create-walle 0.9.21 → 0.9.23
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 +27 -5
- package/package.json +2 -2
- package/template/CLAUDE.md +2 -2
- package/template/LICENSE +1 -1
- package/template/bin/ctm-dev-cleanup.js +24 -3
- package/template/bin/ctm-launch.sh +13 -0
- package/template/bin/dev.sh +156 -18
- package/template/bin/node-bin.sh +84 -0
- package/template/bin/pin-node.sh +51 -0
- package/template/claude-task-manager/api-prompts.js +1203 -182
- package/template/claude-task-manager/api-reviews.js +109 -15
- package/template/claude-task-manager/approval-agent.js +1360 -280
- package/template/claude-task-manager/bin/restart-ctm.sh +64 -23
- package/template/claude-task-manager/bin/storage-migration-supervisor.js +338 -0
- package/template/claude-task-manager/db.js +4417 -295
- package/template/claude-task-manager/docs/app-update-refresh-protocol.md +69 -0
- package/template/claude-task-manager/docs/approval-ai-refinement.md +138 -0
- package/template/claude-task-manager/docs/approval-rescue-loop.md +74 -0
- package/template/claude-task-manager/docs/codex-operational-warning-health.md +107 -0
- package/template/claude-task-manager/docs/codex-resume-state-guard-design.md +17 -12
- package/template/claude-task-manager/docs/codex-terminal-render-controller-handoff.md +311 -0
- package/template/claude-task-manager/docs/coding-agent-hooks-architecture.md +418 -0
- package/template/claude-task-manager/docs/conversation-import-freshness.md +20 -0
- package/template/claude-task-manager/docs/google-workspace-auth-health.md +77 -0
- package/template/claude-task-manager/docs/image-paste-ux.md +13 -0
- package/template/claude-task-manager/docs/ipad-web-preview.md +88 -0
- package/template/claude-task-manager/docs/main-loop-offload-architecture.md +66 -0
- package/template/claude-task-manager/docs/microsoft-dev-tunnel-phone-access-design.md +274 -519
- package/template/claude-task-manager/docs/mobile-live-streaming.md +27 -5
- package/template/claude-task-manager/docs/mobile-remote-submission-lifecycle.md +69 -0
- package/template/claude-task-manager/docs/phone-access-design.md +53 -15
- package/template/claude-task-manager/docs/phone-passkey-identity.md +122 -0
- package/template/claude-task-manager/docs/phone-setup.md +3 -0
- package/template/claude-task-manager/docs/prompt-editing-tree-design.md +25 -1
- package/template/claude-task-manager/docs/remote-desktop-access-design.md +268 -0
- package/template/claude-task-manager/docs/restart-lifecycle-architecture.md +95 -0
- package/template/claude-task-manager/docs/runtime-work-control-plane.md +53 -0
- package/template/claude-task-manager/docs/session-interactive-wait-surfaces.md +38 -0
- package/template/claude-task-manager/docs/session-needs-you-dismissal.md +84 -0
- package/template/claude-task-manager/docs/session-render-state-management-design.md +91 -3
- package/template/claude-task-manager/docs/session-standup-command-center-design.md +25 -1
- package/template/claude-task-manager/docs/session-title-authority.md +32 -0
- package/template/claude-task-manager/docs/session-workspace-binding.md +33 -0
- package/template/claude-task-manager/docs/skill-intent-resolution-design.md +72 -0
- package/template/claude-task-manager/docs/walle-mcp-supervisor-health.md +86 -0
- package/template/claude-task-manager/docs/walle-relay-phone-access-design.md +24 -15
- package/template/claude-task-manager/docs/walle-session-history-hydration.md +114 -0
- package/template/claude-task-manager/docs/walle-session-input-queue.md +104 -0
- package/template/claude-task-manager/docs/walle-session-model-catalog.md +90 -0
- package/template/claude-task-manager/docs/walle-session-model-preferences.md +15 -6
- package/template/claude-task-manager/git-utils.js +897 -27
- package/template/claude-task-manager/lib/agent-capabilities.js +33 -0
- package/template/claude-task-manager/lib/agent-cli-cache.js +37 -7
- package/template/claude-task-manager/lib/agent-hooks-installer.js +26 -2
- package/template/claude-task-manager/lib/agent-presets.js +17 -1
- package/template/claude-task-manager/lib/all-sessions-query.js +108 -0
- package/template/claude-task-manager/lib/approval-ai-refinement.js +488 -0
- package/template/claude-task-manager/lib/approval-self-adapt.js +168 -0
- package/template/claude-task-manager/lib/async-semaphore.js +44 -0
- package/template/claude-task-manager/lib/auth-context.js +5 -0
- package/template/claude-task-manager/lib/auth-rate-limit.js +47 -4
- package/template/claude-task-manager/lib/auth-rules.js +29 -2
- package/template/claude-task-manager/lib/auto-approval-verifier.js +129 -16
- package/template/claude-task-manager/lib/background-llm.js +144 -17
- package/template/claude-task-manager/lib/branch-inventory.js +212 -0
- package/template/claude-task-manager/lib/claude-desktop-sessions.js +15 -3
- package/template/claude-task-manager/lib/coalesce-sync-frames.js +151 -0
- package/template/claude-task-manager/lib/codex-launch-health.js +762 -0
- package/template/claude-task-manager/lib/codex-transcript-pager.js +51 -0
- package/template/claude-task-manager/lib/codex-zst.js +124 -0
- package/template/claude-task-manager/lib/coding-agent-models.js +233 -30
- package/template/claude-task-manager/lib/connection-health.js +232 -0
- package/template/claude-task-manager/lib/conversation-blob-parser.js +42 -0
- package/template/claude-task-manager/lib/conversation-tail-merge.js +89 -26
- package/template/claude-task-manager/lib/ctm-session-context-api.js +39 -10
- package/template/claude-task-manager/lib/cursor-conversation-store.js +354 -0
- package/template/claude-task-manager/lib/db-owner-worker-client.js +315 -0
- package/template/claude-task-manager/lib/document-review.js +141 -6
- package/template/claude-task-manager/lib/escalation-review.js +152 -0
- package/template/claude-task-manager/lib/graceful-shutdown.js +159 -0
- package/template/claude-task-manager/lib/headless-term-service.js +678 -0
- package/template/claude-task-manager/lib/heavy-worker-fallback.js +38 -0
- package/template/claude-task-manager/lib/jsonl-conversation-parser.js +542 -0
- package/template/claude-task-manager/lib/jsonl-range-reader.js +112 -0
- package/template/claude-task-manager/lib/main-db-census.js +216 -0
- package/template/claude-task-manager/lib/message-pagination.js +106 -4
- package/template/claude-task-manager/lib/microsoft-dev-tunnel-setup.js +750 -26
- package/template/claude-task-manager/lib/mobile-auth-api.js +274 -7
- package/template/claude-task-manager/lib/mobile-auth-store.js +592 -10
- package/template/claude-task-manager/lib/mobile-notification-dispatcher.js +15 -0
- package/template/claude-task-manager/lib/model-overview-brain-fallback.js +311 -0
- package/template/claude-task-manager/lib/model-overview-cache.js +141 -0
- package/template/claude-task-manager/lib/models-health-routing-notice.js +126 -0
- package/template/claude-task-manager/lib/node-pin-guard.js +93 -0
- package/template/claude-task-manager/lib/perf-tracker.js +242 -6
- package/template/claude-task-manager/lib/permission-match.js +76 -0
- package/template/claude-task-manager/lib/permission-sync.js +133 -20
- package/template/claude-task-manager/lib/process-title.js +35 -0
- package/template/claude-task-manager/lib/prompt-executions-query.js +25 -0
- package/template/claude-task-manager/lib/prompt-index-disk-cache.js +44 -0
- package/template/claude-task-manager/lib/prompt-intent.js +132 -0
- package/template/claude-task-manager/lib/provider-user-context.js +34 -0
- package/template/claude-task-manager/lib/read-pool-client.js +313 -0
- package/template/claude-task-manager/lib/readpool-breaker.js +31 -0
- package/template/claude-task-manager/lib/recent-sessions-breaker.js +12 -0
- package/template/claude-task-manager/lib/remote-feedback-client.js +72 -0
- package/template/claude-task-manager/lib/remote-relay-protocol.js +37 -4
- package/template/claude-task-manager/lib/remote-relay-store.js +159 -0
- package/template/claude-task-manager/lib/remote-submission-observer.js +278 -0
- package/template/claude-task-manager/lib/restart-guard.js +109 -0
- package/template/claude-task-manager/lib/restore-interruption-detector.js +439 -0
- package/template/claude-task-manager/lib/restore-policy.js +13 -0
- package/template/claude-task-manager/lib/restore-resume-batch.js +74 -0
- package/template/claude-task-manager/lib/restore-runtime.js +68 -0
- package/template/claude-task-manager/lib/restore-storm.js +34 -0
- package/template/claude-task-manager/lib/resume-cwd.js +36 -0
- package/template/claude-task-manager/lib/resume-preflight.js +313 -0
- package/template/claude-task-manager/lib/runtime-work-registry.js +444 -0
- package/template/claude-task-manager/lib/sanitize-openai-auth.js +31 -0
- package/template/claude-task-manager/lib/scheduler.js +21 -1
- package/template/claude-task-manager/lib/scrollback-snapshot-store.js +159 -0
- package/template/claude-task-manager/lib/serial-task-queue.js +64 -0
- package/template/claude-task-manager/lib/server-listeners.js +239 -0
- package/template/claude-task-manager/lib/session-capture.js +42 -7
- package/template/claude-task-manager/lib/session-content-backfill.js +131 -0
- package/template/claude-task-manager/lib/session-history.js +388 -43
- package/template/claude-task-manager/lib/session-host-manager.js +287 -0
- package/template/claude-task-manager/lib/session-image-refs.js +209 -0
- package/template/claude-task-manager/lib/session-jobs.js +399 -59
- package/template/claude-task-manager/lib/session-prompt-index.js +137 -0
- package/template/claude-task-manager/lib/session-restore.js +53 -0
- package/template/claude-task-manager/lib/session-standup.js +123 -23
- package/template/claude-task-manager/lib/session-state-bus.js +14 -0
- package/template/claude-task-manager/lib/session-stream.js +64 -16
- package/template/claude-task-manager/lib/session-timeline-summary.js +260 -0
- package/template/claude-task-manager/lib/session-token-usage.js +494 -0
- package/template/claude-task-manager/lib/session-workspace-binding.js +356 -0
- package/template/claude-task-manager/lib/setup-network-config.js +9 -0
- package/template/claude-task-manager/lib/size-cap.js +45 -0
- package/template/claude-task-manager/lib/size-cap.test.js +62 -0
- package/template/claude-task-manager/lib/skill-autocomplete.js +180 -1
- package/template/claude-task-manager/lib/skill-intent-resolver.js +304 -0
- package/template/claude-task-manager/lib/sqlite-driver.js +19 -3
- package/template/claude-task-manager/lib/standup-attention.js +7 -3
- package/template/claude-task-manager/lib/status-authority.js +39 -0
- package/template/claude-task-manager/lib/status-hooks.js +4 -0
- package/template/claude-task-manager/lib/storage-migration.js +235 -0
- package/template/claude-task-manager/lib/structured-capture.js +298 -0
- package/template/claude-task-manager/lib/sync-io-census.js +163 -0
- package/template/claude-task-manager/lib/tailscale-setup.js +6 -0
- package/template/claude-task-manager/lib/terminal-activity-evidence.js +33 -0
- package/template/claude-task-manager/lib/terminal-choice.js +364 -0
- package/template/claude-task-manager/lib/terminal-control-sanitize.js +17 -0
- package/template/claude-task-manager/lib/terminal-fingerprint.js +48 -0
- package/template/claude-task-manager/lib/terminal-output-flush.js +84 -0
- package/template/claude-task-manager/lib/timeline-order.js +122 -0
- package/template/claude-task-manager/lib/transcript-store.js +348 -43
- package/template/claude-task-manager/lib/transport-security.js +84 -1
- package/template/claude-task-manager/lib/wait-state.js +184 -0
- package/template/claude-task-manager/lib/walle-client.js +47 -5
- package/template/claude-task-manager/lib/walle-ctm-history.js +564 -4
- package/template/claude-task-manager/lib/walle-external-actions.js +135 -16
- package/template/claude-task-manager/lib/walle-history-hydration.js +46 -0
- package/template/claude-task-manager/lib/walle-native-health.js +403 -0
- package/template/claude-task-manager/lib/walle-repair.js +701 -0
- package/template/claude-task-manager/lib/walle-session-cache.js +109 -0
- package/template/claude-task-manager/lib/walle-session-context.js +57 -21
- package/template/claude-task-manager/lib/walle-session-model-catalog.js +34 -0
- package/template/claude-task-manager/lib/walle-supervisor.js +539 -63
- package/template/claude-task-manager/lib/walle-transcript.js +52 -0
- package/template/claude-task-manager/lib/worktree-active-sync.js +11 -7
- package/template/claude-task-manager/lib/worktree-cwd.js +32 -1
- package/template/claude-task-manager/package.json +1 -1
- package/template/claude-task-manager/prompt-harvest.js +89 -66
- package/template/claude-task-manager/providers/claude-code.js +51 -3
- package/template/claude-task-manager/providers/cursor.js +140 -45
- package/template/claude-task-manager/public/css/reviews.css +551 -61
- package/template/claude-task-manager/public/css/setup.css +191 -0
- package/template/claude-task-manager/public/css/walle-session.css +865 -10
- package/template/claude-task-manager/public/css/walle.css +154 -0
- package/template/claude-task-manager/public/designs/ai-providers-consolidation-v2.html +830 -0
- package/template/claude-task-manager/public/index.html +18516 -2058
- package/template/claude-task-manager/public/ipad.html +363 -0
- package/template/claude-task-manager/public/js/document-review-links.js +301 -0
- package/template/claude-task-manager/public/js/image-normalize.js +69 -36
- package/template/claude-task-manager/public/js/message-renderer.js +1265 -77
- package/template/claude-task-manager/public/js/prompts.js +66 -29
- package/template/claude-task-manager/public/js/reviews.js +901 -133
- package/template/claude-task-manager/public/js/session-activity-utils.js +11 -1
- package/template/claude-task-manager/public/js/session-search-utils.js +94 -10
- package/template/claude-task-manager/public/js/session-status-precedence.js +23 -5
- package/template/claude-task-manager/public/js/setup.js +1273 -176
- package/template/claude-task-manager/public/js/stream-view.js +691 -73
- package/template/claude-task-manager/public/js/terminal-reconciler.js +210 -0
- package/template/claude-task-manager/public/js/walle-session.js +2455 -158
- package/template/claude-task-manager/public/js/walle.js +455 -28
- package/template/claude-task-manager/public/m/app.css +2909 -262
- package/template/claude-task-manager/public/m/app.js +6601 -398
- package/template/claude-task-manager/public/m/claim.html +224 -17
- package/template/claude-task-manager/public/m/index.html +117 -21
- package/template/claude-task-manager/public/m/sw.js +3 -1
- package/template/claude-task-manager/public/manifest.json +2 -2
- package/template/claude-task-manager/public/prompts.html +30 -14
- package/template/claude-task-manager/queue-engine.js +507 -28
- package/template/claude-task-manager/scripts/repair-claude-session-images.js +27 -8
- package/template/claude-task-manager/server.js +14341 -2197
- package/template/claude-task-manager/session-integrity.js +160 -18
- package/template/claude-task-manager/session-search-ranking.js +1 -0
- package/template/claude-task-manager/session-utils.js +25 -5
- package/template/claude-task-manager/workers/approval-blocklist.js +96 -6
- package/template/claude-task-manager/workers/approval-widget-validator.js +14 -8
- package/template/claude-task-manager/workers/conversation-import-worker.js +11 -50
- package/template/claude-task-manager/workers/db-owner-worker.js +386 -0
- package/template/claude-task-manager/workers/harvest-worker.js +9 -55
- package/template/claude-task-manager/workers/headless-term-worker.js +9 -530
- package/template/claude-task-manager/workers/read-pool-worker.js +387 -0
- package/template/claude-task-manager/workers/scrollback-worker.js +11 -72
- package/template/claude-task-manager/workers/session-host-process.js +146 -0
- package/template/claude-task-manager/workers/session-integrity-worker.js +10 -54
- package/template/claude-task-manager/workers/state-detectors/base.js +18 -1
- package/template/claude-task-manager/workers/state-detectors/claude-code.js +182 -9
- package/template/claude-task-manager/workers/state-detectors/codex.js +150 -2
- package/template/claude-task-manager/workers/state-detectors/cursor.js +127 -0
- package/template/claude-task-manager/workers/state-detectors/gemini.js +21 -0
- package/template/claude-task-manager/workers/state-detectors/index.js +29 -0
- package/template/claude-task-manager/workers/state-detectors/opencode.js +103 -0
- package/template/docs/design/markdown-review-pane.md +206 -0
- package/template/docs/designs/2026-05-17-portkey-gateway-provider-ux.md +129 -38
- package/template/docs/designs/2026-05-20-mobile-worktree-finish-command.md +27 -0
- package/template/docs/designs/2026-05-22-ai-configuration-consolidation.md +248 -0
- package/template/docs/designs/ai-configuration-consolidation-mock.html +812 -0
- package/template/docs/private-memory-and-pii-policy.md +69 -0
- package/template/package.json +2 -1
- package/template/scripts/check-private-data.js +201 -0
- package/template/shared/sqlite-owner-guard.js +30 -0
- package/template/shared/sqlite-owner-write-queue.js +225 -0
- package/template/shared/sqlite-storage-policy.js +111 -0
- package/template/shared/sqlite-write-lock.js +428 -0
- package/template/wall-e/agent-runners/claude-code.js +5 -0
- package/template/wall-e/agent.js +166 -22
- package/template/wall-e/api-walle.js +524 -70
- package/template/wall-e/auth/provider-flows.js +11 -1
- package/template/wall-e/bin/walle-mcp-stdio.js +341 -17
- package/template/wall-e/brain.js +1614 -141
- package/template/wall-e/chat/attachment-blocks.js +96 -0
- package/template/wall-e/chat/attachments.js +2 -1
- package/template/wall-e/chat/capability-resolver.js +7 -7
- package/template/wall-e/chat/context-messages.js +28 -0
- package/template/wall-e/chat/conversation-frame.js +630 -0
- package/template/wall-e/chat/provider-messages.js +125 -0
- package/template/wall-e/chat.js +1002 -233
- package/template/wall-e/coding/acceptance-contract.js +170 -0
- package/template/wall-e/coding/acp-adapter.js +1 -1
- package/template/wall-e/coding/agent-catalog.js +3 -0
- package/template/wall-e/coding/artifact-store.js +93 -0
- package/template/wall-e/coding/capability-router.js +120 -0
- package/template/wall-e/coding/coding-run-controller.js +423 -0
- package/template/wall-e/coding/compaction-service.js +157 -12
- package/template/wall-e/coding/frontend-verification.js +258 -0
- package/template/wall-e/coding/lifecycle-hooks.js +75 -0
- package/template/wall-e/coding/local-preview-contract.js +157 -0
- package/template/wall-e/coding/permission-service.js +57 -13
- package/template/wall-e/coding/prompt-bundle.js +19 -1
- package/template/wall-e/coding/prompt-section-registry.js +227 -0
- package/template/wall-e/coding/provider-compat.js +15 -0
- package/template/wall-e/coding/runtime-events.js +224 -0
- package/template/wall-e/coding/runtime-mode.js +3 -0
- package/template/wall-e/coding/side-git-snapshot.js +160 -4
- package/template/wall-e/coding/snapshot-service.js +143 -1
- package/template/wall-e/coding/stream-processor.js +388 -34
- package/template/wall-e/coding/task-tool.js +141 -4
- package/template/wall-e/coding/tool-execution-controller.js +365 -0
- package/template/wall-e/coding/tool-registry.js +43 -5
- package/template/wall-e/coding/user-hooks.js +217 -0
- package/template/wall-e/coding-orchestrator.js +1330 -221
- package/template/wall-e/coding-prompts.js +20 -4
- package/template/wall-e/context/context-builder.js +15 -2
- package/template/wall-e/decision/confidence.js +1 -1
- package/template/wall-e/docs/coding-acceptance-contract.md +41 -0
- package/template/wall-e/docs/external-action-controller.md +26 -6
- package/template/wall-e/docs/telemetry-lifecycle.md +8 -2
- package/template/wall-e/embeddings.js +591 -53
- package/template/wall-e/external-action-controller.js +12 -0
- package/template/wall-e/http/auth.js +1 -0
- package/template/wall-e/http/chat-api.js +46 -11
- package/template/wall-e/http/model-admin.js +836 -34
- package/template/wall-e/lib/boot-profile.js +88 -0
- package/template/wall-e/lib/event-loop-monitor.js +93 -0
- package/template/wall-e/lib/service-health.js +194 -0
- package/template/wall-e/llm/anthropic.js +130 -5
- package/template/wall-e/llm/client.js +266 -63
- package/template/wall-e/llm/default-fallback.js +382 -0
- package/template/wall-e/llm/health.js +19 -0
- package/template/wall-e/llm/message-guard.js +78 -0
- package/template/wall-e/llm/model-catalog.js +252 -1
- package/template/wall-e/llm/openai.js +26 -4
- package/template/wall-e/llm/portkey-sync.js +654 -0
- package/template/wall-e/llm/provider-error.js +30 -2
- package/template/wall-e/llm/registry.js +5 -1
- package/template/wall-e/llm/request-compat.js +67 -0
- package/template/wall-e/loops/backfill.js +79 -23
- package/template/wall-e/loops/brain-optimize.js +67 -0
- package/template/wall-e/loops/ingest.js +25 -10
- package/template/wall-e/loops/question-digest.js +160 -0
- package/template/wall-e/loops/reflect.js +6 -4
- package/template/wall-e/loops/think.js +39 -12
- package/template/wall-e/mcp-server.js +318 -36
- package/template/wall-e/memory/ctm-context-client.js +52 -14
- package/template/wall-e/memory/ctm-operational-context.js +237 -0
- package/template/wall-e/memory/ctm-prompt-executions-client.js +128 -0
- package/template/wall-e/memory/ctm-session-context.js +111 -63
- package/template/wall-e/prompts/coding/deepseek.txt +3 -0
- package/template/wall-e/prompts/coding/gemini.txt +6 -0
- package/template/wall-e/prompts/coding/gpt.txt +6 -0
- package/template/wall-e/prompts/coding/local.txt +7 -0
- package/template/wall-e/runtime/decision-hooks.js +115 -0
- package/template/wall-e/runtime/devbox-gateway.js +82 -8
- package/template/wall-e/runtime/prompt-manifest.js +86 -0
- package/template/wall-e/runtime/tool-executor.js +269 -0
- package/template/wall-e/runtime/tool-result-envelope.js +138 -0
- package/template/wall-e/runtime/transcript-projection.js +60 -0
- package/template/wall-e/runtime/walle-runtime.js +224 -0
- package/template/wall-e/scripts/db-optimize/migrate.js +162 -0
- package/template/wall-e/scripts/db-optimize/recall-eval.js +117 -0
- package/template/wall-e/server.js +15 -0
- package/template/wall-e/session-files.js +9 -0
- package/template/wall-e/skills/_bundled/google-calendar/run.js +1 -1
- package/template/wall-e/skills/_bundled/gws-workspace/run.js +1 -1
- package/template/wall-e/skills/_bundled/slack-mentions/run.js +76 -6
- package/template/wall-e/skills/claude-code-reader.js +7 -3
- package/template/wall-e/skills/script-skill-runner.js +10 -0
- package/template/wall-e/skills/skill-planner.js +38 -0
- package/template/wall-e/tools/builtin-middleware.js +19 -9
- package/template/wall-e/tools/local-tools.js +1428 -16
- package/template/wall-e/tools/permission-checker.js +73 -5
- package/template/wall-e/tools/question-manager.js +117 -7
- package/template/wall-e/training/harvester.js +12 -28
- package/template/wall-e/training/replay.js +25 -80
- package/template/website/index.html +10 -10
- package/template/wall-e/eval/ab-test.js +0 -203
- package/template/wall-e/eval/agent-runner.js +0 -772
- package/template/wall-e/eval/agent-scorer.js +0 -461
- package/template/wall-e/eval/aggregator.js +0 -414
- package/template/wall-e/eval/allowed-test-commands.js +0 -34
- package/template/wall-e/eval/benchmark-generator.js +0 -113
- package/template/wall-e/eval/benchmarks/chat-eval.json +0 -1662
- package/template/wall-e/eval/benchmarks/chat.json +0 -82
- package/template/wall-e/eval/benchmarks/coding-agent-real.json +0 -1
- package/template/wall-e/eval/benchmarks/coding-agent.json +0 -1581
- package/template/wall-e/eval/benchmarks/coding.json +0 -122
- package/template/wall-e/eval/benchmarks/memory-retrieval.json +0 -234
- package/template/wall-e/eval/benchmarks/reasoning.json +0 -82
- package/template/wall-e/eval/benchmarks/swebench-lite-30.json +0 -212
- package/template/wall-e/eval/benchmarks.js +0 -669
- package/template/wall-e/eval/cc-replay.js +0 -719
- package/template/wall-e/eval/chat-eval.js +0 -525
- package/template/wall-e/eval/check-keys.js +0 -15
- package/template/wall-e/eval/check-providers.js +0 -42
- package/template/wall-e/eval/codex-cli-baseline.js +0 -669
- package/template/wall-e/eval/coding-agent-real.js +0 -570
- package/template/wall-e/eval/context-compactor.js +0 -251
- package/template/wall-e/eval/debug-agent003.js +0 -68
- package/template/wall-e/eval/diagnostics.js +0 -216
- package/template/wall-e/eval/eval-orchestrator.js +0 -642
- package/template/wall-e/eval/evaluate.js +0 -202
- package/template/wall-e/eval/evaluator.js +0 -373
- package/template/wall-e/eval/exporter.js +0 -212
- package/template/wall-e/eval/fixtures/express-basic/package.json +0 -9
- package/template/wall-e/eval/fixtures/express-basic/server.js +0 -115
- package/template/wall-e/eval/fixtures/express-basic/test.js +0 -83
- package/template/wall-e/eval/fixtures/express-buggy/package.json +0 -9
- package/template/wall-e/eval/fixtures/express-buggy/server.js +0 -113
- package/template/wall-e/eval/fixtures/express-buggy/test.js +0 -83
- package/template/wall-e/eval/fixtures/express-buggy-items/package.json +0 -9
- package/template/wall-e/eval/fixtures/express-buggy-items/server.js +0 -112
- package/template/wall-e/eval/fixtures/express-buggy-items/test.js +0 -83
- package/template/wall-e/eval/fixtures/express-buggy-search/package.json +0 -9
- package/template/wall-e/eval/fixtures/express-buggy-search/server.js +0 -121
- package/template/wall-e/eval/fixtures/express-buggy-search/test.js +0 -83
- package/template/wall-e/eval/fixtures/express-rename-data/data.js +0 -34
- package/template/wall-e/eval/fixtures/express-rename-data/package.json +0 -9
- package/template/wall-e/eval/fixtures/express-rename-data/server.js +0 -97
- package/template/wall-e/eval/fixtures/express-rename-data/test.js +0 -88
- package/template/wall-e/eval/fixtures/express-xss/package.json +0 -12
- package/template/wall-e/eval/fixtures/express-xss/server.js +0 -90
- package/template/wall-e/eval/fixtures/express-xss/test.js +0 -67
- package/template/wall-e/eval/fixtures/express-xss/views/profile.ejs +0 -9
- package/template/wall-e/eval/fixtures/fullstack-app/config/default.js +0 -9
- package/template/wall-e/eval/fixtures/fullstack-app/config/test.js +0 -13
- package/template/wall-e/eval/fixtures/fullstack-app/package.json +0 -11
- package/template/wall-e/eval/fixtures/fullstack-app/public/css/style.css +0 -137
- package/template/wall-e/eval/fixtures/fullstack-app/public/index.html +0 -46
- package/template/wall-e/eval/fixtures/fullstack-app/public/js/app.js +0 -121
- package/template/wall-e/eval/fixtures/fullstack-app/public/js/auth.js +0 -71
- package/template/wall-e/eval/fixtures/fullstack-app/public/js/items.js +0 -80
- package/template/wall-e/eval/fixtures/fullstack-app/public/js/users.js +0 -46
- package/template/wall-e/eval/fixtures/fullstack-app/public/login.html +0 -45
- package/template/wall-e/eval/fixtures/fullstack-app/public/register.html +0 -38
- package/template/wall-e/eval/fixtures/fullstack-app/scripts/migrate.js +0 -23
- package/template/wall-e/eval/fixtures/fullstack-app/scripts/seed.js +0 -46
- package/template/wall-e/eval/fixtures/fullstack-app/server/db.js +0 -99
- package/template/wall-e/eval/fixtures/fullstack-app/server/index.js +0 -94
- package/template/wall-e/eval/fixtures/fullstack-app/server/middleware/auth.js +0 -19
- package/template/wall-e/eval/fixtures/fullstack-app/server/middleware/logger.js +0 -19
- package/template/wall-e/eval/fixtures/fullstack-app/server/router.js +0 -50
- package/template/wall-e/eval/fixtures/fullstack-app/server/routes/auth.js +0 -69
- package/template/wall-e/eval/fixtures/fullstack-app/server/routes/health.js +0 -23
- package/template/wall-e/eval/fixtures/fullstack-app/server/routes/items.js +0 -88
- package/template/wall-e/eval/fixtures/fullstack-app/server/routes/users.js +0 -75
- package/template/wall-e/eval/fixtures/fullstack-app/server/test.js +0 -198
- package/template/wall-e/eval/fixtures/fullstack-app/server/utils/response.js +0 -34
- package/template/wall-e/eval/fixtures/fullstack-app/server/utils/validate.js +0 -26
- package/template/wall-e/eval/fixtures/fullstack-app/server.js +0 -8
- package/template/wall-e/eval/fixtures/fullstack-app/test.js +0 -12
- package/template/wall-e/eval/fixtures/monorepo-basic/package.json +0 -8
- package/template/wall-e/eval/fixtures/monorepo-basic/packages/api/data.js +0 -58
- package/template/wall-e/eval/fixtures/monorepo-basic/packages/api/middleware.js +0 -46
- package/template/wall-e/eval/fixtures/monorepo-basic/packages/api/package.json +0 -8
- package/template/wall-e/eval/fixtures/monorepo-basic/packages/api/routes.js +0 -64
- package/template/wall-e/eval/fixtures/monorepo-basic/packages/api/server.js +0 -56
- package/template/wall-e/eval/fixtures/monorepo-basic/packages/api/test.js +0 -116
- package/template/wall-e/eval/fixtures/monorepo-basic/packages/cli/commands.js +0 -61
- package/template/wall-e/eval/fixtures/monorepo-basic/packages/cli/index.js +0 -62
- package/template/wall-e/eval/fixtures/monorepo-basic/packages/cli/output.js +0 -43
- package/template/wall-e/eval/fixtures/monorepo-basic/packages/cli/package.json +0 -11
- package/template/wall-e/eval/fixtures/monorepo-basic/packages/cli/test.js +0 -44
- package/template/wall-e/eval/fixtures/monorepo-basic/packages/shared/formatters.js +0 -43
- package/template/wall-e/eval/fixtures/monorepo-basic/packages/shared/index.js +0 -12
- package/template/wall-e/eval/fixtures/monorepo-basic/packages/shared/package.json +0 -5
- package/template/wall-e/eval/fixtures/monorepo-basic/packages/shared/test.js +0 -55
- package/template/wall-e/eval/fixtures/monorepo-basic/packages/shared/validators.js +0 -29
- package/template/wall-e/eval/fixtures/monorepo-basic/test.js +0 -46
- package/template/wall-e/eval/fixtures/node-cli/index.js +0 -78
- package/template/wall-e/eval/fixtures/node-cli/package.json +0 -10
- package/template/wall-e/eval/fixtures/node-cli/test.js +0 -57
- package/template/wall-e/eval/fixtures/node-typed/package.json +0 -8
- package/template/wall-e/eval/fixtures/node-typed/src/handlers.js +0 -31
- package/template/wall-e/eval/fixtures/node-typed/src/utils.js +0 -33
- package/template/wall-e/eval/fixtures/node-typed/test.js +0 -36
- package/template/wall-e/eval/fixtures/python-flask/app.py +0 -14
- package/template/wall-e/eval/fixtures/python-flask/requirements.txt +0 -2
- package/template/wall-e/eval/fixtures/python-flask/test_app.py +0 -25
- package/template/wall-e/eval/fixtures/wall-e-subset/brain.js +0 -105
- package/template/wall-e/eval/fixtures/wall-e-subset/eval/aggregator.js +0 -101
- package/template/wall-e/eval/fixtures/wall-e-subset/eval/benchmarks/chat.json +0 -20
- package/template/wall-e/eval/fixtures/wall-e-subset/eval/benchmarks/coding.json +0 -32
- package/template/wall-e/eval/fixtures/wall-e-subset/eval/benchmarks.js +0 -64
- package/template/wall-e/eval/fixtures/wall-e-subset/eval/fixtures/simple-project/package.json +0 -6
- package/template/wall-e/eval/fixtures/wall-e-subset/eval/fixtures/simple-project/server.js +0 -31
- package/template/wall-e/eval/fixtures/wall-e-subset/eval/fixtures/simple-project/test.js +0 -18
- package/template/wall-e/eval/fixtures/wall-e-subset/eval/fixtures/simple-project/utils.js +0 -34
- package/template/wall-e/eval/fixtures/wall-e-subset/eval/runner.js +0 -104
- package/template/wall-e/eval/fixtures/wall-e-subset/eval/scorer.js +0 -73
- package/template/wall-e/eval/fixtures/wall-e-subset/eval/test.js +0 -134
- package/template/wall-e/eval/fixtures/wall-e-subset/llm/client.js +0 -99
- package/template/wall-e/eval/fixtures/wall-e-subset/llm/providers.js +0 -63
- package/template/wall-e/eval/fixtures/wall-e-subset/llm/test.js +0 -70
- package/template/wall-e/eval/fixtures/wall-e-subset/package.json +0 -10
- package/template/wall-e/eval/fixtures/wall-e-subset/test.js +0 -86
- package/template/wall-e/eval/harvester.js +0 -685
- package/template/wall-e/eval/head-to-head.js +0 -388
- package/template/wall-e/eval/humaneval-adapter.js +0 -321
- package/template/wall-e/eval/list-models.js +0 -31
- package/template/wall-e/eval/livecodebench-adapter.js +0 -291
- package/template/wall-e/eval/mail-integration.js +0 -443
- package/template/wall-e/eval/manifest.js +0 -186
- package/template/wall-e/eval/meta-harness/adapters/coding-agent.js +0 -57
- package/template/wall-e/eval/meta-harness/bootstrap-snapshot.js +0 -149
- package/template/wall-e/eval/meta-harness/candidate-store.js +0 -117
- package/template/wall-e/eval/meta-harness/cli.js +0 -86
- package/template/wall-e/eval/meta-harness/domain-spec.js +0 -154
- package/template/wall-e/eval/meta-harness/domains/coding-agent.domain.json +0 -84
- package/template/wall-e/eval/meta-harness/examples/env-bootstrap-candidate.js +0 -29
- package/template/wall-e/eval/meta-harness/experience-store.js +0 -174
- package/template/wall-e/eval/meta-harness/frontier.js +0 -96
- package/template/wall-e/eval/meta-harness/harness-interface.js +0 -90
- package/template/wall-e/eval/meta-harness/leakage-guard.js +0 -80
- package/template/wall-e/eval/meta-harness/optimizer.js +0 -207
- package/template/wall-e/eval/meta-harness/proposer-runner.js +0 -110
- package/template/wall-e/eval/meta-harness/reporting.js +0 -58
- package/template/wall-e/eval/meta-harness/telemetry.js +0 -27
- package/template/wall-e/eval/meta-harness/validation.js +0 -81
- package/template/wall-e/eval/promoter.js +0 -228
- package/template/wall-e/eval/provider-normalizer.js +0 -33
- package/template/wall-e/eval/replay.js +0 -395
- package/template/wall-e/eval/run-agent-benchmarks.js +0 -386
- package/template/wall-e/eval/run-codex-cli-baseline.js +0 -177
- package/template/wall-e/eval/run-coding-agent-real.js +0 -187
- package/template/wall-e/eval/run-eval.js +0 -435
- package/template/wall-e/eval/run-model-comparison.js +0 -142
- package/template/wall-e/eval/session-evaluator.js +0 -187
- package/template/wall-e/eval/session-miner.js +0 -207
- package/template/wall-e/eval/session-retrieval-benchmark.js +0 -150
- package/template/wall-e/eval/session-transcripts.js +0 -509
- package/template/wall-e/eval/shadow.js +0 -161
- package/template/wall-e/eval/swebench-adapter.js +0 -345
- package/template/wall-e/eval/swebench-docker.js +0 -192
- package/template/wall-e/eval/train.py +0 -320
- package/template/wall-e/eval/trainer.js +0 -232
- package/template/wall-e/eval/weekly-eval-loop.js +0 -241
|
@@ -0,0 +1,109 @@
|
|
|
1
|
+
'use strict';
|
|
2
|
+
|
|
3
|
+
const BLOCKING_STATUSES = new Set(['running', 'busy']);
|
|
4
|
+
|
|
5
|
+
function normalizeStatus(value) {
|
|
6
|
+
const text = String(value || '').trim().toLowerCase().replace(/[-\s]+/g, '_');
|
|
7
|
+
if (!text) return '';
|
|
8
|
+
if (text === 'waiting' || text === 'waiting_for_input' || text === 'waiting_input') return 'waiting_input';
|
|
9
|
+
if (text === 'busy') return 'running';
|
|
10
|
+
return text;
|
|
11
|
+
}
|
|
12
|
+
|
|
13
|
+
function sessionLabel(session) {
|
|
14
|
+
return String(
|
|
15
|
+
session?.label ||
|
|
16
|
+
session?.title ||
|
|
17
|
+
session?.name ||
|
|
18
|
+
session?.id ||
|
|
19
|
+
''
|
|
20
|
+
).trim();
|
|
21
|
+
}
|
|
22
|
+
|
|
23
|
+
function sessionId(session) {
|
|
24
|
+
return String(session?.id || session?.sessionId || session?.ctmSessionId || '').trim();
|
|
25
|
+
}
|
|
26
|
+
|
|
27
|
+
function restartSessionSummary(session, status, waitingForInput, waitingReason) {
|
|
28
|
+
const normalized = waitingForInput ? 'waiting_input' : normalizeStatus(status);
|
|
29
|
+
const summary = {
|
|
30
|
+
id: sessionId(session),
|
|
31
|
+
label: sessionLabel(session),
|
|
32
|
+
status: normalized || 'idle',
|
|
33
|
+
};
|
|
34
|
+
if (waitingReason) summary.waitingReason = waitingReason;
|
|
35
|
+
return summary;
|
|
36
|
+
}
|
|
37
|
+
|
|
38
|
+
function normalizeWaitingReason(value) {
|
|
39
|
+
return String(value || '').trim().toLowerCase().replace(/[-\s]+/g, '_');
|
|
40
|
+
}
|
|
41
|
+
|
|
42
|
+
function defaultWaitingReasonForSession(session) {
|
|
43
|
+
return normalizeWaitingReason(
|
|
44
|
+
session?.waitingReason ||
|
|
45
|
+
session?.waiting_reason ||
|
|
46
|
+
session?.reason ||
|
|
47
|
+
session?._waitingForInputReason ||
|
|
48
|
+
''
|
|
49
|
+
);
|
|
50
|
+
}
|
|
51
|
+
|
|
52
|
+
function isBlockingWaitingReason(reason) {
|
|
53
|
+
const text = normalizeWaitingReason(reason);
|
|
54
|
+
return text === 'approval' || text === 'choice';
|
|
55
|
+
}
|
|
56
|
+
|
|
57
|
+
function isBlockingWaitingSession(session, summary, waitingForInput, waitingReason, options) {
|
|
58
|
+
if (typeof options.isBlockingWaitingForInput === 'function') {
|
|
59
|
+
return !!options.isBlockingWaitingForInput(session);
|
|
60
|
+
}
|
|
61
|
+
if (isBlockingWaitingReason(waitingReason)) return true;
|
|
62
|
+
// Older callers did not provide waiting reasons. Preserve the conservative
|
|
63
|
+
// behavior for unknown waiting_input rows instead of silently restarting over
|
|
64
|
+
// a possible approval prompt.
|
|
65
|
+
return summary.status === 'waiting_input' && !waitingReason;
|
|
66
|
+
}
|
|
67
|
+
|
|
68
|
+
function summarizeRestartGuard(sessions = [], options = {}) {
|
|
69
|
+
const items = Array.from(sessions || []).filter(Boolean);
|
|
70
|
+
const blockers = [];
|
|
71
|
+
const restorable = [];
|
|
72
|
+
|
|
73
|
+
for (const session of items) {
|
|
74
|
+
const rawStatus = typeof options.statusForSession === 'function'
|
|
75
|
+
? options.statusForSession(session)
|
|
76
|
+
: (session.standupStatus || session.liveStatus || session.serverState || session.status || '');
|
|
77
|
+
const waitingForInput = typeof options.isWaitingForInput === 'function'
|
|
78
|
+
? !!options.isWaitingForInput(session)
|
|
79
|
+
: !!session.waitingForInput;
|
|
80
|
+
const waitingReason = typeof options.waitingReasonForSession === 'function'
|
|
81
|
+
? normalizeWaitingReason(options.waitingReasonForSession(session))
|
|
82
|
+
: defaultWaitingReasonForSession(session);
|
|
83
|
+
const summary = restartSessionSummary(session, rawStatus, waitingForInput, waitingReason);
|
|
84
|
+
if (
|
|
85
|
+
BLOCKING_STATUSES.has(summary.status) ||
|
|
86
|
+
(summary.status === 'waiting_input' &&
|
|
87
|
+
isBlockingWaitingSession(session, summary, waitingForInput, waitingReason, options))
|
|
88
|
+
) blockers.push(summary);
|
|
89
|
+
else restorable.push(summary);
|
|
90
|
+
}
|
|
91
|
+
|
|
92
|
+
return {
|
|
93
|
+
totalCount: items.length,
|
|
94
|
+
blockingCount: blockers.length,
|
|
95
|
+
restorableCount: restorable.length,
|
|
96
|
+
blockers,
|
|
97
|
+
restorable,
|
|
98
|
+
};
|
|
99
|
+
}
|
|
100
|
+
|
|
101
|
+
module.exports = {
|
|
102
|
+
summarizeRestartGuard,
|
|
103
|
+
_private: {
|
|
104
|
+
normalizeStatus,
|
|
105
|
+
normalizeWaitingReason,
|
|
106
|
+
isBlockingWaitingReason,
|
|
107
|
+
restartSessionSummary,
|
|
108
|
+
},
|
|
109
|
+
};
|
|
@@ -0,0 +1,439 @@
|
|
|
1
|
+
'use strict';
|
|
2
|
+
|
|
3
|
+
const fs = require('fs');
|
|
4
|
+
const path = require('path');
|
|
5
|
+
|
|
6
|
+
const DEFAULT_TAIL_BYTES = 2 * 1024 * 1024;
|
|
7
|
+
const MAX_PARSED_LINES = 400;
|
|
8
|
+
|
|
9
|
+
function _asString(value) {
|
|
10
|
+
return String(value || '').trim();
|
|
11
|
+
}
|
|
12
|
+
|
|
13
|
+
function _safeJson(value) {
|
|
14
|
+
try { return JSON.parse(value); } catch { return null; }
|
|
15
|
+
}
|
|
16
|
+
|
|
17
|
+
function _readJsonlTail(filePath, opts = {}) {
|
|
18
|
+
const target = _asString(filePath);
|
|
19
|
+
if (!target) return { entries: [], error: 'missing_path' };
|
|
20
|
+
try {
|
|
21
|
+
const stat = fs.statSync(target);
|
|
22
|
+
if (!stat.isFile() || stat.size <= 0) return { entries: [], error: 'empty_file' };
|
|
23
|
+
const tailBytes = Math.max(64 * 1024, Number(opts.tailBytes) || DEFAULT_TAIL_BYTES);
|
|
24
|
+
const start = Math.max(0, stat.size - tailBytes);
|
|
25
|
+
const fd = fs.openSync(target, 'r');
|
|
26
|
+
try {
|
|
27
|
+
const len = stat.size - start;
|
|
28
|
+
const buf = Buffer.allocUnsafe(len);
|
|
29
|
+
fs.readSync(fd, buf, 0, len, start);
|
|
30
|
+
let raw = buf.toString('utf8');
|
|
31
|
+
if (start > 0) {
|
|
32
|
+
const nl = raw.indexOf('\n');
|
|
33
|
+
raw = nl >= 0 ? raw.slice(nl + 1) : '';
|
|
34
|
+
}
|
|
35
|
+
const lines = raw.split('\n').filter(Boolean).slice(-MAX_PARSED_LINES);
|
|
36
|
+
const entries = [];
|
|
37
|
+
for (const line of lines) {
|
|
38
|
+
const parsed = _safeJson(line);
|
|
39
|
+
if (parsed && typeof parsed === 'object') entries.push(parsed);
|
|
40
|
+
}
|
|
41
|
+
return { entries, size: stat.size, mtimeMs: stat.mtimeMs || 0 };
|
|
42
|
+
} finally {
|
|
43
|
+
try { fs.closeSync(fd); } catch {}
|
|
44
|
+
}
|
|
45
|
+
} catch (e) {
|
|
46
|
+
return { entries: [], error: e.message || 'read_failed' };
|
|
47
|
+
}
|
|
48
|
+
}
|
|
49
|
+
|
|
50
|
+
function _entryText(entry) {
|
|
51
|
+
const content = entry?.message?.content ?? entry?.content;
|
|
52
|
+
if (typeof content === 'string') return content;
|
|
53
|
+
if (Array.isArray(content)) {
|
|
54
|
+
return content.map((block) => {
|
|
55
|
+
if (!block || typeof block !== 'object') return '';
|
|
56
|
+
if (typeof block.text === 'string') return block.text;
|
|
57
|
+
if (typeof block.content === 'string') return block.content;
|
|
58
|
+
if (block.type === 'tool_result') return `[tool_result:${block.tool_use_id || block.name || ''}]`;
|
|
59
|
+
if (block.type === 'tool_use') return `[tool_use:${block.name || ''}]`;
|
|
60
|
+
return '';
|
|
61
|
+
}).filter(Boolean).join('\n');
|
|
62
|
+
}
|
|
63
|
+
if (typeof entry?.text === 'string') return entry.text;
|
|
64
|
+
if (typeof entry?.summary === 'string') return entry.summary;
|
|
65
|
+
if (typeof entry?.content === 'string') return entry.content;
|
|
66
|
+
return '';
|
|
67
|
+
}
|
|
68
|
+
|
|
69
|
+
function _codexContentText(content, wantedTypes) {
|
|
70
|
+
if (typeof content === 'string') return content;
|
|
71
|
+
if (!Array.isArray(content)) return '';
|
|
72
|
+
const types = new Set(Array.isArray(wantedTypes) ? wantedTypes : []);
|
|
73
|
+
return content.map((block) => {
|
|
74
|
+
if (!block || typeof block !== 'object') return '';
|
|
75
|
+
if (types.size && !types.has(block.type)) return '';
|
|
76
|
+
if (typeof block.text === 'string') return block.text;
|
|
77
|
+
if (typeof block.content === 'string') return block.content;
|
|
78
|
+
return '';
|
|
79
|
+
}).filter(Boolean).join('\n').trim();
|
|
80
|
+
}
|
|
81
|
+
|
|
82
|
+
function _codexUserText(entry) {
|
|
83
|
+
const payload = entry?.payload || {};
|
|
84
|
+
if (payload.type === 'user_message') return _asString(payload.message);
|
|
85
|
+
if (payload.type === 'message' && payload.role === 'user') {
|
|
86
|
+
return _codexContentText(payload.content, ['input_text', 'text']);
|
|
87
|
+
}
|
|
88
|
+
return '';
|
|
89
|
+
}
|
|
90
|
+
|
|
91
|
+
function _isCodexProviderContextUserText(text) {
|
|
92
|
+
const cleaned = _asString(text);
|
|
93
|
+
if (!cleaned) return true;
|
|
94
|
+
return /^<environment_context>/i.test(cleaned) ||
|
|
95
|
+
/^<permissions instructions>/i.test(cleaned) ||
|
|
96
|
+
/^<collaboration_mode>/i.test(cleaned) ||
|
|
97
|
+
/^<skills_instructions>/i.test(cleaned) ||
|
|
98
|
+
/^<plugins_instructions>/i.test(cleaned) ||
|
|
99
|
+
/^<skill>\s*<name>/i.test(cleaned) ||
|
|
100
|
+
/^# AGENTS\.md instructions for\b/i.test(cleaned);
|
|
101
|
+
}
|
|
102
|
+
|
|
103
|
+
function _isCodexTurnAbortedText(text) {
|
|
104
|
+
return /^<turn_aborted>\s*[\s\S]*?<\/turn_aborted>\s*$/i.test(_asString(text));
|
|
105
|
+
}
|
|
106
|
+
|
|
107
|
+
function _codexToolPreview(payload) {
|
|
108
|
+
if (!payload || typeof payload !== 'object') return '';
|
|
109
|
+
if (payload.name) return `[tool:${payload.name}]`;
|
|
110
|
+
if (payload.call_id) return `[tool:${payload.call_id}]`;
|
|
111
|
+
return '';
|
|
112
|
+
}
|
|
113
|
+
|
|
114
|
+
function _contentBlocks(entry) {
|
|
115
|
+
const content = entry?.message?.content ?? entry?.content;
|
|
116
|
+
return Array.isArray(content) ? content : [];
|
|
117
|
+
}
|
|
118
|
+
|
|
119
|
+
function _hasToolResult(entry) {
|
|
120
|
+
if (!entry || typeof entry !== 'object') return false;
|
|
121
|
+
if (entry.type === 'tool_result') return true;
|
|
122
|
+
return _contentBlocks(entry).some((block) => block && block.type === 'tool_result');
|
|
123
|
+
}
|
|
124
|
+
|
|
125
|
+
function _hasToolUse(entry) {
|
|
126
|
+
if (!entry || typeof entry !== 'object') return false;
|
|
127
|
+
if (entry.type === 'tool_use') return true;
|
|
128
|
+
return _contentBlocks(entry).some((block) => block && block.type === 'tool_use');
|
|
129
|
+
}
|
|
130
|
+
|
|
131
|
+
function _isMetadataNoise(entry) {
|
|
132
|
+
const type = _asString(entry?.type).toLowerCase();
|
|
133
|
+
return !type ||
|
|
134
|
+
type === 'summary' ||
|
|
135
|
+
type === 'metadata' ||
|
|
136
|
+
type === 'last-prompt' ||
|
|
137
|
+
type === 'ai-title' ||
|
|
138
|
+
type === 'title' ||
|
|
139
|
+
type === 'message-stat' ||
|
|
140
|
+
type === 'session-meta' ||
|
|
141
|
+
type === 'model';
|
|
142
|
+
}
|
|
143
|
+
|
|
144
|
+
function _semanticEvent(entry) {
|
|
145
|
+
const type = _asString(entry?.type).toLowerCase();
|
|
146
|
+
const timestamp = _asString(entry?.timestamp);
|
|
147
|
+
const payload = entry?.payload || {};
|
|
148
|
+
const payloadType = _asString(payload?.type).toLowerCase();
|
|
149
|
+
|
|
150
|
+
if (type === 'event_msg') {
|
|
151
|
+
if (payloadType === 'user_message') {
|
|
152
|
+
const text = _codexUserText(entry);
|
|
153
|
+
if (!text || _isCodexProviderContextUserText(text)) return null;
|
|
154
|
+
if (_isCodexTurnAbortedText(text)) return { kind: 'codex_turn_aborted', timestamp, text };
|
|
155
|
+
return { kind: 'user', timestamp, text };
|
|
156
|
+
}
|
|
157
|
+
if (payloadType === 'agent_message') {
|
|
158
|
+
const text = _asString(payload.message);
|
|
159
|
+
const phase = _asString(payload.phase).toLowerCase();
|
|
160
|
+
if (!text) return null;
|
|
161
|
+
return {
|
|
162
|
+
kind: phase === 'commentary' ? 'codex_assistant_commentary' : 'assistant',
|
|
163
|
+
timestamp,
|
|
164
|
+
text,
|
|
165
|
+
phase,
|
|
166
|
+
};
|
|
167
|
+
}
|
|
168
|
+
if (payloadType === 'task_complete') {
|
|
169
|
+
const text = _asString(payload.last_agent_message);
|
|
170
|
+
return { kind: 'assistant', timestamp, text, phase: 'task_complete' };
|
|
171
|
+
}
|
|
172
|
+
if (payloadType === 'turn_aborted') {
|
|
173
|
+
return { kind: 'codex_turn_aborted', timestamp, text: payload.reason || payload.message || 'turn_aborted' };
|
|
174
|
+
}
|
|
175
|
+
if (payloadType === 'patch_apply_end') {
|
|
176
|
+
return { kind: 'tool_result', timestamp, text: payload.stdout || payload.stderr || 'patch_apply_end' };
|
|
177
|
+
}
|
|
178
|
+
}
|
|
179
|
+
|
|
180
|
+
if (type === 'response_item') {
|
|
181
|
+
if (payloadType === 'message' && payload.role === 'user') {
|
|
182
|
+
const text = _codexUserText(entry);
|
|
183
|
+
if (!text || _isCodexProviderContextUserText(text)) return null;
|
|
184
|
+
if (_isCodexTurnAbortedText(text)) return { kind: 'codex_turn_aborted', timestamp, text };
|
|
185
|
+
return { kind: 'user', timestamp, text };
|
|
186
|
+
}
|
|
187
|
+
if (payloadType === 'message' && payload.role === 'assistant') {
|
|
188
|
+
const text = _codexContentText(payload.content, ['output_text', 'text']);
|
|
189
|
+
if (!text) return null;
|
|
190
|
+
const phase = _asString(payload.phase).toLowerCase();
|
|
191
|
+
return {
|
|
192
|
+
kind: phase === 'commentary' ? 'codex_assistant_commentary' : 'assistant',
|
|
193
|
+
timestamp,
|
|
194
|
+
text,
|
|
195
|
+
phase,
|
|
196
|
+
};
|
|
197
|
+
}
|
|
198
|
+
if (payloadType === 'function_call' || payloadType === 'custom_tool_call') {
|
|
199
|
+
return { kind: 'assistant_tool_use', timestamp, text: _codexToolPreview(payload), stopReason: 'tool_use' };
|
|
200
|
+
}
|
|
201
|
+
if (payloadType === 'function_call_output' || payloadType === 'custom_tool_call_output') {
|
|
202
|
+
return { kind: 'tool_result', timestamp, text: payload.output || _codexToolPreview(payload) };
|
|
203
|
+
}
|
|
204
|
+
}
|
|
205
|
+
|
|
206
|
+
if (type === 'assistant' || entry?.message?.role === 'assistant') {
|
|
207
|
+
const stopReason = _asString(entry?.message?.stop_reason || entry?.stop_reason).toLowerCase();
|
|
208
|
+
return {
|
|
209
|
+
kind: _hasToolUse(entry) || stopReason === 'tool_use' ? 'assistant_tool_use' : 'assistant',
|
|
210
|
+
timestamp,
|
|
211
|
+
stopReason,
|
|
212
|
+
text: _entryText(entry),
|
|
213
|
+
};
|
|
214
|
+
}
|
|
215
|
+
if (type === 'user' || entry?.message?.role === 'user') {
|
|
216
|
+
if (_hasToolResult(entry)) {
|
|
217
|
+
return { kind: 'tool_result', timestamp, text: _entryText(entry) };
|
|
218
|
+
}
|
|
219
|
+
const text = _entryText(entry);
|
|
220
|
+
if (!text) return null;
|
|
221
|
+
return { kind: 'user', timestamp, text };
|
|
222
|
+
}
|
|
223
|
+
if (type === 'tool_result') {
|
|
224
|
+
return { kind: 'tool_result', timestamp, text: _entryText(entry) };
|
|
225
|
+
}
|
|
226
|
+
if (type === 'queue-operation') {
|
|
227
|
+
const operation = _asString(entry.operation).toLowerCase();
|
|
228
|
+
const content = _entryText(entry);
|
|
229
|
+
if (operation === 'enqueue' && /<status>\s*completed\s*<\/status>|completed/i.test(content)) {
|
|
230
|
+
return { kind: 'background_task_completed', timestamp, text: content };
|
|
231
|
+
}
|
|
232
|
+
return null;
|
|
233
|
+
}
|
|
234
|
+
if (_isMetadataNoise(entry)) return null;
|
|
235
|
+
return null;
|
|
236
|
+
}
|
|
237
|
+
|
|
238
|
+
function _lastSemanticEvent(entries) {
|
|
239
|
+
for (let i = entries.length - 1; i >= 0; i--) {
|
|
240
|
+
const semantic = _semanticEvent(entries[i]);
|
|
241
|
+
if (semantic) return semantic;
|
|
242
|
+
}
|
|
243
|
+
return null;
|
|
244
|
+
}
|
|
245
|
+
|
|
246
|
+
function _reasonForSemantic(semantic) {
|
|
247
|
+
if (!semantic) return null;
|
|
248
|
+
if (semantic.kind === 'tool_result') {
|
|
249
|
+
return {
|
|
250
|
+
interrupted: true,
|
|
251
|
+
reason: 'tool_result_without_assistant_continuation',
|
|
252
|
+
label: 'Interrupted after a tool result',
|
|
253
|
+
detail: 'The last transcript event is a tool result, but no assistant message followed it.',
|
|
254
|
+
};
|
|
255
|
+
}
|
|
256
|
+
if (semantic.kind === 'background_task_completed') {
|
|
257
|
+
return {
|
|
258
|
+
interrupted: true,
|
|
259
|
+
reason: 'background_task_completed_without_assistant_continuation',
|
|
260
|
+
label: 'Interrupted after background work completed',
|
|
261
|
+
detail: 'A background task completion notification was written, but the assistant never processed it.',
|
|
262
|
+
};
|
|
263
|
+
}
|
|
264
|
+
if (semantic.kind === 'assistant_tool_use') {
|
|
265
|
+
return {
|
|
266
|
+
interrupted: true,
|
|
267
|
+
reason: 'assistant_tool_use_without_result',
|
|
268
|
+
label: 'Interrupted while a tool was pending',
|
|
269
|
+
detail: 'The assistant requested a tool and CTM restored before a later continuation was recorded.',
|
|
270
|
+
};
|
|
271
|
+
}
|
|
272
|
+
if (semantic.kind === 'codex_assistant_commentary') {
|
|
273
|
+
return {
|
|
274
|
+
interrupted: true,
|
|
275
|
+
reason: 'codex_commentary_without_final_response',
|
|
276
|
+
label: 'Interrupted after a progress update',
|
|
277
|
+
detail: 'Codex recorded an assistant commentary/progress message, but no final answer or task completion followed it.',
|
|
278
|
+
};
|
|
279
|
+
}
|
|
280
|
+
if (semantic.kind === 'codex_turn_aborted') {
|
|
281
|
+
return {
|
|
282
|
+
interrupted: true,
|
|
283
|
+
reason: 'codex_turn_aborted',
|
|
284
|
+
label: 'Interrupted by Codex turn abort',
|
|
285
|
+
detail: 'Codex recorded a turn-aborted control event before CTM restored the session.',
|
|
286
|
+
};
|
|
287
|
+
}
|
|
288
|
+
if (semantic.kind === 'user') {
|
|
289
|
+
return {
|
|
290
|
+
interrupted: true,
|
|
291
|
+
reason: 'user_prompt_without_assistant_response',
|
|
292
|
+
label: 'Interrupted before the assistant responded',
|
|
293
|
+
detail: 'The last transcript event is a user prompt with no later assistant response.',
|
|
294
|
+
};
|
|
295
|
+
}
|
|
296
|
+
return null;
|
|
297
|
+
}
|
|
298
|
+
|
|
299
|
+
function detectRestoreInterruptionFromEntries(entries, opts = {}) {
|
|
300
|
+
const semantic = _lastSemanticEvent(Array.isArray(entries) ? entries : []);
|
|
301
|
+
const reason = _reasonForSemantic(semantic);
|
|
302
|
+
if (!reason) {
|
|
303
|
+
return {
|
|
304
|
+
interrupted: false,
|
|
305
|
+
lastEventKind: semantic?.kind || '',
|
|
306
|
+
lastEventTimestamp: semantic?.timestamp || '',
|
|
307
|
+
};
|
|
308
|
+
}
|
|
309
|
+
const observedAt = Date.parse(semantic.timestamp || '') || Number(opts.now) || Date.now();
|
|
310
|
+
return {
|
|
311
|
+
...reason,
|
|
312
|
+
lastEventKind: semantic.kind,
|
|
313
|
+
lastEventTimestamp: semantic.timestamp || '',
|
|
314
|
+
observedAt,
|
|
315
|
+
preview: _asString(semantic.text).replace(/\s+/g, ' ').slice(0, 240),
|
|
316
|
+
};
|
|
317
|
+
}
|
|
318
|
+
|
|
319
|
+
function _selectAgentRow(db, ctmSessionId, provider) {
|
|
320
|
+
if (!db || !ctmSessionId) return null;
|
|
321
|
+
try {
|
|
322
|
+
const rows = db.prepare(`
|
|
323
|
+
SELECT *
|
|
324
|
+
FROM agent_sessions
|
|
325
|
+
WHERE ctm_session_id = ?
|
|
326
|
+
ORDER BY COALESCE(NULLIF(modified_at, ''), updated_at, created_at) DESC,
|
|
327
|
+
updated_at DESC,
|
|
328
|
+
created_at DESC
|
|
329
|
+
`).all(ctmSessionId);
|
|
330
|
+
const wanted = _asString(provider).toLowerCase();
|
|
331
|
+
return rows.find((row) => {
|
|
332
|
+
if (!row?.jsonl_path) return false;
|
|
333
|
+
if (!wanted) return true;
|
|
334
|
+
const rowProvider = _asString(row.provider).toLowerCase();
|
|
335
|
+
return !rowProvider || rowProvider === wanted;
|
|
336
|
+
}) || rows.find((row) => row?.jsonl_path) || null;
|
|
337
|
+
} catch {
|
|
338
|
+
return null;
|
|
339
|
+
}
|
|
340
|
+
}
|
|
341
|
+
|
|
342
|
+
function _normalizeArgs(value) {
|
|
343
|
+
if (Array.isArray(value)) return value.map((item) => String(item || ''));
|
|
344
|
+
if (typeof value === 'string') {
|
|
345
|
+
const parsed = _safeJson(value);
|
|
346
|
+
if (Array.isArray(parsed)) return parsed.map((item) => String(item || ''));
|
|
347
|
+
}
|
|
348
|
+
return [];
|
|
349
|
+
}
|
|
350
|
+
|
|
351
|
+
function _resumeIdFromArgs(args) {
|
|
352
|
+
const list = _normalizeArgs(args);
|
|
353
|
+
for (let i = 0; i < list.length; i++) {
|
|
354
|
+
const arg = list[i];
|
|
355
|
+
if ((arg === '--resume' || arg === '-r' || arg === '--session' || arg === '--session-id') && list[i + 1]) {
|
|
356
|
+
return _asString(list[i + 1]);
|
|
357
|
+
}
|
|
358
|
+
const eq = arg.match(/^--(?:resume|session|session-id)=(.+)$/);
|
|
359
|
+
if (eq) return _asString(eq[1]);
|
|
360
|
+
}
|
|
361
|
+
return '';
|
|
362
|
+
}
|
|
363
|
+
|
|
364
|
+
function _encodeProjectEntry(cwd) {
|
|
365
|
+
return String(cwd || '').replace(/\/\./g, '/-').replace(/\//g, '-');
|
|
366
|
+
}
|
|
367
|
+
|
|
368
|
+
function _canonicalClaudeCwd(cwd) {
|
|
369
|
+
const resolved = path.resolve(cwd || process.env.HOME || '.');
|
|
370
|
+
try { return fs.realpathSync(resolved); } catch { return resolved; }
|
|
371
|
+
}
|
|
372
|
+
|
|
373
|
+
function _existingTranscriptPath(basePath) {
|
|
374
|
+
if (!basePath) return '';
|
|
375
|
+
try {
|
|
376
|
+
if (fs.existsSync(basePath)) return basePath;
|
|
377
|
+
const bakPath = `${basePath}.bak`;
|
|
378
|
+
if (fs.existsSync(bakPath)) return bakPath;
|
|
379
|
+
} catch {}
|
|
380
|
+
return '';
|
|
381
|
+
}
|
|
382
|
+
|
|
383
|
+
function _fallbackTranscriptPathFromTask(task, provider, opts = {}) {
|
|
384
|
+
if (!task || typeof task !== 'object') return '';
|
|
385
|
+
const agentId =
|
|
386
|
+
_asString(task.agent_session_id) ||
|
|
387
|
+
_asString(task.claude_session_id) ||
|
|
388
|
+
_asString(task.resume_session_id) ||
|
|
389
|
+
_resumeIdFromArgs(task.args);
|
|
390
|
+
if (!agentId) return '';
|
|
391
|
+
|
|
392
|
+
const explicitProjectDir = _asString(task.agent_project_dir || task.claude_project_dir || task.project_dir);
|
|
393
|
+
if (explicitProjectDir) {
|
|
394
|
+
const explicit = _existingTranscriptPath(path.resolve(explicitProjectDir, `${agentId}.jsonl`));
|
|
395
|
+
if (explicit) return explicit;
|
|
396
|
+
}
|
|
397
|
+
|
|
398
|
+
if (provider === 'claude') {
|
|
399
|
+
const cwd = _asString(task.worktree_path || task.cwd);
|
|
400
|
+
if (!cwd) return '';
|
|
401
|
+
const homeDir = path.resolve(opts.homeDir || process.env.HOME || '');
|
|
402
|
+
if (!homeDir) return '';
|
|
403
|
+
const projectEntry = _encodeProjectEntry(_canonicalClaudeCwd(cwd));
|
|
404
|
+
return _existingTranscriptPath(path.join(homeDir, '.claude', 'projects', projectEntry, `${agentId}.jsonl`));
|
|
405
|
+
}
|
|
406
|
+
|
|
407
|
+
return '';
|
|
408
|
+
}
|
|
409
|
+
|
|
410
|
+
function detectRestoreInterruption({ db, task, agentType, jsonlPath, now, homeDir } = {}) {
|
|
411
|
+
const provider = _asString(agentType || task?.type || task?.cmd).toLowerCase();
|
|
412
|
+
if (!['claude', 'codex'].includes(provider)) {
|
|
413
|
+
return { interrupted: false, reason: 'unsupported_provider' };
|
|
414
|
+
}
|
|
415
|
+
const ctmSessionId = _asString(task?.ctm_session_id || task?.session_id);
|
|
416
|
+
const row = jsonlPath ? null : _selectAgentRow(db, ctmSessionId, provider);
|
|
417
|
+
const filePath = _asString(jsonlPath || row?.jsonl_path || _fallbackTranscriptPathFromTask(task, provider, { homeDir }));
|
|
418
|
+
if (!filePath) return { interrupted: false, reason: 'missing_transcript_path' };
|
|
419
|
+
const tail = _readJsonlTail(path.resolve(filePath));
|
|
420
|
+
if (tail.error) return { interrupted: false, reason: tail.error, jsonlPath: filePath };
|
|
421
|
+
const result = detectRestoreInterruptionFromEntries(tail.entries, { now });
|
|
422
|
+
return {
|
|
423
|
+
...result,
|
|
424
|
+
jsonlPath: filePath,
|
|
425
|
+
fileSize: tail.size || 0,
|
|
426
|
+
fileMtimeMs: tail.mtimeMs || 0,
|
|
427
|
+
agentSessionId: row?.agent_session_id || '',
|
|
428
|
+
};
|
|
429
|
+
}
|
|
430
|
+
|
|
431
|
+
module.exports = {
|
|
432
|
+
detectRestoreInterruption,
|
|
433
|
+
detectRestoreInterruptionFromEntries,
|
|
434
|
+
_private: {
|
|
435
|
+
_readJsonlTail,
|
|
436
|
+
_semanticEvent,
|
|
437
|
+
_lastSemanticEvent,
|
|
438
|
+
},
|
|
439
|
+
};
|
|
@@ -0,0 +1,13 @@
|
|
|
1
|
+
'use strict';
|
|
2
|
+
|
|
3
|
+
function shouldDeferStartupTaskRestore(task, restoredOrdinal, fastSessionCount) {
|
|
4
|
+
if (task?.type === 'walle') return false;
|
|
5
|
+
const ordinal = Number(restoredOrdinal) || 0;
|
|
6
|
+
const fastCount = Number(fastSessionCount) || 0;
|
|
7
|
+
if (ordinal <= fastCount) return false;
|
|
8
|
+
return true;
|
|
9
|
+
}
|
|
10
|
+
|
|
11
|
+
module.exports = {
|
|
12
|
+
shouldDeferStartupTaskRestore,
|
|
13
|
+
};
|
|
@@ -0,0 +1,74 @@
|
|
|
1
|
+
'use strict';
|
|
2
|
+
|
|
3
|
+
// Background auto-resume drain, extracted from server.js _autoResumeDeferredSessions
|
|
4
|
+
// so the batch + coalesced-broadcast invariant is unit-testable without booting the
|
|
5
|
+
// whole server.
|
|
6
|
+
//
|
|
7
|
+
// Drains an ordered list of deferred-session ids in batches of `batchSize`. For each
|
|
8
|
+
// id it re-checks `canResume` (so a tab the user clicked mid-pass is skipped, not
|
|
9
|
+
// double-started) and, if resumable, awaits `startSession`. After each batch it emits
|
|
10
|
+
// EXACTLY ONE coalesced `broadcast()` — but only when >=1 session actually resumed in
|
|
11
|
+
// that batch (an all-skipped batch broadcasts nothing). A `scheduleGap(fn)` callback
|
|
12
|
+
// schedules the next tick after the inter-batch gap; `onComplete` fires once drained.
|
|
13
|
+
//
|
|
14
|
+
// batchSize=1 reproduces the legacy serial cadence (1 resume -> 1 broadcast -> gap):
|
|
15
|
+
// every resumed session is its own batch, so broadcasts == resumes.
|
|
16
|
+
//
|
|
17
|
+
// All effects are injected (no module state), so a test can drive it synchronously
|
|
18
|
+
// with spies + a `scheduleGap` that calls its fn immediately.
|
|
19
|
+
//
|
|
20
|
+
// deps:
|
|
21
|
+
// ids string[] ordered deferred ids (snapshotted once)
|
|
22
|
+
// batchSize number >=1; ids drained per tick
|
|
23
|
+
// getSession (id) => session|undefined
|
|
24
|
+
// canResume (session) => boolean re-checked per id
|
|
25
|
+
// startSession async (session) => any resumes one session
|
|
26
|
+
// onError (id, err) => void non-fatal per-session resume error (optional)
|
|
27
|
+
// broadcast () => void one coalesced list refresh per non-empty batch
|
|
28
|
+
// scheduleGap (fn) => void schedule next tick after the inter-batch gap
|
|
29
|
+
// onComplete () => void all ids drained (optional)
|
|
30
|
+
function drainDeferredInBatches(deps) {
|
|
31
|
+
const {
|
|
32
|
+
ids,
|
|
33
|
+
batchSize,
|
|
34
|
+
getSession,
|
|
35
|
+
canResume,
|
|
36
|
+
startSession,
|
|
37
|
+
onError,
|
|
38
|
+
broadcast,
|
|
39
|
+
scheduleGap,
|
|
40
|
+
onComplete,
|
|
41
|
+
} = deps || {};
|
|
42
|
+
const list = Array.isArray(ids) ? ids : [];
|
|
43
|
+
const size = Math.max(1, Math.floor(Number(batchSize)) || 1);
|
|
44
|
+
let i = 0;
|
|
45
|
+
|
|
46
|
+
async function tick() {
|
|
47
|
+
if (i >= list.length) {
|
|
48
|
+
if (typeof onComplete === 'function') onComplete();
|
|
49
|
+
return;
|
|
50
|
+
}
|
|
51
|
+
let resumedInBatch = 0;
|
|
52
|
+
for (let b = 0; b < size && i < list.length; b++) {
|
|
53
|
+
const id = list[i++];
|
|
54
|
+
const s = getSession(id);
|
|
55
|
+
if (canResume(s)) {
|
|
56
|
+
try {
|
|
57
|
+
await startSession(s);
|
|
58
|
+
resumedInBatch++;
|
|
59
|
+
} catch (e) {
|
|
60
|
+
if (typeof onError === 'function') onError(id, e);
|
|
61
|
+
}
|
|
62
|
+
}
|
|
63
|
+
}
|
|
64
|
+
// One session-list refresh for the whole batch; an all-skipped batch is silent.
|
|
65
|
+
if (resumedInBatch > 0 && typeof broadcast === 'function') broadcast();
|
|
66
|
+
scheduleGap(tick);
|
|
67
|
+
}
|
|
68
|
+
|
|
69
|
+
// First tick is scheduled through the same gap (the original kicked off with
|
|
70
|
+
// setTimeout(resumeNext, gap) after the identity-cache prebuild settled).
|
|
71
|
+
scheduleGap(tick);
|
|
72
|
+
}
|
|
73
|
+
|
|
74
|
+
module.exports = { drainDeferredInBatches };
|