, result);
|
|
224
|
+
}
|
|
225
|
+
}
|
|
226
|
+
}
|
|
227
|
+
else {
|
|
228
|
+
error = 'No advisory returned for command ID';
|
|
229
|
+
}
|
|
230
|
+
}
|
|
231
|
+
catch (e) {
|
|
232
|
+
error = e.message;
|
|
233
|
+
}
|
|
234
|
+
latencyMs = performance.now() - start;
|
|
235
|
+
}
|
|
236
|
+
const record = {
|
|
237
|
+
caseId: c.id,
|
|
238
|
+
command: c.command,
|
|
239
|
+
expected: c.expected,
|
|
240
|
+
predicted: result?.predicted,
|
|
241
|
+
reasoning: result?.reasoning,
|
|
242
|
+
source: result?.source,
|
|
243
|
+
latencyMs,
|
|
244
|
+
error,
|
|
245
|
+
timestamp: new Date().toISOString(),
|
|
246
|
+
model: run.model,
|
|
247
|
+
provider: run.provider,
|
|
248
|
+
promptVersion: SHELL_AUTO_APPROVAL_PROMPT_VERSION,
|
|
249
|
+
category: c.category,
|
|
250
|
+
severity: c.severity,
|
|
251
|
+
cached: !!cachedAdvisory,
|
|
252
|
+
};
|
|
253
|
+
appendFileSync(baseOutputPath + '.jsonl', JSON.stringify(record) + '\n');
|
|
254
|
+
results.push(record);
|
|
255
|
+
completed++;
|
|
256
|
+
const statusIcon = record.error ? '⚠️' : record.predicted === record.expected ? '✅' : '❌';
|
|
257
|
+
const cacheIcon = record.cached ? ' 🧊' : '';
|
|
258
|
+
console.log(`[${completed}/${totalRuns}] ${run.provider}/${run.model} Case ${c.id}: ${statusIcon}${cacheIcon} (${Math.round(latencyMs)}ms)`);
|
|
259
|
+
}
|
|
260
|
+
while (queue.length > 0 || active.length > 0) {
|
|
261
|
+
while (active.length < flags.concurrency && queue.length > 0) {
|
|
262
|
+
const c = queue.shift();
|
|
263
|
+
const p = runCase(c).then(() => {
|
|
264
|
+
active.splice(active.indexOf(p), 1);
|
|
265
|
+
});
|
|
266
|
+
active.push(p);
|
|
267
|
+
}
|
|
268
|
+
if (active.length > 0) {
|
|
269
|
+
await Promise.race(active);
|
|
270
|
+
}
|
|
271
|
+
}
|
|
272
|
+
}
|
|
273
|
+
for (const run of modelRuns) {
|
|
274
|
+
if (modelRuns.length > 1) {
|
|
275
|
+
console.log(`\n--- Running ${run.provider}/${run.model} ---`);
|
|
276
|
+
}
|
|
277
|
+
await runModelEvaluation(run);
|
|
278
|
+
}
|
|
279
|
+
if (reportFormats.includes('json')) {
|
|
280
|
+
writeFileSync(baseOutputPath + '.json', JSON.stringify(results, null, 2));
|
|
281
|
+
console.log(`JSON results saved to ${baseOutputPath}.json`);
|
|
282
|
+
}
|
|
283
|
+
if (reportFormats.includes('jsonl') && !existsSync(baseOutputPath + '.jsonl')) {
|
|
284
|
+
// Already written incrementally, but ensure file exists if it was empty for some reason
|
|
285
|
+
// or if we want to explicitly move/rename it. Currently it's written as .jsonl always.
|
|
286
|
+
}
|
|
287
|
+
if (reportFormats.includes('md')) {
|
|
288
|
+
const metrics = computeMetrics(results);
|
|
289
|
+
const report = generateReport(results, metrics, { promptVersion: SHELL_AUTO_APPROVAL_PROMPT_VERSION });
|
|
290
|
+
writeFileSync(baseOutputPath + '.md', report);
|
|
291
|
+
console.log(`Markdown report saved to ${baseOutputPath}.md`);
|
|
292
|
+
}
|
|
293
|
+
const leaderboardPaths = getModelLeaderboardPaths(reportsRoot);
|
|
294
|
+
const persistedModelResults = loadPersistedModelResults(reportsRoot);
|
|
295
|
+
const mergedModelResults = mergeModelResults(persistedModelResults, results);
|
|
296
|
+
const modelLeaderboard = buildModelLeaderboard(mergedModelResults, {
|
|
297
|
+
promptVersion: SHELL_AUTO_APPROVAL_PROMPT_VERSION,
|
|
298
|
+
});
|
|
299
|
+
saveModelResults(leaderboardPaths.recordsPath, mergedModelResults);
|
|
300
|
+
saveModelLeaderboardJson(leaderboardPaths.jsonPath, modelLeaderboard, mergedModelResults, {
|
|
301
|
+
promptVersion: SHELL_AUTO_APPROVAL_PROMPT_VERSION,
|
|
302
|
+
});
|
|
303
|
+
saveModelLeaderboardMarkdown(leaderboardPaths.markdownPath, modelLeaderboard, mergedModelResults, {
|
|
304
|
+
promptVersion: SHELL_AUTO_APPROVAL_PROMPT_VERSION,
|
|
305
|
+
});
|
|
306
|
+
console.log(`Model leaderboard updated: ${leaderboardPaths.markdownPath}`);
|
|
307
|
+
// Cleanup .jsonl if it's not a requested format
|
|
308
|
+
if (!reportFormats.includes('jsonl') && existsSync(baseOutputPath + '.jsonl')) {
|
|
309
|
+
const { unlinkSync } = await import('node:fs');
|
|
310
|
+
unlinkSync(baseOutputPath + '.jsonl');
|
|
311
|
+
}
|
|
312
|
+
console.log('\nPruning cache...');
|
|
313
|
+
cache.prune(500);
|
|
314
|
+
console.log(`\nDone!`);
|
|
315
|
+
const metrics = computeMetrics(results);
|
|
316
|
+
console.log(`Cache hits: ${cacheHits}/${totalRuns}`);
|
|
317
|
+
console.log(`Accuracy: ${(metrics.accuracy * 100).toFixed(1)}%`);
|
|
318
|
+
if (modelLeaderboard[0]) {
|
|
319
|
+
console.log(`Top model: ${modelLeaderboard[0].provider}/${modelLeaderboard[0].model} (score ${modelLeaderboard[0].score.toFixed(2)}, ${modelLeaderboard[0].passed}/${modelLeaderboard[0].casesRun} passed)`);
|
|
320
|
+
}
|
|
321
|
+
}
|
|
322
|
+
run().catch((err) => {
|
|
323
|
+
console.error('Fatal error:', err);
|
|
324
|
+
process.exit(1);
|
|
325
|
+
});
|
|
326
|
+
//# sourceMappingURL=runner.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"runner.js","sourceRoot":"","sources":["../../../source/scripts/eval-auto-approval/runner.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,MAAM,CAAC;AACxB,OAAO,EAAE,WAAW,EAAE,aAAa,EAAQ,MAAM,cAAc,CAAC;AAChE,OAAO,EAAE,yBAAyB,EAAE,MAAM,yCAAyC,CAAC;AACpF,OAAO,EAAE,cAAc,EAAE,MAAM,mCAAmC,CAAC;AACnE,OAAO,EAAE,iBAAiB,EAAE,MAAM,kCAAkC,CAAC;AACrE,OAAO,EACL,mCAAmC,EACnC,kCAAkC,GACnC,MAAM,iDAAiD,CAAC;AACzD,OAAO,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAC3C,OAAO,EACL,qBAAqB,EACrB,wBAAwB,EACxB,yBAAyB,EACzB,iBAAiB,EAEjB,wBAAwB,EACxB,4BAA4B,EAC5B,gBAAgB,GACjB,MAAM,kBAAkB,CAAC;AAC1B,OAAO,EAAE,cAAc,EAAE,MAAM,cAAc,CAAC;AAC9C,OAAO,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAC7C,OAAO,EAAE,cAAc,EAAE,qBAAqB,EAAE,gBAAgB,EAAE,qBAAqB,EAAE,MAAM,mBAAmB,CAAC;AACnH,OAAO,EAAE,cAAc,EAAE,aAAa,EAAE,SAAS,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AAC/E,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAC1C,OAAO,EAAE,WAAW,EAAE,MAAM,iBAAiB,CAAC;AAE9C,MAAM,GAAG,GAAG,IAAI,CACd;;;;;;;;;;;;;;;;;;;;;;CAsBD,EACC;IACE,UAAU,EAAE,MAAM,CAAC,IAAI;IACvB,KAAK,EAAE;QACL,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QACzB,QAAQ,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,QAAQ,EAAE;QAC/C,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,iCAAiC,EAAE;QACvE,UAAU,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,gCAAgC,EAAE;QACzE,SAAS,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAC7B,SAAS,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAC7B,WAAW,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,EAAE;QAC3C,MAAM,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,EAAE;QACtC,MAAM,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAC1B,YAAY,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE;QACjD,MAAM,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;QAC3B,GAAG,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;QACxB,KAAK,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,IAAI,EAAE;QACzC,UAAU,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;KAChC;CACF,CACF,CAAC;AAEF,KAAK,UAAU,GAAG;IAChB,MAAM,KAAK,GAAG,GAAG,CAAC,KAAK,CAAC;IAExB,IAAI,KAAK,CAAC,UAAU,EAAE,CAAC;QACrB,MAAM,KAAK,GAAG,IAAI,aAAa,CAAC,2BAA2B,CAAC,CAAC;QAC7D,KAAK,CAAC,KAAK,EAAE,CAAC;QACd,OAAO,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAC;QAC9B,IAAI,KAAK,CAAC,MAAM;YAAE,OAAO;IAC3B,CAAC;IAED,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,eAAe,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC;QACnD,OAAO,CAAC,KAAK,CAAC,kFAAkF,CAAC,CAAC;QAClG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;IAED,MAAM,WAAW,GAAG,KAAK,CAAC,OAAO,CAAC;IAClC,qBAAqB,CAAC;QACpB,WAAW,EAAE,KAAK,CAAC,WAAW;QAC9B,MAAM,EAAE,KAAK,CAAC,MAAM;KACrB,CAAC,CAAC;IAEH,MAAM,KAAK,GAAG,WAAW,CAAC,WAAW,CAAC,CAAC;IACvC,MAAM,aAAa,GAAG,aAAa,CAAC,KAAK,EAAE;QACzC,QAAQ,EAAE,KAAK,CAAC,SAAS;QACzB,QAAQ,EAAE,KAAK,CAAC,SAAS;KAC1B,CAAC,CAAC;IAEH,IAAI,aAAa,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC/B,OAAO,CAAC,GAAG,CAAC,+BAA+B,CAAC,CAAC;QAC7C,OAAO;IACT,CAAC;IAED,MAAM,SAAS,GACb,KAAK,CAAC,KAAK,IAAI,KAAK,CAAC,UAAU,KAAK,gCAAgC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,UAAU,CAAC;QACnG,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC;QACpD,CAAC,CAAC,KAAK,CAAC,UAAU,IAAI,UAAU,CAAC,KAAK,CAAC,UAAU,CAAC;YAClD,CAAC,CAAC,qBAAqB,CAAC,KAAK,CAAC,UAAU,CAAC;YACzC,CAAC,CAAC,KAAK,CAAC,KAAK;gBACb,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC;gBACpD,CAAC,CAAC,EAAE,CAAC;IAET,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC3B,IAAI,CAAC,KAAK,CAAC,KAAK,IAAI,CAAC,CAAC,KAAK,CAAC,UAAU,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,EAAE,CAAC;YACzE,OAAO,CAAC,KAAK,CAAC,+EAA+E,CAAC,CAAC;YAC/F,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QACD,OAAO,CAAC,GAAG,CAAC,2BAA2B,CAAC,CAAC;QACzC,OAAO;IACT,CAAC;IAED,MAAM,SAAS,GAAG,aAAa,CAAC,MAAM,GAAG,KAAK,CAAC,MAAM,GAAG,SAAS,CAAC,MAAM,CAAC;IAEzE,OAAO,CAAC,GAAG,CAAC,mBAAmB,CAAC,CAAC;IACjC,IAAI,KAAK,CAAC,UAAU,EAAE,CAAC;QACrB,OAAO,CAAC,GAAG,CAAC,iBAAiB,KAAK,CAAC,UAAU,EAAE,CAAC,CAAC;QACjD,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC;QAC3B,KAAK,MAAM,GAAG,IAAI,SAAS,EAAE,CAAC;YAC5B,OAAO,CAAC,GAAG,CAAC,OAAO,GAAG,CAAC,QAAQ,IAAI,GAAG,CAAC,KAAK,EAAE,CAAC,CAAC;QAClD,CAAC;IACH,CAAC;SAAM,CAAC;QACN,OAAO,CAAC,GAAG,CAAC,gBAAgB,SAAS,CAAC,CAAC,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC;QAClD,OAAO,CAAC,GAAG,CAAC,gBAAgB,SAAS,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC;IACvD,CAAC;IACD,OAAO,CAAC,GAAG,CAAC,gBAAgB,WAAW,EAAE,CAAC,CAAC;IAC3C,OAAO,CAAC,GAAG,CAAC,gBAAgB,aAAa,CAAC,MAAM,EAAE,CAAC,CAAC;IACpD,OAAO,CAAC,GAAG,CAAC,gBAAgB,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC;IAC5C,OAAO,CAAC,GAAG,CAAC,gBAAgB,SAAS,EAAE,CAAC,CAAC;IACzC,OAAO,CAAC,GAAG,CAAC,gBAAgB,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC;IACjD,OAAO,CAAC,GAAG,CAAC,gBAAgB,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC;IACpE,OAAO,CAAC,GAAG,CAAC,mBAAmB,CAAC,CAAC;IAEjC,MAAM,eAAe,GAAG,IAAI,CAAC;IAC7B,MAAM,YAAY,GAAG,IAAI,CAAC,CAAC,kCAAkC;IAC7D,MAAM,YAAY,GAAG,CAAC,SAAS,GAAG,eAAe,GAAG,YAAY,CAAC,GAAG,IAAI,CAAC;IAEzE,OAAO,CAAC,GAAG,CAAC,qBAAqB,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,wBAAwB,CAAC,CAAC;IAElF,IAAI,KAAK,CAAC,MAAM,EAAE,CAAC;QACjB,OAAO,CAAC,GAAG,CAAC,mBAAmB,CAAC,CAAC;QACjC,OAAO;IACT,CAAC;IAED,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC;QACf,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,eAAe,CAAC,CAAC;QAC/C,MAAM,EAAE,GAAG,QAAQ,CAAC,eAAe,CAAC;YAClC,KAAK,EAAE,OAAO,CAAC,KAAK;YACpB,MAAM,EAAE,OAAO,CAAC,MAAM;SACvB,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,MAAM,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,EAAE,CAAC,QAAQ,CAAC,mBAAmB,EAAE,OAAO,CAAC,CAAC,CAAC;QACzF,EAAE,CAAC,KAAK,EAAE,CAAC;QAEX,IAAI,MAAM,CAAC,MAAM,CAAC,CAAC,WAAW,EAAE,KAAK,GAAG,EAAE,CAAC;YACzC,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC;YACxB,OAAO;QACT,CAAC;IACH,CAAC;IAED,MAAM,IAAI,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;IACpD,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC,OAAO,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC;IACjE,MAAM,aAAa,GAAI,KAAK,CAAC,YAAuB,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;IACrF,MAAM,WAAW,GAAG,IAAI,CAAC,OAAO,CAAC,WAAW,CAAC,EAAE,SAAS,CAAC,CAAC;IAC1D,MAAM,UAAU,GAAG,IAAI,CAAC,WAAW,EAAE,IAAK,CAAC,CAAC;IAE5C,IAAI,CAAC,UAAU,CAAC,UAAU,CAAC,EAAE,CAAC;QAC5B,SAAS,CAAC,UAAU,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAC7C,CAAC;IAED,MAAM,cAAc,GAAG,KAAK,CAAC,MAAM,IAAI,IAAI,CAAC,UAAU,EAAE,WAAW,SAAS,EAAE,CAAC,CAAC;IAEhF,OAAO,CAAC,GAAG,CAAC,sBAAsB,cAAc,KAAK,aAAa,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IAEjF,MAAM,MAAM,GAAG,IAAI,cAAc,CAAC,EAAE,cAAc,EAAE,IAAI,EAAE,CAAC,CAAC;IAE5D,MAAM,KAAK,GAAG,IAAI,aAAa,CAAC,2BAA2B,CAAC,CAAC;IAC7D,MAAM,OAAO,GAAwB,EAAE,CAAC;IAExC,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,SAAS,GAAG,CAAC,CAAC;IAElB,KAAK,UAAU,kBAAkB,CAAC,GAAwC;QACxE,MAAM,eAAe,GAAG,yBAAyB,CAAC;YAChD,wBAAwB,EAAE,GAAG,CAAC,KAAK;YACnC,2BAA2B,EAAE,GAAG,CAAC,QAAQ;YACzC,uBAAuB,EAAE,MAAM;YAC/B,wEAAwE;YACxE,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC,CAAC,CAAC,CAAC,EAAE,yBAAyB,EAAE,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAC/G,CAAC,CAAC;QAEH,MAAM,WAAW,GAAG,IAAI,iBAAiB,CAAC;YACxC,KAAK,EAAE,GAAG,CAAC,KAAK;YAChB,IAAI,EAAE;gBACJ,MAAM;gBACN,QAAQ,EAAE,eAAe;aAC1B;SACF,CAAC,CAAC;QAEH,MAAM,KAAK,GAAG,CAAC,GAAG,aAAa,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QAC/F,MAAM,MAAM,GAAoB,EAAE,CAAC;QAEnC,KAAK,UAAU,OAAO,CAAC,CAAO;YAC5B,MAAM,QAAQ,GAAG,cAAc,CAAC;gBAC9B,KAAK,EAAE,GAAG,CAAC,KAAK;gBAChB,QAAQ,EAAE,GAAG,CAAC,QAAQ;gBACtB,aAAa,EAAE,kCAAkC;gBACjD,OAAO,EAAE,CAAC,CAAC,OAAO;gBAClB,OAAO,EAAE,CAAC,CAAC,OAAO;aACnB,CAAC,CAAC;YAEH,IAAI,cAAc,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;YAC9D,IAAI,MAAW,CAAC;YAChB,IAAI,KAAyB,CAAC;YAC9B,IAAI,SAAS,GAAG,CAAC,CAAC;YAElB,IAAI,cAAc,EAAE,CAAC;gBACnB,SAAS,EAAE,CAAC;gBACZ,MAAM,GAAG,cAAc,CAAC;YAC1B,CAAC;iBAAM,CAAC;gBACN,MAAM,KAAK,GAAG,WAAW,CAAC,GAAG,EAAE,CAAC;gBAChC,IAAI,CAAC;oBACH,MAAM,UAAU,GAAG,MAAM,gBAAgB,CAAC;wBACxC,SAAS,EAAE,GAAG,EAAE,CACd,mCAAmC,CAAC;4BAClC,QAAQ,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC,EAAE,EAAE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC;4BAC5C,OAAO,EAAE,CAAC,CAAC,OAAc;4BACzB,eAAe;4BACf,WAAW;4BACX,MAAM;4BACN,YAAY,EAAE,IAAI;yBACnB,CAAC;wBACJ,UAAU,EAAE,CAAC;wBACb,OAAO,EAAE,CAAC,EAAE,OAAO,EAAE,gBAAgB,EAAE,OAAO,EAAE,KAAK,EAAE,UAAU,EAAE,EAAE,EAAE;4BACrE,OAAO,CAAC,IAAI,CACV,QAAQ,CAAC,CAAC,EAAE,8BAA8B,OAAO,oBAAoB,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KACtF,gBAAgB,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,gBAAgB,wBAAwB,CAAC,CAAC,CAAC,EACzE,GAAG,CACJ,CAAC;4BACF,MAAM,CAAC,IAAI,CAAC,8BAA8B,EAAE;gCAC1C,MAAM,EAAE,CAAC,CAAC,EAAE;gCACZ,OAAO;gCACP,gBAAgB;gCAChB,OAAO;gCACP,KAAK,EAAE,UAAU,YAAY,KAAK,CAAC,CAAC,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,UAAU,CAAC;6BAC7E,CAAC,CAAC;wBACL,CAAC;qBACF,CAAC,CAAC;oBAEH,MAAM,QAAQ,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;oBACtC,IAAI,QAAQ,EAAE,CAAC;wBACb,IAAI,QAAQ,CAAC,OAAO,EAAE,CAAC;4BACrB,KAAK,GAAG,QAAQ,CAAC,SAAS,CAAC;wBAC7B,CAAC;6BAAM,CAAC;4BACN,MAAM,GAAG;gCACP,SAAS,EAAE,QAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,QAAQ;gCACnD,SAAS,EAAE,QAAQ,CAAC,SAAS;gCAC7B,MAAM,EAAE,QAAQ,CAAC,MAAM;6BACxB,CAAC;4BACF,IAAI,KAAK,CAAC,KAAK,IAAI,QAAQ,CAAC,MAAM,KAAK,KAAK,EAAE,CAAC;gCAC7C,KAAK,CAAC,GAAG,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;4BAC9B,CAAC;wBACH,CAAC;oBACH,CAAC;yBAAM,CAAC;wBACN,KAAK,GAAG,qCAAqC,CAAC;oBAChD,CAAC;gBACH,CAAC;gBAAC,OAAO,CAAM,EAAE,CAAC;oBAChB,KAAK,GAAG,CAAC,CAAC,OAAO,CAAC;gBACpB,CAAC;gBACD,SAAS,GAAG,WAAW,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC;YACxC,CAAC;YAED,MAAM,MAAM,GAAsB;gBAChC,MAAM,EAAE,CAAC,CAAC,EAAE;gBACZ,OAAO,EAAE,CAAC,CAAC,OAAO;gBAClB,QAAQ,EAAE,CAAC,CAAC,QAAQ;gBACpB,SAAS,EAAE,MAAM,EAAE,SAAS;gBAC5B,SAAS,EAAE,MAAM,EAAE,SAAS;gBAC5B,MAAM,EAAE,MAAM,EAAE,MAAM;gBACtB,SAAS;gBACT,KAAK;gBACL,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;gBACnC,KAAK,EAAE,GAAG,CAAC,KAAK;gBAChB,QAAQ,EAAE,GAAG,CAAC,QAAQ;gBACtB,aAAa,EAAE,kCAAkC;gBACjD,QAAQ,EAAE,CAAC,CAAC,QAAQ;gBACpB,QAAQ,EAAE,CAAC,CAAC,QAAQ;gBACpB,MAAM,EAAE,CAAC,CAAC,cAAc;aACzB,CAAC;YAEF,cAAc,CAAC,cAAc,GAAG,QAAQ,EAAE,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,GAAG,IAAI,CAAC,CAAC;YACzE,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;YACrB,SAAS,EAAE,CAAC;YACZ,MAAM,UAAU,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,SAAS,KAAK,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;YAC1F,MAAM,SAAS,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC;YAC7C,OAAO,CAAC,GAAG,CACT,IAAI,SAAS,IAAI,SAAS,KAAK,GAAG,CAAC,QAAQ,IAAI,GAAG,CAAC,KAAK,SAAS,CAAC,CAAC,EAAE,KAAK,UAAU,GAAG,SAAS,KAAK,IAAI,CAAC,KAAK,CAC7G,SAAS,CACV,KAAK,CACP,CAAC;QACJ,CAAC;QAED,OAAO,KAAK,CAAC,MAAM,GAAG,CAAC,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC7C,OAAO,MAAM,CAAC,MAAM,GAAG,KAAK,CAAC,WAAW,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBAC7D,MAAM,CAAC,GAAG,KAAK,CAAC,KAAK,EAAG,CAAC;gBACzB,MAAM,CAAC,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE;oBAC7B,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;gBACtC,CAAC,CAAC,CAAC;gBACH,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YACjB,CAAC;YACD,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBACtB,MAAM,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;YAC7B,CAAC;QACH,CAAC;IACH,CAAC;IAED,KAAK,MAAM,GAAG,IAAI,SAAS,EAAE,CAAC;QAC5B,IAAI,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACzB,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,CAAC,QAAQ,IAAI,GAAG,CAAC,KAAK,MAAM,CAAC,CAAC;QAChE,CAAC;QACD,MAAM,kBAAkB,CAAC,GAAG,CAAC,CAAC;IAChC,CAAC;IAED,IAAI,aAAa,CAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC;QACnC,aAAa,CAAC,cAAc,GAAG,OAAO,EAAE,IAAI,CAAC,SAAS,CAAC,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;QAC1E,OAAO,CAAC,GAAG,CAAC,yBAAyB,cAAc,OAAO,CAAC,CAAC;IAC9D,CAAC;IAED,IAAI,aAAa,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,UAAU,CAAC,cAAc,GAAG,QAAQ,CAAC,EAAE,CAAC;QAC9E,wFAAwF;QACxF,uFAAuF;IACzF,CAAC;IAED,IAAI,aAAa,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC;QACjC,MAAM,OAAO,GAAG,cAAc,CAAC,OAAO,CAAC,CAAC;QACxC,MAAM,MAAM,GAAG,cAAc,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE,aAAa,EAAE,kCAAkC,EAAE,CAAC,CAAC;QACvG,aAAa,CAAC,cAAc,GAAG,KAAK,EAAE,MAAM,CAAC,CAAC;QAC9C,OAAO,CAAC,GAAG,CAAC,4BAA4B,cAAc,KAAK,CAAC,CAAC;IAC/D,CAAC;IAED,MAAM,gBAAgB,GAAG,wBAAwB,CAAC,WAAW,CAAC,CAAC;IAC/D,MAAM,qBAAqB,GAAG,yBAAyB,CAAC,WAAW,CAAC,CAAC;IACrE,MAAM,kBAAkB,GAAG,iBAAiB,CAAC,qBAAqB,EAAE,OAAO,CAAC,CAAC;IAC7E,MAAM,gBAAgB,GAAG,qBAAqB,CAAC,kBAAkB,EAAE;QACjE,aAAa,EAAE,kCAAkC;KAClD,CAAC,CAAC;IAEH,gBAAgB,CAAC,gBAAgB,CAAC,WAAW,EAAE,kBAAkB,CAAC,CAAC;IACnE,wBAAwB,CAAC,gBAAgB,CAAC,QAAQ,EAAE,gBAAgB,EAAE,kBAAkB,EAAE;QACxF,aAAa,EAAE,kCAAkC;KAClD,CAAC,CAAC;IACH,4BAA4B,CAAC,gBAAgB,CAAC,YAAY,EAAE,gBAAgB,EAAE,kBAAkB,EAAE;QAChG,aAAa,EAAE,kCAAkC;KAClD,CAAC,CAAC;IAEH,OAAO,CAAC,GAAG,CAAC,8BAA8B,gBAAgB,CAAC,YAAY,EAAE,CAAC,CAAC;IAE3E,gDAAgD;IAChD,IAAI,CAAC,aAAa,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,UAAU,CAAC,cAAc,GAAG,QAAQ,CAAC,EAAE,CAAC;QAC9E,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,SAAS,CAAC,CAAC;QAC/C,UAAU,CAAC,cAAc,GAAG,QAAQ,CAAC,CAAC;IACxC,CAAC;IAED,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC,CAAC;IAClC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IAEjB,OAAO,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;IACvB,MAAM,OAAO,GAAG,cAAc,CAAC,OAAO,CAAC,CAAC;IACxC,OAAO,CAAC,GAAG,CAAC,eAAe,SAAS,IAAI,SAAS,EAAE,CAAC,CAAC;IACrD,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,OAAO,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;IACjE,IAAI,gBAAgB,CAAC,CAAC,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,GAAG,CACT,cAAc,gBAAgB,CAAC,CAAC,CAAC,CAAC,QAAQ,IACxC,gBAAgB,CAAC,CAAC,CAAC,CAAC,KACtB,WAAW,gBAAgB,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,gBAAgB,CAAC,CAAC,CAAC,CAAC,MAAM,IAC5E,gBAAgB,CAAC,CAAC,CAAC,CAAC,QACtB,UAAU,CACX,CAAC;IACJ,CAAC;AACH,CAAC;AAED,GAAG,EAAE,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE;IAClB,OAAO,CAAC,KAAK,CAAC,cAAc,EAAE,GAAG,CAAC,CAAC;IACnC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAClB,CAAC,CAAC,CAAC"}
|
|
@@ -0,0 +1,11 @@
|
|
|
1
|
+
export type AgentStream = AsyncIterable<unknown> & {
|
|
2
|
+
completed: Promise<unknown>;
|
|
3
|
+
rawResponses?: unknown[];
|
|
4
|
+
lastResponseId?: string | null;
|
|
5
|
+
interruptions?: unknown[];
|
|
6
|
+
state?: unknown;
|
|
7
|
+
newItems?: unknown[];
|
|
8
|
+
history?: unknown[];
|
|
9
|
+
finalOutput?: string;
|
|
10
|
+
};
|
|
11
|
+
//# sourceMappingURL=agent-stream.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agent-stream.d.ts","sourceRoot":"","sources":["../../source/services/agent-stream.ts"],"names":[],"mappings":"AAAA,MAAM,MAAM,WAAW,GAAG,aAAa,CAAC,OAAO,CAAC,GAAG;IACjD,SAAS,EAAE,OAAO,CAAC,OAAO,CAAC,CAAC;IAC5B,YAAY,CAAC,EAAE,OAAO,EAAE,CAAC;IACzB,cAAc,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,aAAa,CAAC,EAAE,OAAO,EAAE,CAAC;IAC1B,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,QAAQ,CAAC,EAAE,OAAO,EAAE,CAAC;IACrB,OAAO,CAAC,EAAE,OAAO,EAAE,CAAC;IACpB,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agent-stream.js","sourceRoot":"","sources":["../../source/services/agent-stream.ts"],"names":[],"mappings":""}
|
|
@@ -0,0 +1,43 @@
|
|
|
1
|
+
import type { OpenAIAgentClient } from '../lib/openai-agent-client.js';
|
|
2
|
+
import type { ILoggingService } from './service-interfaces.js';
|
|
3
|
+
import type { ConversationEvent } from './conversation-events.js';
|
|
4
|
+
import { ApprovalState, type AbortedApprovalContext, type PendingApprovalContext } from './approval-state.js';
|
|
5
|
+
export interface ApprovalFlowCoordinatorDeps {
|
|
6
|
+
agentClient: OpenAIAgentClient;
|
|
7
|
+
approvalState: ApprovalState;
|
|
8
|
+
logger: ILoggingService;
|
|
9
|
+
sessionId: string;
|
|
10
|
+
}
|
|
11
|
+
export interface AbortResolutionPlan {
|
|
12
|
+
abortedContext: AbortedApprovalContext;
|
|
13
|
+
/** Cleanup that removes the rejection interceptor; always safe to call. */
|
|
14
|
+
removeInterceptor: () => void;
|
|
15
|
+
}
|
|
16
|
+
export interface ContinuationPlan {
|
|
17
|
+
pendingApprovalContext: PendingApprovalContext;
|
|
18
|
+
toolStartedEvent?: ConversationEvent;
|
|
19
|
+
/** Cleanup that removes the rejection interceptor; always safe to call. */
|
|
20
|
+
removeInterceptor: () => void;
|
|
21
|
+
}
|
|
22
|
+
export declare class ApprovalFlowCoordinator {
|
|
23
|
+
private readonly deps;
|
|
24
|
+
constructor(deps: ApprovalFlowCoordinatorDeps);
|
|
25
|
+
abort(): boolean;
|
|
26
|
+
consumeAborted(): AbortedApprovalContext | null;
|
|
27
|
+
/**
|
|
28
|
+
* Drive an aborted-approval resolution: install rejection interceptor and approve the
|
|
29
|
+
* interruption so the agent gets the rejection text via the interceptor.
|
|
30
|
+
*/
|
|
31
|
+
prepareAbortResolution(abortedContext: AbortedApprovalContext, userText: string): AbortResolutionPlan;
|
|
32
|
+
/**
|
|
33
|
+
* Prepare for a continuation after the user makes an approval decision.
|
|
34
|
+
* Returns null if there is no pending approval. Caller is responsible for invoking
|
|
35
|
+
* `removeInterceptor` in a finally block once the continuation stream completes.
|
|
36
|
+
*/
|
|
37
|
+
prepareContinuation(answer: string, rejectionReason: string | undefined): ContinuationPlan | null;
|
|
38
|
+
recordPending(pending: PendingApprovalContext): void;
|
|
39
|
+
clearPending(): void;
|
|
40
|
+
getPending(): PendingApprovalContext | null;
|
|
41
|
+
getPendingInterruption(): unknown;
|
|
42
|
+
}
|
|
43
|
+
//# sourceMappingURL=approval-flow-coordinator.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"approval-flow-coordinator.d.ts","sourceRoot":"","sources":["../../source/services/approval-flow-coordinator.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,+BAA+B,CAAC;AACvE,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,yBAAyB,CAAC;AAC/D,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,0BAA0B,CAAC;AAClE,OAAO,EAAE,aAAa,EAAE,KAAK,sBAAsB,EAAE,KAAK,sBAAsB,EAAE,MAAM,qBAAqB,CAAC;AAe9G,MAAM,WAAW,2BAA2B;IAC1C,WAAW,EAAE,iBAAiB,CAAC;IAC/B,aAAa,EAAE,aAAa,CAAC;IAC7B,MAAM,EAAE,eAAe,CAAC;IACxB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,mBAAmB;IAClC,cAAc,EAAE,sBAAsB,CAAC;IACvC,2EAA2E;IAC3E,iBAAiB,EAAE,MAAM,IAAI,CAAC;CAC/B;AAED,MAAM,WAAW,gBAAgB;IAC/B,sBAAsB,EAAE,sBAAsB,CAAC;IAC/C,gBAAgB,CAAC,EAAE,iBAAiB,CAAC;IACrC,2EAA2E;IAC3E,iBAAiB,EAAE,MAAM,IAAI,CAAC;CAC/B;AAED,qBAAa,uBAAuB;IACtB,OAAO,CAAC,QAAQ,CAAC,IAAI;gBAAJ,IAAI,EAAE,2BAA2B;IAE9D,KAAK,IAAI,OAAO;IAehB,cAAc,IAAI,sBAAsB,GAAG,IAAI;IAI/C;;;OAGG;IACH,sBAAsB,CAAC,cAAc,EAAE,sBAAsB,EAAE,QAAQ,EAAE,MAAM,GAAG,mBAAmB;IAkBrG;;;;OAIG;IACH,mBAAmB,CAAC,MAAM,EAAE,MAAM,EAAE,eAAe,EAAE,MAAM,GAAG,SAAS,GAAG,gBAAgB,GAAG,IAAI;IAoEjG,aAAa,CAAC,OAAO,EAAE,sBAAsB,GAAG,IAAI;IAIpD,YAAY,IAAI,IAAI;IAIpB,UAAU,IAAI,sBAAsB,GAAG,IAAI;IAI3C,sBAAsB,IAAI,OAAO;CAGlC"}
|