self-bench 0.3.6 → 0.3.8
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/.dockerignore +11 -1
- package/Dockerfile +9 -9
- package/Dockerfile.sandbox +5 -4
- package/README.md +27 -12
- package/bun.lock +462 -95
- package/compose.yaml +64 -20
- package/dist/agent-feed.d.ts +8 -0
- package/dist/agent-feed.d.ts.map +1 -0
- package/dist/agent-feed.js +56 -0
- package/dist/agent-feed.js.map +1 -0
- package/dist/api/http.d.ts +18 -0
- package/dist/api/http.d.ts.map +1 -0
- package/dist/api/http.js +111 -0
- package/dist/api/http.js.map +1 -0
- package/dist/api/run-request.d.ts +37 -0
- package/dist/api/run-request.d.ts.map +1 -0
- package/dist/api/run-request.js +59 -0
- package/dist/api/run-request.js.map +1 -0
- package/dist/api/status.d.ts +4 -0
- package/dist/api/status.d.ts.map +1 -0
- package/dist/api/status.js +36 -0
- package/dist/api/status.js.map +1 -0
- package/dist/api/viewer-routes.d.ts +9 -0
- package/dist/api/viewer-routes.d.ts.map +1 -0
- package/dist/api/viewer-routes.js +98 -0
- package/dist/api/viewer-routes.js.map +1 -0
- package/dist/api-main.js +5 -1
- package/dist/api-main.js.map +1 -1
- package/dist/api.d.ts +7 -25
- package/dist/api.d.ts.map +1 -1
- package/dist/api.js +80 -148
- package/dist/api.js.map +1 -1
- package/dist/archive.d.ts +14 -0
- package/dist/archive.d.ts.map +1 -0
- package/dist/archive.js +114 -0
- package/dist/archive.js.map +1 -0
- package/dist/artifacts/common.d.ts +15 -0
- package/dist/artifacts/common.d.ts.map +1 -0
- package/dist/artifacts/common.js +58 -0
- package/dist/artifacts/common.js.map +1 -0
- package/dist/artifacts/gcs.d.ts +18 -0
- package/dist/artifacts/gcs.d.ts.map +1 -0
- package/dist/artifacts/gcs.js +178 -0
- package/dist/artifacts/gcs.js.map +1 -0
- package/dist/artifacts/local.d.ts +17 -0
- package/dist/artifacts/local.d.ts.map +1 -0
- package/dist/artifacts/local.js +205 -0
- package/dist/artifacts/local.js.map +1 -0
- package/dist/artifacts/types.d.ts +18 -0
- package/dist/artifacts/types.d.ts.map +1 -0
- package/dist/artifacts/types.js +2 -0
- package/dist/artifacts/types.js.map +1 -0
- package/dist/artifacts.d.ts +5 -28
- package/dist/artifacts.d.ts.map +1 -1
- package/dist/artifacts.js +5 -302
- package/dist/artifacts.js.map +1 -1
- package/dist/audit.d.ts.map +1 -1
- package/dist/audit.js +8 -2
- package/dist/audit.js.map +1 -1
- package/dist/auth/api-key-routes.d.ts +17 -0
- package/dist/auth/api-key-routes.d.ts.map +1 -0
- package/dist/auth/api-key-routes.js +57 -0
- package/dist/auth/api-key-routes.js.map +1 -0
- package/dist/auth/api-keys.d.ts +47 -0
- package/dist/auth/api-keys.d.ts.map +1 -0
- package/dist/auth/api-keys.js +102 -0
- package/dist/auth/api-keys.js.map +1 -0
- package/dist/auth/config.d.ts +17 -0
- package/dist/auth/config.d.ts.map +1 -0
- package/dist/auth/config.js +38 -0
- package/dist/auth/config.js.map +1 -0
- package/dist/auth/cookies.d.ts +12 -0
- package/dist/auth/cookies.d.ts.map +1 -0
- package/dist/auth/cookies.js +45 -0
- package/dist/auth/cookies.js.map +1 -0
- package/dist/auth/crypto.d.ts +19 -0
- package/dist/auth/crypto.d.ts.map +1 -0
- package/dist/auth/crypto.js +48 -0
- package/dist/auth/crypto.js.map +1 -0
- package/dist/auth/github.d.ts +37 -0
- package/dist/auth/github.d.ts.map +1 -0
- package/dist/auth/github.js +127 -0
- package/dist/auth/github.js.map +1 -0
- package/dist/auth/routes.d.ts +41 -0
- package/dist/auth/routes.d.ts.map +1 -0
- package/dist/auth/routes.js +153 -0
- package/dist/auth/routes.js.map +1 -0
- package/dist/auth/session-expired.d.ts +4 -0
- package/dist/auth/session-expired.d.ts.map +1 -0
- package/dist/auth/session-expired.js +17 -0
- package/dist/auth/session-expired.js.map +1 -0
- package/dist/auth/session.d.ts +23 -0
- package/dist/auth/session.d.ts.map +1 -0
- package/dist/auth/session.js +56 -0
- package/dist/auth/session.js.map +1 -0
- package/dist/auth/users.d.ts +49 -0
- package/dist/auth/users.d.ts.map +1 -0
- package/dist/auth/users.js +125 -0
- package/dist/auth/users.js.map +1 -0
- package/dist/build-metadata.js +1 -1
- package/dist/cli/api-client.d.ts +8 -0
- package/dist/cli/api-client.d.ts.map +1 -0
- package/dist/cli/api-client.js +71 -0
- package/dist/cli/api-client.js.map +1 -0
- package/dist/cli/associate.d.ts +2 -0
- package/dist/cli/associate.d.ts.map +1 -0
- package/dist/cli/associate.js +51 -0
- package/dist/cli/associate.js.map +1 -0
- package/dist/cli/help.d.ts +2 -0
- package/dist/cli/help.d.ts.map +1 -0
- package/dist/cli/help.js +52 -0
- package/dist/cli/help.js.map +1 -0
- package/dist/cli/main.d.ts +2 -0
- package/dist/cli/main.d.ts.map +1 -0
- package/dist/cli/main.js +55 -0
- package/dist/cli/main.js.map +1 -0
- package/dist/cli/replay.d.ts +3 -0
- package/dist/cli/replay.d.ts.map +1 -0
- package/dist/cli/replay.js +59 -0
- package/dist/cli/replay.js.map +1 -0
- package/dist/cli/repository.d.ts +6 -0
- package/dist/cli/repository.d.ts.map +1 -0
- package/dist/cli/repository.js +36 -0
- package/dist/cli/repository.js.map +1 -0
- package/dist/cli/run.d.ts +2 -0
- package/dist/cli/run.d.ts.map +1 -0
- package/dist/cli/run.js +93 -0
- package/dist/cli/run.js.map +1 -0
- package/dist/cli/stack-environment.d.ts +26 -0
- package/dist/cli/stack-environment.d.ts.map +1 -0
- package/dist/cli/stack-environment.js +86 -0
- package/dist/cli/stack-environment.js.map +1 -0
- package/dist/cli/stack.d.ts +6 -0
- package/dist/cli/stack.d.ts.map +1 -0
- package/dist/cli/stack.js +131 -0
- package/dist/cli/stack.js.map +1 -0
- package/dist/cli/values.d.ts +8 -0
- package/dist/cli/values.d.ts.map +1 -0
- package/dist/cli/values.js +30 -0
- package/dist/cli/values.js.map +1 -0
- package/dist/cli/view.d.ts +2 -0
- package/dist/cli/view.d.ts.map +1 -0
- package/dist/cli/view.js +40 -0
- package/dist/cli/view.js.map +1 -0
- package/dist/cli.js +2 -429
- package/dist/cli.js.map +1 -1
- package/dist/config.d.ts +16 -2
- package/dist/config.d.ts.map +1 -1
- package/dist/config.js +60 -22
- package/dist/config.js.map +1 -1
- package/dist/contracts/common.d.ts +22 -0
- package/dist/contracts/common.d.ts.map +1 -0
- package/dist/contracts/common.js +29 -0
- package/dist/contracts/common.js.map +1 -0
- package/dist/contracts/results.d.ts +66 -0
- package/dist/contracts/results.d.ts.map +1 -0
- package/dist/contracts/results.js +2 -0
- package/dist/contracts/results.js.map +1 -0
- package/dist/contracts/run.d.ts +145 -0
- package/dist/contracts/run.d.ts.map +1 -0
- package/dist/contracts/run.js +74 -0
- package/dist/contracts/run.js.map +1 -0
- package/dist/contracts/task.d.ts +232 -0
- package/dist/contracts/task.d.ts.map +1 -0
- package/dist/contracts/task.js +198 -0
- package/dist/contracts/task.js.map +1 -0
- package/dist/contracts/verify.d.ts +331 -0
- package/dist/contracts/verify.d.ts.map +1 -0
- package/dist/contracts/verify.js +84 -0
- package/dist/contracts/verify.js.map +1 -0
- package/dist/contracts.d.ts +5 -243
- package/dist/contracts.d.ts.map +1 -1
- package/dist/contracts.js +5 -128
- package/dist/contracts.js.map +1 -1
- package/dist/db/client.d.ts +12 -0
- package/dist/db/client.d.ts.map +1 -0
- package/dist/db/client.js +16 -0
- package/dist/db/client.js.map +1 -0
- package/dist/db/schema.d.ts +1256 -0
- package/dist/db/schema.d.ts.map +1 -0
- package/dist/db/schema.js +128 -0
- package/dist/db/schema.js.map +1 -0
- package/dist/environment.d.ts +9 -0
- package/dist/environment.d.ts.map +1 -0
- package/dist/environment.js +101 -0
- package/dist/environment.js.map +1 -0
- package/dist/evaluation/account.d.ts +30 -0
- package/dist/evaluation/account.d.ts.map +1 -0
- package/dist/evaluation/account.js +34 -0
- package/dist/evaluation/account.js.map +1 -0
- package/dist/evaluation/activities.d.ts +9 -0
- package/dist/evaluation/activities.d.ts.map +1 -0
- package/dist/evaluation/activities.js +40 -0
- package/dist/evaluation/activities.js.map +1 -0
- package/dist/evaluation/catalog-pricing.d.ts +3 -0
- package/dist/evaluation/catalog-pricing.d.ts.map +1 -0
- package/dist/evaluation/catalog-pricing.js +37 -0
- package/dist/evaluation/catalog-pricing.js.map +1 -0
- package/dist/evaluation/catalog.d.ts +17 -0
- package/dist/evaluation/catalog.d.ts.map +1 -0
- package/dist/evaluation/catalog.js +126 -0
- package/dist/evaluation/catalog.js.map +1 -0
- package/dist/evaluation/codex-login-process.d.ts +12 -0
- package/dist/evaluation/codex-login-process.d.ts.map +1 -0
- package/dist/evaluation/codex-login-process.js +121 -0
- package/dist/evaluation/codex-login-process.js.map +1 -0
- package/dist/evaluation/codex-login-routes.d.ts +5 -0
- package/dist/evaluation/codex-login-routes.d.ts.map +1 -0
- package/dist/evaluation/codex-login-routes.js +55 -0
- package/dist/evaluation/codex-login-routes.js.map +1 -0
- package/dist/evaluation/codex-login.d.ts +42 -0
- package/dist/evaluation/codex-login.d.ts.map +1 -0
- package/dist/evaluation/codex-login.js +143 -0
- package/dist/evaluation/codex-login.js.map +1 -0
- package/dist/evaluation/comparisons.d.ts +65 -0
- package/dist/evaluation/comparisons.d.ts.map +1 -0
- package/dist/evaluation/comparisons.js +161 -0
- package/dist/evaluation/comparisons.js.map +1 -0
- package/dist/evaluation/config.d.ts +43 -0
- package/dist/evaluation/config.d.ts.map +1 -0
- package/dist/evaluation/config.js +116 -0
- package/dist/evaluation/config.js.map +1 -0
- package/dist/evaluation/cost.d.ts +3 -0
- package/dist/evaluation/cost.d.ts.map +1 -0
- package/dist/evaluation/cost.js +95 -0
- package/dist/evaluation/cost.js.map +1 -0
- package/dist/evaluation/credential-execution.d.ts +10 -0
- package/dist/evaluation/credential-execution.d.ts.map +1 -0
- package/dist/evaluation/credential-execution.js +65 -0
- package/dist/evaluation/credential-execution.js.map +1 -0
- package/dist/evaluation/credentials.d.ts +40 -0
- package/dist/evaluation/credentials.d.ts.map +1 -0
- package/dist/evaluation/credentials.js +146 -0
- package/dist/evaluation/credentials.js.map +1 -0
- package/dist/evaluation/encrypted-records.d.ts +15 -0
- package/dist/evaluation/encrypted-records.d.ts.map +1 -0
- package/dist/evaluation/encrypted-records.js +57 -0
- package/dist/evaluation/encrypted-records.js.map +1 -0
- package/dist/evaluation/gateway-execution.d.ts +18 -0
- package/dist/evaluation/gateway-execution.d.ts.map +1 -0
- package/dist/evaluation/gateway-execution.js +38 -0
- package/dist/evaluation/gateway-execution.js.map +1 -0
- package/dist/evaluation/harbor-cost.d.ts +3 -0
- package/dist/evaluation/harbor-cost.d.ts.map +1 -0
- package/dist/evaluation/harbor-cost.js +39 -0
- package/dist/evaluation/harbor-cost.js.map +1 -0
- package/dist/evaluation/harbor_e2b.py +20 -0
- package/dist/evaluation/harbor_gateway.py +22 -0
- package/dist/evaluation/harnesses.d.ts +8 -0
- package/dist/evaluation/harnesses.d.ts.map +1 -0
- package/dist/evaluation/harnesses.js +10 -0
- package/dist/evaluation/harnesses.js.map +1 -0
- package/dist/evaluation/model-options.d.ts +9 -0
- package/dist/evaluation/model-options.d.ts.map +1 -0
- package/dist/evaluation/model-options.js +81 -0
- package/dist/evaluation/model-options.js.map +1 -0
- package/dist/evaluation/org-credential-routes.d.ts +5 -0
- package/dist/evaluation/org-credential-routes.d.ts.map +1 -0
- package/dist/evaluation/org-credential-routes.js +56 -0
- package/dist/evaluation/org-credential-routes.js.map +1 -0
- package/dist/evaluation/org-records.d.ts +4 -0
- package/dist/evaluation/org-records.d.ts.map +1 -0
- package/dist/evaluation/org-records.js +14 -0
- package/dist/evaluation/org-records.js.map +1 -0
- package/dist/evaluation/output.d.ts +9 -0
- package/dist/evaluation/output.d.ts.map +1 -0
- package/dist/evaluation/output.js +171 -0
- package/dist/evaluation/output.js.map +1 -0
- package/dist/evaluation/platform-routes.d.ts +5 -0
- package/dist/evaluation/platform-routes.d.ts.map +1 -0
- package/dist/evaluation/platform-routes.js +123 -0
- package/dist/evaluation/platform-routes.js.map +1 -0
- package/dist/evaluation/profile-routes.d.ts +16 -0
- package/dist/evaluation/profile-routes.d.ts.map +1 -0
- package/dist/evaluation/profile-routes.js +47 -0
- package/dist/evaluation/profile-routes.js.map +1 -0
- package/dist/evaluation/profiles.d.ts +48 -0
- package/dist/evaluation/profiles.d.ts.map +1 -0
- package/dist/evaluation/profiles.js +134 -0
- package/dist/evaluation/profiles.js.map +1 -0
- package/dist/evaluation/providers.d.ts +11 -0
- package/dist/evaluation/providers.d.ts.map +1 -0
- package/dist/evaluation/providers.js +30 -0
- package/dist/evaluation/providers.js.map +1 -0
- package/dist/evaluation/routes.d.ts +23 -0
- package/dist/evaluation/routes.d.ts.map +1 -0
- package/dist/evaluation/routes.js +212 -0
- package/dist/evaluation/routes.js.map +1 -0
- package/dist/evaluation/runner.d.ts +16 -0
- package/dist/evaluation/runner.d.ts.map +1 -0
- package/dist/evaluation/runner.js +241 -0
- package/dist/evaluation/runner.js.map +1 -0
- package/dist/evaluation/store.d.ts +8 -0
- package/dist/evaluation/store.d.ts.map +1 -0
- package/dist/evaluation/store.js +60 -0
- package/dist/evaluation/store.js.map +1 -0
- package/dist/evaluation/types.d.ts +98 -0
- package/dist/evaluation/types.d.ts.map +1 -0
- package/dist/evaluation/types.js +2 -0
- package/dist/evaluation/types.js.map +1 -0
- package/dist/evaluation/worker-main.d.ts +2 -0
- package/dist/evaluation/worker-main.d.ts.map +1 -0
- package/dist/evaluation/worker-main.js +26 -0
- package/dist/evaluation/worker-main.js.map +1 -0
- package/dist/evaluation/worker-records.d.ts +5 -0
- package/dist/evaluation/worker-records.d.ts.map +1 -0
- package/dist/evaluation/worker-records.js +11 -0
- package/dist/evaluation/worker-records.js.map +1 -0
- package/dist/evaluation/workflow.d.ts +3 -0
- package/dist/evaluation/workflow.d.ts.map +1 -0
- package/dist/evaluation/workflow.js +20 -0
- package/dist/evaluation/workflow.js.map +1 -0
- package/dist/execution-environment.d.ts +3 -0
- package/dist/execution-environment.d.ts.map +1 -0
- package/dist/execution-environment.js +9 -0
- package/dist/execution-environment.js.map +1 -0
- package/dist/extension-authoring.bundle.js +347 -0
- package/dist/extension-verifier.bundle.js +73 -0
- package/dist/extensions/authoring.d.ts.map +1 -1
- package/dist/extensions/authoring.js +70 -59
- package/dist/extensions/authoring.js.map +1 -1
- package/dist/extensions/shared/deliverable.d.ts +33 -0
- package/dist/extensions/shared/deliverable.d.ts.map +1 -0
- package/dist/extensions/shared/deliverable.js +123 -0
- package/dist/extensions/shared/deliverable.js.map +1 -0
- package/dist/extensions/shared/mailbox.d.ts +33 -0
- package/dist/extensions/shared/mailbox.d.ts.map +1 -0
- package/dist/extensions/shared/mailbox.js +102 -0
- package/dist/extensions/shared/mailbox.js.map +1 -0
- package/dist/extensions/shared/schemas.d.ts +90 -0
- package/dist/extensions/shared/schemas.d.ts.map +1 -0
- package/dist/extensions/shared/schemas.js +71 -0
- package/dist/extensions/shared/schemas.js.map +1 -0
- package/dist/extensions/shared/static-check.d.ts +38 -0
- package/dist/extensions/shared/static-check.d.ts.map +1 -0
- package/dist/extensions/shared/static-check.js +52 -0
- package/dist/extensions/shared/static-check.js.map +1 -0
- package/dist/extensions/verifier.d.ts +4 -0
- package/dist/extensions/verifier.d.ts.map +1 -0
- package/dist/extensions/verifier.js +70 -0
- package/dist/extensions/verifier.js.map +1 -0
- package/dist/github.d.ts +1 -1
- package/dist/github.d.ts.map +1 -1
- package/dist/github.js +3 -3
- package/dist/github.js.map +1 -1
- package/dist/harbor-environment.d.ts +13 -1
- package/dist/harbor-environment.d.ts.map +1 -1
- package/dist/harbor-environment.js +37 -4
- package/dist/harbor-environment.js.map +1 -1
- package/dist/harbor-results.d.ts +2 -0
- package/dist/harbor-results.d.ts.map +1 -1
- package/dist/harbor-results.js +2 -0
- package/dist/harbor-results.js.map +1 -1
- package/dist/harbor-task/compiler.d.ts +10 -0
- package/dist/harbor-task/compiler.d.ts.map +1 -0
- package/dist/harbor-task/compiler.js +174 -0
- package/dist/harbor-task/compiler.js.map +1 -0
- package/dist/harbor-task/constants.d.ts +3 -0
- package/dist/harbor-task/constants.d.ts.map +1 -0
- package/dist/harbor-task/constants.js +3 -0
- package/dist/harbor-task/constants.js.map +1 -0
- package/dist/harbor-task/dependencies.d.ts +3 -0
- package/dist/harbor-task/dependencies.d.ts.map +1 -0
- package/dist/harbor-task/dependencies.js +20 -0
- package/dist/harbor-task/dependencies.js.map +1 -0
- package/dist/harbor-task/paths.d.ts +6 -0
- package/dist/harbor-task/paths.d.ts.map +1 -0
- package/dist/harbor-task/paths.js +37 -0
- package/dist/harbor-task/paths.js.map +1 -0
- package/dist/harbor-task/render.d.ts +11 -0
- package/dist/harbor-task/render.d.ts.map +1 -0
- package/dist/harbor-task/render.js +194 -0
- package/dist/harbor-task/render.js.map +1 -0
- package/dist/harbor-task/runtime/command.sh +26 -0
- package/dist/harbor-task/runtime/junit.py +69 -0
- package/dist/harbor-task/runtime-assets.d.ts +2 -0
- package/dist/harbor-task/runtime-assets.d.ts.map +1 -0
- package/dist/harbor-task/runtime-assets.js +8 -0
- package/dist/harbor-task/runtime-assets.js.map +1 -0
- package/dist/harbor-task/test-patch.d.ts +4 -0
- package/dist/harbor-task/test-patch.d.ts.map +1 -0
- package/dist/harbor-task/test-patch.js +15 -0
- package/dist/harbor-task/test-patch.js.map +1 -0
- package/dist/harbor-task/verifier.d.ts +4 -0
- package/dist/harbor-task/verifier.d.ts.map +1 -0
- package/dist/harbor-task/verifier.js +105 -0
- package/dist/harbor-task/verifier.js.map +1 -0
- package/dist/harbor-task.d.ts +3 -11
- package/dist/harbor-task.d.ts.map +1 -1
- package/dist/harbor-task.js +2 -463
- package/dist/harbor-task.js.map +1 -1
- package/dist/log-excerpt.d.ts +15 -0
- package/dist/log-excerpt.d.ts.map +1 -0
- package/dist/log-excerpt.js +104 -0
- package/dist/log-excerpt.js.map +1 -0
- package/dist/patch-check.d.ts +26 -0
- package/dist/patch-check.d.ts.map +1 -0
- package/dist/patch-check.js +137 -0
- package/dist/patch-check.js.map +1 -0
- package/dist/pi-event-feed.d.ts +16 -0
- package/dist/pi-event-feed.d.ts.map +1 -0
- package/dist/pi-event-feed.js +86 -0
- package/dist/pi-event-feed.js.map +1 -0
- package/dist/pi-session.d.ts +32 -0
- package/dist/pi-session.d.ts.map +1 -0
- package/dist/pi-session.js +175 -0
- package/dist/pi-session.js.map +1 -0
- package/dist/process.d.ts +5 -0
- package/dist/process.d.ts.map +1 -1
- package/dist/process.js +31 -9
- package/dist/process.js.map +1 -1
- package/dist/provenance/github.d.ts +4 -0
- package/dist/provenance/github.d.ts.map +1 -0
- package/dist/provenance/github.js +65 -0
- package/dist/provenance/github.js.map +1 -0
- package/dist/provenance/local.d.ts +4 -0
- package/dist/provenance/local.d.ts.map +1 -0
- package/dist/provenance/local.js +82 -0
- package/dist/provenance/local.js.map +1 -0
- package/dist/provenance/redact.d.ts +2 -0
- package/dist/provenance/redact.d.ts.map +1 -0
- package/dist/provenance/redact.js +28 -0
- package/dist/provenance/redact.js.map +1 -0
- package/dist/provenance/selection.d.ts +4 -0
- package/dist/provenance/selection.d.ts.map +1 -0
- package/dist/provenance/selection.js +29 -0
- package/dist/provenance/selection.js.map +1 -0
- package/dist/provenance/session.d.ts +3 -0
- package/dist/provenance/session.d.ts.map +1 -0
- package/dist/provenance/session.js +214 -0
- package/dist/provenance/session.js.map +1 -0
- package/dist/provenance/shared.d.ts +7 -0
- package/dist/provenance/shared.d.ts.map +1 -0
- package/dist/provenance/shared.js +10 -0
- package/dist/provenance/shared.js.map +1 -0
- package/dist/provenance/types.d.ts +34 -0
- package/dist/provenance/types.d.ts.map +1 -0
- package/dist/provenance/types.js +22 -0
- package/dist/provenance/types.js.map +1 -0
- package/dist/provenance-associations/apply.d.ts +3 -0
- package/dist/provenance-associations/apply.d.ts.map +1 -0
- package/dist/provenance-associations/apply.js +66 -0
- package/dist/provenance-associations/apply.js.map +1 -0
- package/dist/provenance-associations/shared.d.ts +45 -0
- package/dist/provenance-associations/shared.d.ts.map +1 -0
- package/dist/provenance-associations/shared.js +42 -0
- package/dist/provenance-associations/shared.js.map +1 -0
- package/dist/provenance-associations.d.ts +3 -30
- package/dist/provenance-associations.d.ts.map +1 -1
- package/dist/provenance-associations.js +4 -103
- package/dist/provenance-associations.js.map +1 -1
- package/dist/provenance.d.ts +7 -41
- package/dist/provenance.d.ts.map +1 -1
- package/dist/provenance.js +6 -432
- package/dist/provenance.js.map +1 -1
- package/dist/providers.d.ts +25 -2
- package/dist/providers.d.ts.map +1 -1
- package/dist/providers.js +23 -2
- package/dist/providers.js.map +1 -1
- package/dist/repair.d.ts +1 -6
- package/dist/repair.d.ts.map +1 -1
- package/dist/repair.js +17 -27
- package/dist/repair.js.map +1 -1
- package/dist/review/assets/index-BHHztQ9x.js +1679 -0
- package/dist/review/assets/index-KQ1eFQ0t.css +1 -0
- package/dist/review/index.html +11 -8
- package/dist/runtime/command.sh +26 -0
- package/dist/runtime/junit.py +69 -0
- package/dist/sandbox/contracts.d.ts +36 -1
- package/dist/sandbox/contracts.d.ts.map +1 -1
- package/dist/sandbox/contracts.js +3 -0
- package/dist/sandbox/contracts.js.map +1 -1
- package/dist/sandbox/index.d.ts +1 -1
- package/dist/sandbox/index.d.ts.map +1 -1
- package/dist/sandbox/index.js +13 -2
- package/dist/sandbox/index.js.map +1 -1
- package/dist/sandbox/live.d.ts +18 -0
- package/dist/sandbox/live.d.ts.map +1 -0
- package/dist/sandbox/live.js +55 -0
- package/dist/sandbox/live.js.map +1 -0
- package/dist/sandbox/output-retry.d.ts +16 -0
- package/dist/sandbox/output-retry.d.ts.map +1 -0
- package/dist/sandbox/output-retry.js +36 -0
- package/dist/sandbox/output-retry.js.map +1 -0
- package/dist/sandbox/programs/author.js +16 -6
- package/dist/sandbox/programs/author.js.map +1 -1
- package/dist/sandbox/programs/check.d.ts +3 -0
- package/dist/sandbox/programs/check.d.ts.map +1 -0
- package/dist/sandbox/programs/check.js +76 -0
- package/dist/sandbox/programs/check.js.map +1 -0
- package/dist/sandbox/programs/prepare-task.d.ts +12 -0
- package/dist/sandbox/programs/prepare-task.d.ts.map +1 -0
- package/dist/sandbox/programs/{prepare-repair.js → prepare-task.js} +11 -9
- package/dist/sandbox/programs/prepare-task.js.map +1 -0
- package/dist/sandbox/programs/verifier.d.ts +3 -0
- package/dist/sandbox/programs/verifier.d.ts.map +1 -0
- package/dist/sandbox/programs/verifier.js +9 -0
- package/dist/sandbox/programs/verifier.js.map +1 -0
- package/dist/sandbox/providers/docker/executor.d.ts +4 -1
- package/dist/sandbox/providers/docker/executor.d.ts.map +1 -1
- package/dist/sandbox/providers/docker/executor.js +75 -25
- package/dist/sandbox/providers/docker/executor.js.map +1 -1
- package/dist/sandbox/providers/e2b/bytes.d.ts +3 -0
- package/dist/sandbox/providers/e2b/bytes.d.ts.map +1 -0
- package/dist/sandbox/providers/e2b/bytes.js +8 -0
- package/dist/sandbox/providers/e2b/bytes.js.map +1 -0
- package/dist/sandbox/providers/e2b/cleanup.d.ts +8 -0
- package/dist/sandbox/providers/e2b/cleanup.d.ts.map +1 -0
- package/dist/sandbox/providers/e2b/cleanup.js +124 -0
- package/dist/sandbox/providers/e2b/cleanup.js.map +1 -0
- package/dist/sandbox/providers/e2b/command.d.ts +23 -0
- package/dist/sandbox/providers/e2b/command.d.ts.map +1 -0
- package/dist/sandbox/providers/e2b/command.js +160 -0
- package/dist/sandbox/providers/e2b/command.js.map +1 -0
- package/dist/sandbox/providers/e2b/config.d.ts +13 -0
- package/dist/sandbox/providers/e2b/config.d.ts.map +1 -0
- package/dist/sandbox/providers/e2b/config.js +2 -0
- package/dist/sandbox/providers/e2b/config.js.map +1 -0
- package/dist/sandbox/providers/e2b/executor.d.ts +15 -0
- package/dist/sandbox/providers/e2b/executor.d.ts.map +1 -0
- package/dist/sandbox/providers/e2b/executor.js +207 -0
- package/dist/sandbox/providers/e2b/executor.js.map +1 -0
- package/dist/sandbox/providers/e2b/lifecycle.d.ts +8 -0
- package/dist/sandbox/providers/e2b/lifecycle.d.ts.map +1 -0
- package/dist/sandbox/providers/e2b/lifecycle.js +37 -0
- package/dist/sandbox/providers/e2b/lifecycle.js.map +1 -0
- package/dist/sandbox/providers/e2b/outcome.d.ts +11 -0
- package/dist/sandbox/providers/e2b/outcome.d.ts.map +1 -0
- package/dist/sandbox/providers/e2b/outcome.js +72 -0
- package/dist/sandbox/providers/e2b/outcome.js.map +1 -0
- package/dist/sandbox/providers/e2b/stage-files.d.ts +10 -0
- package/dist/sandbox/providers/e2b/stage-files.d.ts.map +1 -0
- package/dist/sandbox/providers/e2b/stage-files.js +45 -0
- package/dist/sandbox/providers/e2b/stage-files.js.map +1 -0
- package/dist/sandbox/providers/e2b/startup.d.ts +10 -0
- package/dist/sandbox/providers/e2b/startup.d.ts.map +1 -0
- package/dist/sandbox/providers/e2b/startup.js +26 -0
- package/dist/sandbox/providers/e2b/startup.js.map +1 -0
- package/dist/sandbox/providers/e2b/types.d.ts +9 -0
- package/dist/sandbox/providers/e2b/types.d.ts.map +1 -0
- package/dist/sandbox/providers/e2b/types.js +2 -0
- package/dist/sandbox/providers/e2b/types.js.map +1 -0
- package/dist/sandbox/providers/e2b/validation.d.ts +14 -0
- package/dist/sandbox/providers/e2b/validation.d.ts.map +1 -0
- package/dist/sandbox/providers/e2b/validation.js +60 -0
- package/dist/sandbox/providers/e2b/validation.js.map +1 -0
- package/dist/{modal-auth.d.ts → sandbox/providers/modal/auth.d.ts} +1 -1
- package/dist/sandbox/providers/modal/auth.d.ts.map +1 -0
- package/dist/{modal-auth.js → sandbox/providers/modal/auth.js} +1 -1
- package/dist/sandbox/providers/modal/auth.js.map +1 -0
- package/dist/sandbox/providers/modal/executor.d.ts +4 -1
- package/dist/sandbox/providers/modal/executor.d.ts.map +1 -1
- package/dist/sandbox/providers/modal/executor.js +33 -9
- package/dist/sandbox/providers/modal/executor.js.map +1 -1
- package/dist/sandbox/providers/modal/live.d.ts +7 -0
- package/dist/sandbox/providers/modal/live.d.ts.map +1 -0
- package/dist/sandbox/providers/modal/live.js +50 -0
- package/dist/sandbox/providers/modal/live.js.map +1 -0
- package/dist/sandbox/providers/vercel/command.d.ts +6 -0
- package/dist/sandbox/providers/vercel/command.d.ts.map +1 -1
- package/dist/sandbox/providers/vercel/command.js +44 -11
- package/dist/sandbox/providers/vercel/command.js.map +1 -1
- package/dist/sandbox/providers/vercel/executor.d.ts +6 -7
- package/dist/sandbox/providers/vercel/executor.d.ts.map +1 -1
- package/dist/sandbox/providers/vercel/executor.js +21 -187
- package/dist/sandbox/providers/vercel/executor.js.map +1 -1
- package/dist/sandbox/providers/vercel/lifecycle.d.ts +13 -0
- package/dist/sandbox/providers/vercel/lifecycle.d.ts.map +1 -0
- package/dist/sandbox/providers/vercel/lifecycle.js +140 -0
- package/dist/sandbox/providers/vercel/lifecycle.js.map +1 -0
- package/dist/sandbox/providers/vercel/validation.d.ts +11 -0
- package/dist/sandbox/providers/vercel/validation.d.ts.map +1 -0
- package/dist/sandbox/providers/vercel/validation.js +30 -0
- package/dist/sandbox/providers/vercel/validation.js.map +1 -0
- package/dist/sandbox/remote-files.d.ts +11 -0
- package/dist/sandbox/remote-files.d.ts.map +1 -0
- package/dist/sandbox/remote-files.js +41 -0
- package/dist/sandbox/remote-files.js.map +1 -0
- package/dist/sandbox/request-validation.d.ts +5 -0
- package/dist/sandbox/request-validation.d.ts.map +1 -0
- package/dist/sandbox/request-validation.js +43 -0
- package/dist/sandbox/request-validation.js.map +1 -0
- package/dist/sandbox/supervisor.d.ts +42 -0
- package/dist/sandbox/supervisor.d.ts.map +1 -0
- package/dist/sandbox/supervisor.js +126 -0
- package/dist/sandbox/supervisor.js.map +1 -0
- package/dist/sandbox/timeout.d.ts +6 -1
- package/dist/sandbox/timeout.d.ts.map +1 -1
- package/dist/sandbox/timeout.js +12 -0
- package/dist/sandbox/timeout.js.map +1 -1
- package/dist/sandbox-author.bundle.js +47 -14879
- package/dist/{sandbox-validation-repair.bundle.js → sandbox-check.bundle.js} +7261 -1893
- package/dist/sandbox-verifier.bundle.js +307 -0
- package/dist/setup/e2b/build.d.ts +24 -0
- package/dist/setup/e2b/build.d.ts.map +1 -0
- package/dist/setup/e2b/build.js +46 -0
- package/dist/setup/e2b/build.js.map +1 -0
- package/dist/setup/e2b/cli.d.ts +2 -0
- package/dist/setup/e2b/cli.d.ts.map +1 -0
- package/dist/setup/e2b/cli.js +51 -0
- package/dist/setup/e2b/cli.js.map +1 -0
- package/dist/setup/e2b/index.d.ts +5 -0
- package/dist/setup/e2b/index.d.ts.map +1 -0
- package/dist/setup/e2b/index.js +4 -0
- package/dist/setup/e2b/index.js.map +1 -0
- package/dist/setup/e2b/template.d.ts +4 -0
- package/dist/setup/e2b/template.d.ts.map +1 -0
- package/dist/setup/e2b/template.js +53 -0
- package/dist/setup/e2b/template.js.map +1 -0
- package/dist/setup/vercel/cli-helpers.d.ts +22 -0
- package/dist/setup/vercel/cli-helpers.d.ts.map +1 -0
- package/dist/setup/vercel/cli-helpers.js +87 -0
- package/dist/setup/vercel/cli-helpers.js.map +1 -0
- package/dist/setup/vercel/cli-runner.d.ts +14 -0
- package/dist/setup/vercel/cli-runner.d.ts.map +1 -0
- package/dist/setup/vercel/cli-runner.js +24 -0
- package/dist/setup/vercel/cli-runner.js.map +1 -0
- package/dist/setup/vercel/cli-types.d.ts +24 -0
- package/dist/setup/vercel/cli-types.d.ts.map +1 -0
- package/dist/setup/vercel/cli-types.js +18 -0
- package/dist/setup/vercel/cli-types.js.map +1 -0
- package/dist/setup/vercel/cli.d.ts +5 -31
- package/dist/setup/vercel/cli.d.ts.map +1 -1
- package/dist/setup/vercel/cli.js +15 -125
- package/dist/setup/vercel/cli.js.map +1 -1
- package/dist/setup/vercel/profile-files.d.ts +6 -0
- package/dist/setup/vercel/profile-files.d.ts.map +1 -0
- package/dist/setup/vercel/profile-files.js +100 -0
- package/dist/setup/vercel/profile-files.js.map +1 -0
- package/dist/setup/vercel/profile-schema.d.ts +48 -0
- package/dist/setup/vercel/profile-schema.d.ts.map +1 -0
- package/dist/setup/vercel/profile-schema.js +39 -0
- package/dist/setup/vercel/profile-schema.js.map +1 -0
- package/dist/setup/vercel/profile.d.ts +3 -19
- package/dist/setup/vercel/profile.d.ts.map +1 -1
- package/dist/setup/vercel/profile.js +4 -130
- package/dist/setup/vercel/profile.js.map +1 -1
- package/dist/setup/vercel/selection.d.ts +18 -0
- package/dist/setup/vercel/selection.d.ts.map +1 -0
- package/dist/setup/vercel/selection.js +113 -0
- package/dist/setup/vercel/selection.js.map +1 -0
- package/dist/setup/vercel/setup-types.d.ts +34 -0
- package/dist/setup/vercel/setup-types.d.ts.map +1 -0
- package/dist/setup/vercel/setup-types.js +2 -0
- package/dist/setup/vercel/setup-types.js.map +1 -0
- package/dist/setup/vercel/setup.d.ts +3 -28
- package/dist/setup/vercel/setup.d.ts.map +1 -1
- package/dist/setup/vercel/setup.js +3 -111
- package/dist/setup/vercel/setup.js.map +1 -1
- package/dist/site/batch-activity.d.ts +5 -0
- package/dist/site/batch-activity.d.ts.map +1 -0
- package/dist/site/batch-activity.js +31 -0
- package/dist/site/batch-activity.js.map +1 -0
- package/dist/site/batch-progress.d.ts +16 -0
- package/dist/site/batch-progress.d.ts.map +1 -0
- package/dist/site/batch-progress.js +52 -0
- package/dist/site/batch-progress.js.map +1 -0
- package/dist/site/batch-routes.d.ts +30 -0
- package/dist/site/batch-routes.d.ts.map +1 -0
- package/dist/site/batch-routes.js +117 -0
- package/dist/site/batch-routes.js.map +1 -0
- package/dist/site/batch-start.d.ts +60 -0
- package/dist/site/batch-start.d.ts.map +1 -0
- package/dist/site/batch-start.js +39 -0
- package/dist/site/batch-start.js.map +1 -0
- package/dist/site/connected-repos.d.ts +25 -0
- package/dist/site/connected-repos.d.ts.map +1 -0
- package/dist/site/connected-repos.js +109 -0
- package/dist/site/connected-repos.js.map +1 -0
- package/dist/site/evaluation-start.d.ts +4 -0
- package/dist/site/evaluation-start.d.ts.map +1 -0
- package/dist/site/evaluation-start.js +19 -0
- package/dist/site/evaluation-start.js.map +1 -0
- package/dist/site/generation-config.d.ts +4 -0
- package/dist/site/generation-config.d.ts.map +1 -0
- package/dist/site/generation-config.js +16 -0
- package/dist/site/generation-config.js.map +1 -0
- package/dist/site/generation-credentials.d.ts +6 -0
- package/dist/site/generation-credentials.d.ts.map +1 -0
- package/dist/site/generation-credentials.js +101 -0
- package/dist/site/generation-credentials.js.map +1 -0
- package/dist/site/generation-run.d.ts +6 -0
- package/dist/site/generation-run.d.ts.map +1 -0
- package/dist/site/generation-run.js +22 -0
- package/dist/site/generation-run.js.map +1 -0
- package/dist/site/generation-settings.d.ts +80 -0
- package/dist/site/generation-settings.d.ts.map +1 -0
- package/dist/site/generation-settings.js +84 -0
- package/dist/site/generation-settings.js.map +1 -0
- package/dist/site/generation-subscription.d.ts +3 -0
- package/dist/site/generation-subscription.d.ts.map +1 -0
- package/dist/site/generation-subscription.js +38 -0
- package/dist/site/generation-subscription.js.map +1 -0
- package/dist/site/github-repos.d.ts +35 -0
- package/dist/site/github-repos.d.ts.map +1 -0
- package/dist/site/github-repos.js +121 -0
- package/dist/site/github-repos.js.map +1 -0
- package/dist/site/pr-candidate.d.ts +26 -0
- package/dist/site/pr-candidate.d.ts.map +1 -0
- package/dist/site/pr-candidate.js +101 -0
- package/dist/site/pr-candidate.js.map +1 -0
- package/dist/site/pr-list.d.ts +13 -0
- package/dist/site/pr-list.d.ts.map +1 -0
- package/dist/site/pr-list.js +24 -0
- package/dist/site/pr-list.js.map +1 -0
- package/dist/site/pr-routes.d.ts +26 -0
- package/dist/site/pr-routes.d.ts.map +1 -0
- package/dist/site/pr-routes.js +139 -0
- package/dist/site/pr-routes.js.map +1 -0
- package/dist/site/repo-store.d.ts +40 -0
- package/dist/site/repo-store.d.ts.map +1 -0
- package/dist/site/repo-store.js +74 -0
- package/dist/site/repo-store.js.map +1 -0
- package/dist/site/run-store.d.ts +17 -0
- package/dist/site/run-store.d.ts.map +1 -0
- package/dist/site/run-store.js +42 -0
- package/dist/site/run-store.js.map +1 -0
- package/dist/site/runtime.d.ts +27 -0
- package/dist/site/runtime.d.ts.map +1 -0
- package/dist/site/runtime.js +97 -0
- package/dist/site/runtime.js.map +1 -0
- package/dist/site/task-deletion.d.ts +7 -0
- package/dist/site/task-deletion.d.ts.map +1 -0
- package/dist/site/task-deletion.js +29 -0
- package/dist/site/task-deletion.js.map +1 -0
- package/dist/site/task-start-reservation.d.ts +4 -0
- package/dist/site/task-start-reservation.d.ts.map +1 -0
- package/dist/site/task-start-reservation.js +18 -0
- package/dist/site/task-start-reservation.js.map +1 -0
- package/dist/site/task-start.d.ts +33 -0
- package/dist/site/task-start.d.ts.map +1 -0
- package/dist/site/task-start.js +50 -0
- package/dist/site/task-start.js.map +1 -0
- package/dist/site/task-status.d.ts +36 -0
- package/dist/site/task-status.d.ts.map +1 -0
- package/dist/site/task-status.js +113 -0
- package/dist/site/task-status.js.map +1 -0
- package/dist/site/task-store.d.ts +78 -0
- package/dist/site/task-store.d.ts.map +1 -0
- package/dist/site/task-store.js +186 -0
- package/dist/site/task-store.js.map +1 -0
- package/dist/site/task-sync.d.ts +30 -0
- package/dist/site/task-sync.d.ts.map +1 -0
- package/dist/site/task-sync.js +121 -0
- package/dist/site/task-sync.js.map +1 -0
- package/dist/site/tasks.d.ts +47 -0
- package/dist/site/tasks.d.ts.map +1 -0
- package/dist/site/tasks.js +164 -0
- package/dist/site/tasks.js.map +1 -0
- package/dist/site/temporal-status.d.ts +8 -0
- package/dist/site/temporal-status.d.ts.map +1 -0
- package/dist/site/temporal-status.js +41 -0
- package/dist/site/temporal-status.js.map +1 -0
- package/dist/site/tenant.d.ts +4 -0
- package/dist/site/tenant.d.ts.map +1 -0
- package/dist/site/tenant.js +6 -0
- package/dist/site/tenant.js.map +1 -0
- package/dist/static-check.d.ts +38 -0
- package/dist/static-check.d.ts.map +1 -0
- package/dist/static-check.js +131 -0
- package/dist/static-check.js.map +1 -0
- package/dist/submission-hash.d.ts +18 -0
- package/dist/submission-hash.d.ts.map +1 -0
- package/dist/submission-hash.js +16 -0
- package/dist/submission-hash.js.map +1 -0
- package/dist/subscription-auth.d.ts +0 -7
- package/dist/subscription-auth.d.ts.map +1 -1
- package/dist/subscription-auth.js +3 -23
- package/dist/subscription-auth.js.map +1 -1
- package/dist/temporal/activities/agent-feed.d.ts +5 -0
- package/dist/temporal/activities/agent-feed.d.ts.map +1 -0
- package/dist/temporal/activities/agent-feed.js +41 -0
- package/dist/temporal/activities/agent-feed.js.map +1 -0
- package/dist/temporal/activities/agent-scripts.d.ts +16 -0
- package/dist/temporal/activities/agent-scripts.d.ts.map +1 -0
- package/dist/temporal/activities/agent-scripts.js +162 -0
- package/dist/temporal/activities/agent-scripts.js.map +1 -0
- package/dist/temporal/activities/authoring-round.d.ts +13 -0
- package/dist/temporal/activities/authoring-round.d.ts.map +1 -0
- package/dist/temporal/activities/authoring-round.js +160 -0
- package/dist/temporal/activities/authoring-round.js.map +1 -0
- package/dist/temporal/activities/constants.d.ts +5 -0
- package/dist/temporal/activities/constants.d.ts.map +1 -0
- package/dist/temporal/activities/constants.js +5 -0
- package/dist/temporal/activities/constants.js.map +1 -0
- package/dist/temporal/activities/discovery.d.ts +17 -0
- package/dist/temporal/activities/discovery.d.ts.map +1 -0
- package/dist/temporal/activities/discovery.js +153 -0
- package/dist/temporal/activities/discovery.js.map +1 -0
- package/dist/temporal/activities/drafts.d.ts +15 -0
- package/dist/temporal/activities/drafts.d.ts.map +1 -0
- package/dist/temporal/activities/drafts.js +51 -0
- package/dist/temporal/activities/drafts.js.map +1 -0
- package/dist/temporal/activities/excluded-source-prs.d.ts +9 -0
- package/dist/temporal/activities/excluded-source-prs.d.ts.map +1 -0
- package/dist/temporal/activities/excluded-source-prs.js +132 -0
- package/dist/temporal/activities/excluded-source-prs.js.map +1 -0
- package/dist/temporal/activities/export.d.ts +38 -0
- package/dist/temporal/activities/export.d.ts.map +1 -0
- package/dist/temporal/activities/export.js +79 -0
- package/dist/temporal/activities/export.js.map +1 -0
- package/dist/temporal/activities/factory.d.ts +5 -0
- package/dist/temporal/activities/factory.d.ts.map +1 -0
- package/dist/temporal/activities/factory.js +26 -0
- package/dist/temporal/activities/factory.js.map +1 -0
- package/dist/temporal/activities/gate-logs.d.ts +20 -0
- package/dist/temporal/activities/gate-logs.d.ts.map +1 -0
- package/dist/temporal/activities/gate-logs.js +62 -0
- package/dist/temporal/activities/gate-logs.js.map +1 -0
- package/dist/temporal/activities/generation-runtime.d.ts +6 -0
- package/dist/temporal/activities/generation-runtime.d.ts.map +1 -0
- package/dist/temporal/activities/generation-runtime.js +56 -0
- package/dist/temporal/activities/generation-runtime.js.map +1 -0
- package/dist/temporal/activities/harbor.d.ts +11 -0
- package/dist/temporal/activities/harbor.d.ts.map +1 -0
- package/dist/temporal/activities/harbor.js +112 -0
- package/dist/temporal/activities/harbor.js.map +1 -0
- package/dist/temporal/activities/modal-build-log.d.ts +12 -0
- package/dist/temporal/activities/modal-build-log.d.ts.map +1 -0
- package/dist/temporal/activities/modal-build-log.js +57 -0
- package/dist/temporal/activities/modal-build-log.js.map +1 -0
- package/dist/temporal/activities/prompts-authoring.d.ts +6 -0
- package/dist/temporal/activities/prompts-authoring.d.ts.map +1 -0
- package/dist/temporal/activities/prompts-authoring.js +87 -0
- package/dist/temporal/activities/prompts-authoring.js.map +1 -0
- package/dist/temporal/activities/prompts-verifier.d.ts +17 -0
- package/dist/temporal/activities/prompts-verifier.d.ts.map +1 -0
- package/dist/temporal/activities/prompts-verifier.js +66 -0
- package/dist/temporal/activities/prompts-verifier.js.map +1 -0
- package/dist/temporal/activities/provenance.d.ts +4 -0
- package/dist/temporal/activities/provenance.d.ts.map +1 -0
- package/dist/temporal/activities/provenance.js +18 -0
- package/dist/temporal/activities/provenance.js.map +1 -0
- package/dist/temporal/activities/replay.d.ts +20 -0
- package/dist/temporal/activities/replay.d.ts.map +1 -0
- package/dist/temporal/activities/replay.js +135 -0
- package/dist/temporal/activities/replay.js.map +1 -0
- package/dist/temporal/activities/round-outcome.d.ts +63 -0
- package/dist/temporal/activities/round-outcome.d.ts.map +1 -0
- package/dist/temporal/activities/round-outcome.js +127 -0
- package/dist/temporal/activities/round-outcome.js.map +1 -0
- package/dist/temporal/activities/runtime.d.ts +38 -0
- package/dist/temporal/activities/runtime.d.ts.map +1 -0
- package/dist/temporal/activities/runtime.js +146 -0
- package/dist/temporal/activities/runtime.js.map +1 -0
- package/dist/temporal/activities/session-verify.d.ts +45 -0
- package/dist/temporal/activities/session-verify.d.ts.map +1 -0
- package/dist/temporal/activities/session-verify.js +91 -0
- package/dist/temporal/activities/session-verify.js.map +1 -0
- package/dist/temporal/activities/submissions.d.ts +9 -0
- package/dist/temporal/activities/submissions.d.ts.map +1 -0
- package/dist/temporal/activities/submissions.js +28 -0
- package/dist/temporal/activities/submissions.js.map +1 -0
- package/dist/temporal/activities/task-compiler.d.ts +21 -0
- package/dist/temporal/activities/task-compiler.d.ts.map +1 -0
- package/dist/temporal/activities/task-compiler.js +63 -0
- package/dist/temporal/activities/task-compiler.js.map +1 -0
- package/dist/temporal/activities/types.d.ts +57 -0
- package/dist/temporal/activities/types.d.ts.map +1 -0
- package/dist/temporal/activities/types.js +2 -0
- package/dist/temporal/activities/types.js.map +1 -0
- package/dist/temporal/activities/verifier-material.d.ts +18 -0
- package/dist/temporal/activities/verifier-material.d.ts.map +1 -0
- package/dist/temporal/activities/verifier-material.js +49 -0
- package/dist/temporal/activities/verifier-material.js.map +1 -0
- package/dist/temporal/activities/verifier-outcome.d.ts +19 -0
- package/dist/temporal/activities/verifier-outcome.d.ts.map +1 -0
- package/dist/temporal/activities/verifier-outcome.js +70 -0
- package/dist/temporal/activities/verifier-outcome.js.map +1 -0
- package/dist/temporal/activities/verifier-round.d.ts +8 -0
- package/dist/temporal/activities/verifier-round.d.ts.map +1 -0
- package/dist/temporal/activities/verifier-round.js +110 -0
- package/dist/temporal/activities/verifier-round.js.map +1 -0
- package/dist/temporal/activities/verify-harbor.d.ts +15 -0
- package/dist/temporal/activities/verify-harbor.d.ts.map +1 -0
- package/dist/temporal/activities/verify-harbor.js +193 -0
- package/dist/temporal/activities/verify-harbor.js.map +1 -0
- package/dist/temporal/activities/verify.d.ts +11 -0
- package/dist/temporal/activities/verify.d.ts.map +1 -0
- package/dist/temporal/activities/verify.js +151 -0
- package/dist/temporal/activities/verify.js.map +1 -0
- package/dist/temporal/activities.d.ts +3 -51
- package/dist/temporal/activities.d.ts.map +1 -1
- package/dist/temporal/activities.js +2 -886
- package/dist/temporal/activities.js.map +1 -1
- package/dist/temporal/connection.d.ts.map +1 -1
- package/dist/temporal/connection.js +3 -2
- package/dist/temporal/connection.js.map +1 -1
- package/dist/temporal/worker-main.js +19 -3
- package/dist/temporal/worker-main.js.map +1 -1
- package/dist/temporal/workflow/activity-proxies.d.ts +3 -0
- package/dist/temporal/workflow/activity-proxies.d.ts.map +1 -0
- package/dist/temporal/workflow/activity-proxies.js +45 -0
- package/dist/temporal/workflow/activity-proxies.js.map +1 -0
- package/dist/temporal/workflow/authoring-stage.d.ts +5 -0
- package/dist/temporal/workflow/authoring-stage.d.ts.map +1 -0
- package/dist/temporal/workflow/authoring-stage.js +64 -0
- package/dist/temporal/workflow/authoring-stage.js.map +1 -0
- package/dist/temporal/workflow/candidate-tracker.d.ts +28 -0
- package/dist/temporal/workflow/candidate-tracker.d.ts.map +1 -0
- package/dist/temporal/workflow/candidate-tracker.js +85 -0
- package/dist/temporal/workflow/candidate-tracker.js.map +1 -0
- package/dist/temporal/workflow/candidate.d.ts +6 -0
- package/dist/temporal/workflow/candidate.d.ts.map +1 -0
- package/dist/temporal/workflow/candidate.js +54 -0
- package/dist/temporal/workflow/candidate.js.map +1 -0
- package/dist/temporal/workflow/discovery.d.ts +19 -0
- package/dist/temporal/workflow/discovery.d.ts.map +1 -0
- package/dist/temporal/workflow/discovery.js +137 -0
- package/dist/temporal/workflow/discovery.js.map +1 -0
- package/dist/temporal/workflow/failures.d.ts +5 -0
- package/dist/temporal/workflow/failures.d.ts.map +1 -0
- package/dist/temporal/workflow/failures.js +28 -0
- package/dist/temporal/workflow/failures.js.map +1 -0
- package/dist/temporal/workflow/run.d.ts +5 -0
- package/dist/temporal/workflow/run.d.ts.map +1 -0
- package/dist/temporal/workflow/run.js +154 -0
- package/dist/temporal/workflow/run.js.map +1 -0
- package/dist/temporal/workflow/stage.d.ts +26 -0
- package/dist/temporal/workflow/stage.d.ts.map +1 -0
- package/dist/temporal/workflow/stage.js +16 -0
- package/dist/temporal/workflow/stage.js.map +1 -0
- package/dist/temporal/workflow/verification-stage.d.ts +11 -0
- package/dist/temporal/workflow/verification-stage.d.ts.map +1 -0
- package/dist/temporal/workflow/verification-stage.js +20 -0
- package/dist/temporal/workflow/verification-stage.js.map +1 -0
- package/dist/temporal/workflow.d.ts +12 -4
- package/dist/temporal/workflow.d.ts.map +1 -1
- package/dist/temporal/workflow.js +38 -445
- package/dist/temporal/workflow.js.map +1 -1
- package/dist/verifier-fix.d.ts +20 -0
- package/dist/verifier-fix.d.ts.map +1 -0
- package/dist/verifier-fix.js +62 -0
- package/dist/verifier-fix.js.map +1 -0
- package/dist/verify-report.d.ts +9 -0
- package/dist/verify-report.d.ts.map +1 -0
- package/dist/verify-report.js +135 -0
- package/dist/verify-report.js.map +1 -0
- package/dist/viewer/archived-keys.d.ts +9 -0
- package/dist/viewer/archived-keys.d.ts.map +1 -0
- package/dist/viewer/archived-keys.js +36 -0
- package/dist/viewer/archived-keys.js.map +1 -0
- package/dist/viewer/archived.d.ts +20 -0
- package/dist/viewer/archived.d.ts.map +1 -0
- package/dist/viewer/archived.js +238 -0
- package/dist/viewer/archived.js.map +1 -0
- package/dist/viewer/artifacts.d.ts +8 -0
- package/dist/viewer/artifacts.d.ts.map +1 -0
- package/dist/viewer/artifacts.js +96 -0
- package/dist/viewer/artifacts.js.map +1 -0
- package/dist/viewer/bundle.d.ts +9 -0
- package/dist/viewer/bundle.d.ts.map +1 -0
- package/dist/viewer/bundle.js +79 -0
- package/dist/viewer/bundle.js.map +1 -0
- package/dist/viewer/candidates.d.ts +9 -0
- package/dist/viewer/candidates.d.ts.map +1 -0
- package/dist/viewer/candidates.js +137 -0
- package/dist/viewer/candidates.js.map +1 -0
- package/dist/viewer/local-server.d.ts +11 -0
- package/dist/viewer/local-server.d.ts.map +1 -0
- package/dist/viewer/local-server.js +64 -0
- package/dist/viewer/local-server.js.map +1 -0
- package/dist/viewer/task-files.d.ts +10 -0
- package/dist/viewer/task-files.d.ts.map +1 -0
- package/dist/viewer/task-files.js +120 -0
- package/dist/viewer/task-files.js.map +1 -0
- package/dist/viewer/types.d.ts +78 -0
- package/dist/viewer/types.d.ts.map +1 -0
- package/dist/viewer/types.js +36 -0
- package/dist/viewer/types.js.map +1 -0
- package/docs/api.md +168 -0
- package/docs/evaluations.md +42 -15
- package/docs/operations.md +254 -26
- package/docs/solver-evaluations.md +63 -0
- package/docs/task-construction.md +21 -11
- package/drizzle/0000_glossy_johnny_blaze.sql +97 -0
- package/drizzle/0001_evaluation_records.sql +5 -0
- package/drizzle/0002_task_deletion.sql +1 -0
- package/drizzle/0003_drop_repo_runs.sql +3 -0
- package/drizzle/0004_preserve_batch_runs.sql +8 -0
- package/drizzle/0005_api_keys.sql +15 -0
- package/drizzle/meta/0000_snapshot.json +776 -0
- package/drizzle/meta/0001_snapshot.json +807 -0
- package/drizzle/meta/0002_snapshot.json +813 -0
- package/drizzle/meta/0003_snapshot.json +726 -0
- package/drizzle/meta/0004_snapshot.json +813 -0
- package/drizzle/meta/0005_snapshot.json +919 -0
- package/drizzle/meta/_journal.json +48 -0
- package/package.json +39 -25
- package/src/extensions/authoring.ts +83 -72
- package/src/extensions/shared/deliverable.ts +153 -0
- package/src/extensions/shared/mailbox.ts +128 -0
- package/src/extensions/shared/schemas.ts +101 -0
- package/src/extensions/shared/static-check.ts +103 -0
- package/src/extensions/verifier.ts +88 -0
- package/src/skills/selfbench/SKILL.md +71 -13
- package/biome.json +0 -24
- package/dist/agent-smoke-main.d.ts +0 -3
- package/dist/agent-smoke-main.d.ts.map +0 -1
- package/dist/agent-smoke-main.js +0 -55
- package/dist/agent-smoke-main.js.map +0 -1
- package/dist/agent-smoke.d.ts +0 -18
- package/dist/agent-smoke.d.ts.map +0 -1
- package/dist/agent-smoke.js +0 -103
- package/dist/agent-smoke.js.map +0 -1
- package/dist/codex-review.d.ts +0 -41
- package/dist/codex-review.d.ts.map +0 -1
- package/dist/codex-review.js +0 -193
- package/dist/codex-review.js.map +0 -1
- package/dist/eval-main.d.ts +0 -3
- package/dist/eval-main.d.ts.map +0 -1
- package/dist/eval-main.js +0 -81
- package/dist/eval-main.js.map +0 -1
- package/dist/evaluate.d.ts +0 -25
- package/dist/evaluate.d.ts.map +0 -1
- package/dist/evaluate.js +0 -214
- package/dist/evaluate.js.map +0 -1
- package/dist/modal-auth.d.ts.map +0 -1
- package/dist/modal-auth.js.map +0 -1
- package/dist/reaudit-main.d.ts +0 -3
- package/dist/reaudit-main.d.ts.map +0 -1
- package/dist/reaudit-main.js +0 -158
- package/dist/reaudit-main.js.map +0 -1
- package/dist/repair-main.d.ts +0 -3
- package/dist/repair-main.d.ts.map +0 -1
- package/dist/repair-main.js +0 -171
- package/dist/repair-main.js.map +0 -1
- package/dist/review/assets/index-CZ3G759O.js +0 -1568
- package/dist/review/assets/index-DgSLzX8l.css +0 -1
- package/dist/sandbox/programs/prepare-repair.d.ts +0 -7
- package/dist/sandbox/programs/prepare-repair.d.ts.map +0 -1
- package/dist/sandbox/programs/prepare-repair.js.map +0 -1
- package/dist/sandbox/programs/repair.d.ts +0 -3
- package/dist/sandbox/programs/repair.d.ts.map +0 -1
- package/dist/sandbox/programs/repair.js +0 -100
- package/dist/sandbox/programs/repair.js.map +0 -1
- package/dist/sandbox/programs/review.d.ts +0 -2
- package/dist/sandbox/programs/review.d.ts.map +0 -1
- package/dist/sandbox/programs/review.js +0 -18
- package/dist/sandbox/programs/review.js.map +0 -1
- package/dist/sandbox/programs/validation-repair.d.ts +0 -3
- package/dist/sandbox/programs/validation-repair.d.ts.map +0 -1
- package/dist/sandbox/programs/validation-repair.js +0 -115
- package/dist/sandbox/programs/validation-repair.js.map +0 -1
- package/dist/sandbox-repair.bundle.js +0 -323
- package/dist/sandbox-review.bundle.js +0 -14462
- package/dist/validate-main.d.ts +0 -3
- package/dist/validate-main.d.ts.map +0 -1
- package/dist/validate-main.js +0 -143
- package/dist/validate-main.js.map +0 -1
- package/dist/validation-repair.d.ts +0 -10
- package/dist/validation-repair.d.ts.map +0 -1
- package/dist/validation-repair.js +0 -72
- package/dist/validation-repair.js.map +0 -1
- package/review/index.html +0 -14
- package/review/src/App.tsx +0 -338
- package/review/src/export-loader.test.ts +0 -57
- package/review/src/export-loader.ts +0 -151
- package/review/src/main.tsx +0 -15
- package/review/src/styles.css +0 -162
- package/review/src/types.ts +0 -47
- package/review/tsconfig.json +0 -16
- package/review/vite.config.ts +0 -20
- package/scripts/build-sandbox-programs.ts +0 -22
- package/scripts/verify-package.ts +0 -102
- package/scripts/write-build-metadata.ts +0 -27
- package/src/agent-smoke-main.ts +0 -64
- package/src/agent-smoke.ts +0 -138
- package/src/api-main.ts +0 -12
- package/src/api.ts +0 -291
- package/src/artifacts.ts +0 -357
- package/src/audit.ts +0 -106
- package/src/build-metadata.ts +0 -3
- package/src/cli.ts +0 -504
- package/src/codex-review.ts +0 -229
- package/src/config.ts +0 -239
- package/src/contracts.ts +0 -233
- package/src/coupling.ts +0 -259
- package/src/eval-main.ts +0 -93
- package/src/evaluate.ts +0 -300
- package/src/github.ts +0 -26
- package/src/harbor-environment.ts +0 -17
- package/src/harbor-results.ts +0 -144
- package/src/harbor-task.ts +0 -533
- package/src/hash.ts +0 -5
- package/src/modal-auth.ts +0 -11
- package/src/parallel.ts +0 -24
- package/src/process.ts +0 -167
- package/src/project-paths.ts +0 -15
- package/src/provenance-associations.ts +0 -311
- package/src/provenance.ts +0 -561
- package/src/providers.ts +0 -22
- package/src/reaudit-main.ts +0 -198
- package/src/repair-main.ts +0 -209
- package/src/repair.ts +0 -55
- package/src/run-wait.ts +0 -40
- package/src/sandbox/contracts.ts +0 -52
- package/src/sandbox/index.ts +0 -23
- package/src/sandbox/programs/author.ts +0 -19
- package/src/sandbox/programs/prepare-repair.ts +0 -44
- package/src/sandbox/programs/repair.ts +0 -132
- package/src/sandbox/programs/review.ts +0 -19
- package/src/sandbox/programs/validation-repair.ts +0 -153
- package/src/sandbox/providers/docker/executor.ts +0 -115
- package/src/sandbox/providers/modal/executor.ts +0 -217
- package/src/sandbox/providers/vercel/command.ts +0 -127
- package/src/sandbox/providers/vercel/executor.ts +0 -448
- package/src/sandbox/providers/vercel/fetch.ts +0 -45
- package/src/sandbox/timeout.ts +0 -56
- package/src/setup/vercel/cli.ts +0 -403
- package/src/setup/vercel/index.ts +0 -3
- package/src/setup/vercel/probe.ts +0 -232
- package/src/setup/vercel/profile.ts +0 -322
- package/src/setup/vercel/runtime-image.ts +0 -204
- package/src/setup/vercel/setup.ts +0 -419
- package/src/subscription-auth.ts +0 -80
- package/src/temporal/activities.ts +0 -1348
- package/src/temporal/connection.ts +0 -23
- package/src/temporal/worker-main.ts +0 -28
- package/src/temporal/workflow.ts +0 -568
- package/src/terminal-prompts.ts +0 -197
- package/src/terminal-reporter.ts +0 -126
- package/src/validate-main.ts +0 -177
- package/src/validation-repair.ts +0 -94
- package/tsconfig.build.json +0 -13
- package/tsconfig.json +0 -21
|
@@ -0,0 +1,161 @@
|
|
|
1
|
+
import { randomUUID } from "node:crypto";
|
|
2
|
+
import { z } from "zod";
|
|
3
|
+
import { readAccount, updateAccount } from "./account.js";
|
|
4
|
+
import { catalog, hostedSandboxes } from "./catalog.js";
|
|
5
|
+
import { harnessIds } from "./harnesses.js";
|
|
6
|
+
import { routeFor, thinkingLevels, thinkingOptions } from "./model-options.js";
|
|
7
|
+
import { modelIdPattern } from "./providers.js";
|
|
8
|
+
import { getEvaluation } from "./store.js";
|
|
9
|
+
export const comparisonSchema = z
|
|
10
|
+
.object({
|
|
11
|
+
id: z.uuid(),
|
|
12
|
+
tasks: z
|
|
13
|
+
.array(z
|
|
14
|
+
.object({ runId: z.string().min(1).max(100), taskId: z.string().min(1).max(200) })
|
|
15
|
+
.strict())
|
|
16
|
+
.min(1),
|
|
17
|
+
models: z
|
|
18
|
+
.array(z
|
|
19
|
+
.object({
|
|
20
|
+
catalogId: z.string().max(80),
|
|
21
|
+
customModel: z.string().regex(modelIdPattern).optional(),
|
|
22
|
+
credentialId: z.uuid(),
|
|
23
|
+
thinking: z.enum(thinkingLevels).optional(),
|
|
24
|
+
harnesses: z.array(z.enum(harnessIds)).min(1).max(harnessIds.length),
|
|
25
|
+
})
|
|
26
|
+
.strict())
|
|
27
|
+
.min(1)
|
|
28
|
+
.max(12),
|
|
29
|
+
sandbox: z.enum(hostedSandboxes),
|
|
30
|
+
sandboxCredentialId: z.uuid(),
|
|
31
|
+
})
|
|
32
|
+
.strict();
|
|
33
|
+
export async function createComparison(records, tasks, scope, draft) {
|
|
34
|
+
const selection = comparisonSchema.parse(draft);
|
|
35
|
+
const signature = JSON.stringify(selection);
|
|
36
|
+
const previous = (await readAccount(records, scope.ownerId)).comparisons.find((entry) => entry.id === selection.id);
|
|
37
|
+
if (previous) {
|
|
38
|
+
if (previous.repoId !== scope.repoId || previous.signature !== signature)
|
|
39
|
+
throw new Error("Comparison ID belongs to a different selection");
|
|
40
|
+
return previous;
|
|
41
|
+
}
|
|
42
|
+
if (new Set(selection.tasks.map((task) => `${task.runId}/${task.taskId}`)).size !==
|
|
43
|
+
selection.tasks.length)
|
|
44
|
+
throw new Error("Duplicate task selection");
|
|
45
|
+
const bundles = await Promise.all(selection.tasks.map((task) => tasks.find(scope.repoId, task.runId, task.taskId)));
|
|
46
|
+
if (bundles.some((task) => !task?.bundleKey ||
|
|
47
|
+
task.pipelineStatus !== "accepted" ||
|
|
48
|
+
task.review?.decision !== "approve"))
|
|
49
|
+
throw new Error("Every task must be human-approved in this repository");
|
|
50
|
+
const frozen = bundles.map((task) => {
|
|
51
|
+
if (!task?.bundleKey)
|
|
52
|
+
throw new Error("Task bundle unavailable");
|
|
53
|
+
return { runId: task.runId, taskId: task.taskId, bundleKey: task.bundleKey };
|
|
54
|
+
});
|
|
55
|
+
return updateAccount(records, scope.ownerId, async (account) => {
|
|
56
|
+
const existing = account.comparisons.find((entry) => entry.id === selection.id);
|
|
57
|
+
if (existing) {
|
|
58
|
+
if (existing.repoId !== scope.repoId || existing.signature !== signature)
|
|
59
|
+
throw new Error("Comparison ID belongs to a different selection");
|
|
60
|
+
return existing;
|
|
61
|
+
}
|
|
62
|
+
if (account.comparisons.length >= 500)
|
|
63
|
+
throw new Error("Comparison retention limit reached; contact an operator");
|
|
64
|
+
const sandbox = account.credentials.find((entry) => entry.id === selection.sandboxCredentialId && !entry.deleted);
|
|
65
|
+
if (sandbox?.kind !== selection.sandbox)
|
|
66
|
+
throw new Error("Select your saved sandbox credential");
|
|
67
|
+
const seen = new Set();
|
|
68
|
+
const createdAt = new Date().toISOString();
|
|
69
|
+
const inputs = selection.models.map((selected) => {
|
|
70
|
+
const model = selected.catalogId === "custom" && selected.customModel
|
|
71
|
+
? {
|
|
72
|
+
id: "custom",
|
|
73
|
+
label: "Custom model",
|
|
74
|
+
source: "",
|
|
75
|
+
provider: "custom",
|
|
76
|
+
model: selected.customModel,
|
|
77
|
+
harnesses: ["pi"],
|
|
78
|
+
}
|
|
79
|
+
: catalog.find((entry) => entry.id === selected.catalogId);
|
|
80
|
+
if (!model || (selected.catalogId !== "custom" && selected.customModel))
|
|
81
|
+
throw new Error("Unknown model");
|
|
82
|
+
const credential = account.credentials.find((entry) => entry.id === selected.credentialId && !entry.deleted);
|
|
83
|
+
const route = credential ? routeFor(model, credential.kind) : undefined;
|
|
84
|
+
if (!credential || !route)
|
|
85
|
+
throw new Error("Select your matching provider credential");
|
|
86
|
+
if (credential.auth === "codex-login" &&
|
|
87
|
+
selected.harnesses.some((harness) => harness !== "codex"))
|
|
88
|
+
throw new Error("Codex sign-in can only be used with the Codex harness");
|
|
89
|
+
if (selected.harnesses.some((harness) => !route.harnesses.includes(harness)) ||
|
|
90
|
+
new Set(selected.harnesses).size !== selected.harnesses.length)
|
|
91
|
+
throw new Error("Unsupported or repeated harness");
|
|
92
|
+
const levels = thinkingOptions(model, selected.harnesses);
|
|
93
|
+
const thinking = selected.thinking ?? (levels.includes("high") ? "high" : "default");
|
|
94
|
+
if (!levels.includes(thinking))
|
|
95
|
+
throw new Error("Unsupported thinking level for this model and harness");
|
|
96
|
+
const identity = JSON.stringify([route.provider, route.model, credential.id, thinking]);
|
|
97
|
+
for (const harness of selected.harnesses) {
|
|
98
|
+
const pair = `${identity}/${harness}`;
|
|
99
|
+
if (seen.has(pair))
|
|
100
|
+
throw new Error("Select each model, harness, credential, and thinking configuration once");
|
|
101
|
+
seen.add(pair);
|
|
102
|
+
}
|
|
103
|
+
return {
|
|
104
|
+
id: randomUUID(),
|
|
105
|
+
model: selected.catalogId,
|
|
106
|
+
modelName: `${route.provider === "custom" ? "openai" : route.provider}/${route.model}`,
|
|
107
|
+
thinking,
|
|
108
|
+
harnesses: selected.harnesses,
|
|
109
|
+
sandbox: selection.sandbox,
|
|
110
|
+
tasks: frozen,
|
|
111
|
+
repoId: scope.repoId,
|
|
112
|
+
tenant: scope.tenant,
|
|
113
|
+
startedBy: scope.login,
|
|
114
|
+
createdAt,
|
|
115
|
+
...(route.pricing ? { pricing: route.pricing } : {}),
|
|
116
|
+
credentialOwnerId: scope.ownerId,
|
|
117
|
+
...(scope.credentialOrgId ? { credentialOrgId: scope.credentialOrgId } : {}),
|
|
118
|
+
comparisonId: selection.id,
|
|
119
|
+
credentials: {
|
|
120
|
+
modelCredentialId: credential.id,
|
|
121
|
+
sandboxCredentialId: sandbox.id,
|
|
122
|
+
provider: route.provider,
|
|
123
|
+
},
|
|
124
|
+
};
|
|
125
|
+
});
|
|
126
|
+
const record = {
|
|
127
|
+
id: selection.id,
|
|
128
|
+
repoId: scope.repoId,
|
|
129
|
+
ownerId: scope.ownerId,
|
|
130
|
+
createdAt,
|
|
131
|
+
signature,
|
|
132
|
+
inputs,
|
|
133
|
+
};
|
|
134
|
+
account.comparisons.push(record);
|
|
135
|
+
return record;
|
|
136
|
+
});
|
|
137
|
+
}
|
|
138
|
+
export async function comparisonStatus(store, record) {
|
|
139
|
+
const runs = await Promise.all(record.inputs.map(async (input) => {
|
|
140
|
+
const run = await getEvaluation(store, record.repoId, input.id);
|
|
141
|
+
return {
|
|
142
|
+
id: input.id,
|
|
143
|
+
model: input.modelName,
|
|
144
|
+
thinking: input.thinking,
|
|
145
|
+
harnesses: input.harnesses,
|
|
146
|
+
status: run?.status ?? "pending",
|
|
147
|
+
completed: run?.trials.filter((trial) => trial.status === "completed" || trial.status === "failed")
|
|
148
|
+
.length ?? 0,
|
|
149
|
+
trials: input.tasks.length * input.harnesses.length,
|
|
150
|
+
};
|
|
151
|
+
}));
|
|
152
|
+
return { id: record.id, createdAt: record.createdAt, runs };
|
|
153
|
+
}
|
|
154
|
+
export async function dispatchComparison(store, record, start) {
|
|
155
|
+
for (const input of record.inputs) {
|
|
156
|
+
const run = await getEvaluation(store, record.repoId, input.id);
|
|
157
|
+
if (!run || run.status === "queued")
|
|
158
|
+
await start(input);
|
|
159
|
+
}
|
|
160
|
+
}
|
|
161
|
+
//# sourceMappingURL=comparisons.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"comparisons.js","sourceRoot":"","sources":["../../src/evaluation/comparisons.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAGxB,OAAO,EAAyB,WAAW,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AACjF,OAAO,EAAqB,OAAO,EAAE,eAAe,EAAE,MAAM,cAAc,CAAC;AAE3E,OAAO,EAAE,UAAU,EAAE,MAAM,gBAAgB,CAAC;AAC5C,OAAO,EAAE,QAAQ,EAAE,cAAc,EAAE,eAAe,EAAE,MAAM,oBAAoB,CAAC;AAC/E,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAChD,OAAO,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAG3C,MAAM,CAAC,MAAM,gBAAgB,GAAG,CAAC;KAC9B,MAAM,CAAC;IACN,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE;IACZ,KAAK,EAAE,CAAC;SACL,KAAK,CACJ,CAAC;SACE,MAAM,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;SACjF,MAAM,EAAE,CACZ;SACA,GAAG,CAAC,CAAC,CAAC;IACT,MAAM,EAAE,CAAC;SACN,KAAK,CACJ,CAAC;SACE,MAAM,CAAC;QACN,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC;QAC7B,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,KAAK,CAAC,cAAc,CAAC,CAAC,QAAQ,EAAE;QACxD,YAAY,EAAE,CAAC,CAAC,IAAI,EAAE;QACtB,QAAQ,EAAE,CAAC,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC,QAAQ,EAAE;QAC3C,SAAS,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,UAAU,CAAC,MAAM,CAAC;KACrE,CAAC;SACD,MAAM,EAAE,CACZ;SACA,GAAG,CAAC,CAAC,CAAC;SACN,GAAG,CAAC,EAAE,CAAC;IACV,OAAO,EAAE,CAAC,CAAC,IAAI,CAAC,eAAe,CAAC;IAChC,mBAAmB,EAAE,CAAC,CAAC,IAAI,EAAE;CAC9B,CAAC;KACD,MAAM,EAAE,CAAC;AASZ,MAAM,CAAC,KAAK,UAAU,gBAAgB,CACpC,OAA6B,EAC7B,KAAgB,EAChB,KAAsB,EACtB,KAAsB;IAEtB,MAAM,SAAS,GAAG,gBAAgB,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;IAChD,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,CAAC,SAAS,CAAC,CAAC;IAC5C,MAAM,QAAQ,GAAG,CAAC,MAAM,WAAW,CAAC,OAAO,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,WAAW,CAAC,IAAI,CAC3E,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,SAAS,CAAC,EAAE,CACrC,CAAC;IACF,IAAI,QAAQ,EAAE,CAAC;QACb,IAAI,QAAQ,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,IAAI,QAAQ,CAAC,SAAS,KAAK,SAAS;YACtE,MAAM,IAAI,KAAK,CAAC,gDAAgD,CAAC,CAAC;QACpE,OAAO,QAAQ,CAAC;IAClB,CAAC;IACD,IACE,IAAI,GAAG,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,GAAG,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,IAAI;QAC3E,SAAS,CAAC,KAAK,CAAC,MAAM;QAEtB,MAAM,IAAI,KAAK,CAAC,0BAA0B,CAAC,CAAC;IAC9C,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,GAAG,CAC/B,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,IAAI,CAAC,KAAK,EAAE,IAAI,CAAC,MAAM,CAAC,CAAC,CACjF,CAAC;IACF,IACE,OAAO,CAAC,IAAI,CACV,CAAC,IAAI,EAAE,EAAE,CACP,CAAC,IAAI,EAAE,SAAS;QAChB,IAAI,CAAC,cAAc,KAAK,UAAU;QAClC,IAAI,CAAC,MAAM,EAAE,QAAQ,KAAK,SAAS,CACtC;QAED,MAAM,IAAI,KAAK,CAAC,sDAAsD,CAAC,CAAC;IAC1E,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;QAClC,IAAI,CAAC,IAAI,EAAE,SAAS;YAAE,MAAM,IAAI,KAAK,CAAC,yBAAyB,CAAC,CAAC;QACjE,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,CAAC,MAAM,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,CAAC;IAC/E,CAAC,CAAC,CAAC;IACH,OAAO,aAAa,CAAC,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE;QAC7D,MAAM,QAAQ,GAAG,OAAO,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,SAAS,CAAC,EAAE,CAAC,CAAC;QAChF,IAAI,QAAQ,EAAE,CAAC;YACb,IAAI,QAAQ,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,IAAI,QAAQ,CAAC,SAAS,KAAK,SAAS;gBACtE,MAAM,IAAI,KAAK,CAAC,gDAAgD,CAAC,CAAC;YACpE,OAAO,QAAQ,CAAC;QAClB,CAAC;QACD,IAAI,OAAO,CAAC,WAAW,CAAC,MAAM,IAAI,GAAG;YACnC,MAAM,IAAI,KAAK,CAAC,yDAAyD,CAAC,CAAC;QAC7E,MAAM,OAAO,GAAG,OAAO,CAAC,WAAW,CAAC,IAAI,CACtC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,SAAS,CAAC,mBAAmB,IAAI,CAAC,KAAK,CAAC,OAAO,CACxE,CAAC;QACF,IAAI,OAAO,EAAE,IAAI,KAAK,SAAS,CAAC,OAAO;YACrC,MAAM,IAAI,KAAK,CAAC,sCAAsC,CAAC,CAAC;QAC1D,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;QAC/B,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;QAC3C,MAAM,MAAM,GAAsB,SAAS,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE;YAClE,MAAM,KAAK,GACT,QAAQ,CAAC,SAAS,KAAK,QAAQ,IAAI,QAAQ,CAAC,WAAW;gBACrD,CAAC,CAAC;oBACE,EAAE,EAAE,QAAQ;oBACZ,KAAK,EAAE,cAAc;oBACrB,MAAM,EAAE,EAAE;oBACV,QAAQ,EAAE,QAAQ;oBAClB,KAAK,EAAE,QAAQ,CAAC,WAAW;oBAC3B,SAAS,EAAE,CAAC,IAAI,CAAC;iBAClB;gBACH,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,QAAQ,CAAC,SAAS,CAAC,CAAC;YAC/D,IAAI,CAAC,KAAK,IAAI,CAAC,QAAQ,CAAC,SAAS,KAAK,QAAQ,IAAI,QAAQ,CAAC,WAAW,CAAC;gBACrE,MAAM,IAAI,KAAK,CAAC,eAAe,CAAC,CAAC;YACnC,MAAM,UAAU,GAAG,OAAO,CAAC,WAAW,CAAC,IAAI,CACzC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,QAAQ,CAAC,YAAY,IAAI,CAAC,KAAK,CAAC,OAAO,CAChE,CAAC;YACF,MAAM,KAAK,GAAG,UAAU,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,EAAE,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;YACxE,IAAI,CAAC,UAAU,IAAI,CAAC,KAAK;gBAAE,MAAM,IAAI,KAAK,CAAC,0CAA0C,CAAC,CAAC;YACvF,IACE,UAAU,CAAC,IAAI,KAAK,aAAa;gBACjC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,KAAK,OAAO,CAAC;gBAEzD,MAAM,IAAI,KAAK,CAAC,uDAAuD,CAAC,CAAC;YAC3E,IACE,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,SAAS,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;gBACxE,IAAI,GAAG,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC,IAAI,KAAK,QAAQ,CAAC,SAAS,CAAC,MAAM;gBAE9D,MAAM,IAAI,KAAK,CAAC,iCAAiC,CAAC,CAAC;YACrD,MAAM,MAAM,GAAG,eAAe,CAAC,KAAK,EAAE,QAAQ,CAAC,SAAS,CAAC,CAAC;YAC1D,MAAM,QAAQ,GAAG,QAAQ,CAAC,QAAQ,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC;YACrF,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC;gBAC5B,MAAM,IAAI,KAAK,CAAC,uDAAuD,CAAC,CAAC;YAC3E,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,CAAC,KAAK,CAAC,QAAQ,EAAE,KAAK,CAAC,KAAK,EAAE,UAAU,CAAC,EAAE,EAAE,QAAQ,CAAC,CAAC,CAAC;YACxF,KAAK,MAAM,OAAO,IAAI,QAAQ,CAAC,SAAS,EAAE,CAAC;gBACzC,MAAM,IAAI,GAAG,GAAG,QAAQ,IAAI,OAAO,EAAE,CAAC;gBACtC,IAAI,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC;oBAChB,MAAM,IAAI,KAAK,CACb,yEAAyE,CAC1E,CAAC;gBACJ,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;YACjB,CAAC;YACD,OAAO;gBACL,EAAE,EAAE,UAAU,EAAE;gBAChB,KAAK,EAAE,QAAQ,CAAC,SAAS;gBACzB,SAAS,EAAE,GAAG,KAAK,CAAC,QAAQ,KAAK,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,QAAQ,IAAI,KAAK,CAAC,KAAK,EAAE;gBACtF,QAAQ;gBACR,SAAS,EAAE,QAAQ,CAAC,SAAS;gBAC7B,OAAO,EAAE,SAAS,CAAC,OAAO;gBAC1B,KAAK,EAAE,MAAM;gBACb,MAAM,EAAE,KAAK,CAAC,MAAM;gBACpB,MAAM,EAAE,KAAK,CAAC,MAAM;gBACpB,SAAS,EAAE,KAAK,CAAC,KAAK;gBACtB,SAAS;gBACT,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACpD,iBAAiB,EAAE,KAAK,CAAC,OAAO;gBAChC,GAAG,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,KAAK,CAAC,eAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC5E,YAAY,EAAE,SAAS,CAAC,EAAE;gBAC1B,WAAW,EAAE;oBACX,iBAAiB,EAAE,UAAU,CAAC,EAAE;oBAChC,mBAAmB,EAAE,OAAO,CAAC,EAAE;oBAC/B,QAAQ,EAAE,KAAK,CAAC,QAAQ;iBACzB;aACF,CAAC;QACJ,CAAC,CAAC,CAAC;QACH,MAAM,MAAM,GAAqB;YAC/B,EAAE,EAAE,SAAS,CAAC,EAAE;YAChB,MAAM,EAAE,KAAK,CAAC,MAAM;YACpB,OAAO,EAAE,KAAK,CAAC,OAAO;YACtB,SAAS;YACT,SAAS;YACT,MAAM;SACP,CAAC;QACF,OAAO,CAAC,WAAW,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACjC,OAAO,MAAM,CAAC;IAChB,CAAC,CAAC,CAAC;AACL,CAAC;AACD,MAAM,CAAC,KAAK,UAAU,gBAAgB,CAAC,KAAoB,EAAE,MAAwB;IACnF,MAAM,IAAI,GAAG,MAAM,OAAO,CAAC,GAAG,CAC5B,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE;QAChC,MAAM,GAAG,GAAG,MAAM,aAAa,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,KAAK,CAAC,EAAE,CAAC,CAAC;QAChE,OAAO;YACL,EAAE,EAAE,KAAK,CAAC,EAAE;YACZ,KAAK,EAAE,KAAK,CAAC,SAAS;YACtB,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,MAAM,EAAE,GAAG,EAAE,MAAM,IAAI,SAAS;YAChC,SAAS,EACP,GAAG,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,MAAM,KAAK,WAAW,IAAI,KAAK,CAAC,MAAM,KAAK,QAAQ,CAAC;iBACrF,MAAM,IAAI,CAAC;YAChB,MAAM,EAAE,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,KAAK,CAAC,SAAS,CAAC,MAAM;SACpD,CAAC;IACJ,CAAC,CAAC,CACH,CAAC;IACF,OAAO,EAAE,EAAE,EAAE,MAAM,CAAC,EAAE,EAAE,SAAS,EAAE,MAAM,CAAC,SAAS,EAAE,IAAI,EAAE,CAAC;AAC9D,CAAC;AACD,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACtC,KAAoB,EACpB,MAAwB,EACxB,KAAgD;IAEhD,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC;QAClC,MAAM,GAAG,GAAG,MAAM,aAAa,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,KAAK,CAAC,EAAE,CAAC,CAAC;QAChE,IAAI,CAAC,GAAG,IAAI,GAAG,CAAC,MAAM,KAAK,QAAQ;YAAE,MAAM,KAAK,CAAC,KAAK,CAAC,CAAC;IAC1D,CAAC;AACH,CAAC"}
|
|
@@ -0,0 +1,43 @@
|
|
|
1
|
+
import type { EvaluationChoices, EvaluationInput } from "./types.js";
|
|
2
|
+
export declare const HARBOR_VERSION = "0.23.0";
|
|
3
|
+
export declare const evaluationSandboxes: readonly ["docker", "modal", "e2b", "daytona"];
|
|
4
|
+
export declare function evaluationConfig(env?: NodeJS.ProcessEnv): {
|
|
5
|
+
profiles: {
|
|
6
|
+
id: string;
|
|
7
|
+
label: string;
|
|
8
|
+
model: string;
|
|
9
|
+
harnesses: ("codex" | "claude-code" | "pi" | "mini-swe-agent" | "terminus-2")[];
|
|
10
|
+
tenants: string[];
|
|
11
|
+
credentialEnv: string;
|
|
12
|
+
pricing?: {
|
|
13
|
+
input: number;
|
|
14
|
+
output: number;
|
|
15
|
+
cacheRead: number;
|
|
16
|
+
cacheWrite: number;
|
|
17
|
+
source: string;
|
|
18
|
+
asOf: string;
|
|
19
|
+
} | undefined;
|
|
20
|
+
}[];
|
|
21
|
+
sandboxes: ("docker" | "modal" | "e2b" | "daytona")[];
|
|
22
|
+
};
|
|
23
|
+
export declare function evaluationChoices(tenant: string, env?: NodeJS.ProcessEnv): EvaluationChoices;
|
|
24
|
+
export declare function solverEnvironment(input: EvaluationInput, home: string, env?: NodeJS.ProcessEnv): {
|
|
25
|
+
profile: {
|
|
26
|
+
id: string;
|
|
27
|
+
label: string;
|
|
28
|
+
model: string;
|
|
29
|
+
harnesses: ("codex" | "claude-code" | "pi" | "mini-swe-agent" | "terminus-2")[];
|
|
30
|
+
tenants: string[];
|
|
31
|
+
credentialEnv: string;
|
|
32
|
+
pricing?: {
|
|
33
|
+
input: number;
|
|
34
|
+
output: number;
|
|
35
|
+
cacheRead: number;
|
|
36
|
+
cacheWrite: number;
|
|
37
|
+
source: string;
|
|
38
|
+
asOf: string;
|
|
39
|
+
} | undefined;
|
|
40
|
+
};
|
|
41
|
+
child: NodeJS.ProcessEnv;
|
|
42
|
+
};
|
|
43
|
+
//# sourceMappingURL=config.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"config.d.ts","sourceRoot":"","sources":["../../src/evaluation/config.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,iBAAiB,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AAErE,eAAO,MAAM,cAAc,WAAW,CAAC;AACvC,eAAO,MAAM,mBAAmB,gDAAiD,CAAC;AAqClF,wBAAgB,gBAAgB,CAAC,GAAG,GAAE,MAAM,CAAC,UAAwB;;;;;;;;;;;;;;;;;;EAapE;AACD,wBAAgB,iBAAiB,CAC/B,MAAM,EAAE,MAAM,EACd,GAAG,GAAE,MAAM,CAAC,UAAwB,GACnC,iBAAiB,CAiBnB;AACD,wBAAgB,iBAAiB,CAC/B,KAAK,EAAE,eAAe,EACtB,IAAI,EAAE,MAAM,EACZ,GAAG,GAAE,MAAM,CAAC,UAAwB;;;;;;;;;;;;;;;;;;EAiDrC"}
|
|
@@ -0,0 +1,116 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
import { harnessIds } from "./harnesses.js";
|
|
3
|
+
import { modelIdPattern, modelProvider, providers } from "./providers.js";
|
|
4
|
+
export const HARBOR_VERSION = "0.23.0";
|
|
5
|
+
export const evaluationSandboxes = ["docker", "modal", "e2b", "daytona"];
|
|
6
|
+
const profileSchema = z
|
|
7
|
+
.object({
|
|
8
|
+
id: z.string().regex(/^[a-z0-9-]{1,60}$/),
|
|
9
|
+
label: z.string().min(1).max(100),
|
|
10
|
+
model: z
|
|
11
|
+
.string()
|
|
12
|
+
.refine((value) => !!modelProvider(value) &&
|
|
13
|
+
modelIdPattern.test(value.slice(value.indexOf("/") + 1)) &&
|
|
14
|
+
value.includes("/")),
|
|
15
|
+
harnesses: z.array(z.enum(harnessIds)).min(1).max(harnessIds.length),
|
|
16
|
+
tenants: z.array(z.string().regex(/^[A-Za-z0-9_.-]+$/)).min(1),
|
|
17
|
+
credentialEnv: z.string().regex(/^SELFBENCH_EVAL_SECRET_[A-Z0-9_]+$/),
|
|
18
|
+
pricing: z
|
|
19
|
+
.object({
|
|
20
|
+
input: z.number().nonnegative(),
|
|
21
|
+
output: z.number().nonnegative(),
|
|
22
|
+
cacheRead: z.number().nonnegative(),
|
|
23
|
+
cacheWrite: z.number().nonnegative(),
|
|
24
|
+
source: z.url(),
|
|
25
|
+
asOf: z.iso.date(),
|
|
26
|
+
})
|
|
27
|
+
.strict()
|
|
28
|
+
.optional(),
|
|
29
|
+
})
|
|
30
|
+
.strict()
|
|
31
|
+
.superRefine((profile, context) => {
|
|
32
|
+
if (profile.harnesses.includes("codex") && !profile.model.startsWith("openai/")) {
|
|
33
|
+
context.addIssue({ code: "custom", message: "Codex requires an OpenAI model" });
|
|
34
|
+
}
|
|
35
|
+
if (profile.harnesses.includes("claude-code") && !profile.model.startsWith("anthropic/")) {
|
|
36
|
+
context.addIssue({ code: "custom", message: "Claude Code requires an Anthropic model" });
|
|
37
|
+
}
|
|
38
|
+
});
|
|
39
|
+
export function evaluationConfig(env = process.env) {
|
|
40
|
+
const profiles = z
|
|
41
|
+
.array(profileSchema)
|
|
42
|
+
.max(50)
|
|
43
|
+
.parse(JSON.parse(env.SELFBENCH_EVAL_MODELS ?? "[]"));
|
|
44
|
+
if (new Set(profiles.map((profile) => profile.id)).size !== profiles.length) {
|
|
45
|
+
throw new Error("Duplicate evaluation model IDs");
|
|
46
|
+
}
|
|
47
|
+
const sandboxes = z
|
|
48
|
+
.array(z.enum(evaluationSandboxes))
|
|
49
|
+
.max(4)
|
|
50
|
+
.parse(JSON.parse(env.SELFBENCH_EVAL_SANDBOXES ?? "[]"));
|
|
51
|
+
return { profiles, sandboxes };
|
|
52
|
+
}
|
|
53
|
+
export function evaluationChoices(tenant, env = process.env) {
|
|
54
|
+
const config = evaluationConfig(env);
|
|
55
|
+
return {
|
|
56
|
+
harborVersion: HARBOR_VERSION,
|
|
57
|
+
sandboxes: [...new Set(config.sandboxes)],
|
|
58
|
+
models: config.profiles
|
|
59
|
+
.filter((profile) => profile.tenants.some((allowed) => allowed.toLowerCase() === tenant.toLowerCase()))
|
|
60
|
+
.map(({ id, label, model, harnesses, pricing }) => ({
|
|
61
|
+
id,
|
|
62
|
+
label,
|
|
63
|
+
model,
|
|
64
|
+
harnesses: [...new Set(harnesses)],
|
|
65
|
+
...(pricing ? { pricing } : {}),
|
|
66
|
+
})),
|
|
67
|
+
};
|
|
68
|
+
}
|
|
69
|
+
export function solverEnvironment(input, home, env = process.env) {
|
|
70
|
+
const config = evaluationConfig(env);
|
|
71
|
+
const profile = config.profiles.find((candidate) => candidate.id === input.model &&
|
|
72
|
+
candidate.tenants.some((tenant) => tenant.toLowerCase() === input.tenant.toLowerCase()));
|
|
73
|
+
if (!profile ||
|
|
74
|
+
profile.model !== input.modelName ||
|
|
75
|
+
!config.sandboxes.includes(input.sandbox) ||
|
|
76
|
+
input.harnesses.some((harness) => !profile.harnesses.includes(harness))) {
|
|
77
|
+
throw new Error("Evaluation configuration is no longer available");
|
|
78
|
+
}
|
|
79
|
+
const secret = env[profile.credentialEnv];
|
|
80
|
+
if (!secret)
|
|
81
|
+
throw new Error("Evaluation model credential is not configured on the worker");
|
|
82
|
+
const child = {
|
|
83
|
+
PATH: env.PATH,
|
|
84
|
+
HOME: home,
|
|
85
|
+
TMPDIR: home,
|
|
86
|
+
LANG: "C.UTF-8",
|
|
87
|
+
PYTHONUNBUFFERED: "1",
|
|
88
|
+
};
|
|
89
|
+
const provider = modelProvider(profile.model);
|
|
90
|
+
if (!provider)
|
|
91
|
+
throw new Error("Unsupported model provider");
|
|
92
|
+
child[providers[provider].credential] = secret;
|
|
93
|
+
if (input.sandbox === "docker" && env.DOCKER_HOST)
|
|
94
|
+
child.DOCKER_HOST = env.DOCKER_HOST;
|
|
95
|
+
if (input.sandbox === "modal") {
|
|
96
|
+
if (!env.SELFBENCH_EVAL_SECRET_MODAL_TOKEN_ID ||
|
|
97
|
+
!env.SELFBENCH_EVAL_SECRET_MODAL_TOKEN_SECRET) {
|
|
98
|
+
throw new Error("Evaluation Modal credentials are not configured on the worker");
|
|
99
|
+
}
|
|
100
|
+
child.MODAL_TOKEN_ID = env.SELFBENCH_EVAL_SECRET_MODAL_TOKEN_ID;
|
|
101
|
+
child.MODAL_TOKEN_SECRET = env.SELFBENCH_EVAL_SECRET_MODAL_TOKEN_SECRET;
|
|
102
|
+
}
|
|
103
|
+
for (const [sandbox, name] of [
|
|
104
|
+
["e2b", "E2B_API_KEY"],
|
|
105
|
+
["daytona", "DAYTONA_API_KEY"],
|
|
106
|
+
]) {
|
|
107
|
+
if (input.sandbox !== sandbox)
|
|
108
|
+
continue;
|
|
109
|
+
const value = env[`SELFBENCH_EVAL_SECRET_${name}`];
|
|
110
|
+
if (!value)
|
|
111
|
+
throw new Error(`Evaluation ${sandbox} API key is missing`);
|
|
112
|
+
child[name] = value;
|
|
113
|
+
}
|
|
114
|
+
return { profile, child };
|
|
115
|
+
}
|
|
116
|
+
//# sourceMappingURL=config.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"config.js","sourceRoot":"","sources":["../../src/evaluation/config.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,UAAU,EAAE,MAAM,gBAAgB,CAAC;AAC5C,OAAO,EAAE,cAAc,EAAE,aAAa,EAAE,SAAS,EAAE,MAAM,gBAAgB,CAAC;AAG1E,MAAM,CAAC,MAAM,cAAc,GAAG,QAAQ,CAAC;AACvC,MAAM,CAAC,MAAM,mBAAmB,GAAG,CAAC,QAAQ,EAAE,OAAO,EAAE,KAAK,EAAE,SAAS,CAAU,CAAC;AAClF,MAAM,aAAa,GAAG,CAAC;KACpB,MAAM,CAAC;IACN,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,KAAK,CAAC,mBAAmB,CAAC;IACzC,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC;IACjC,KAAK,EAAE,CAAC;SACL,MAAM,EAAE;SACR,MAAM,CACL,CAAC,KAAK,EAAE,EAAE,CACR,CAAC,CAAC,aAAa,CAAC,KAAK,CAAC;QACtB,cAAc,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC;QACxD,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CACtB;IACH,SAAS,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,UAAU,CAAC,MAAM,CAAC;IACpE,OAAO,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,KAAK,CAAC,mBAAmB,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC9D,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,KAAK,CAAC,oCAAoC,CAAC;IACrE,OAAO,EAAE,CAAC;SACP,MAAM,CAAC;QACN,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,WAAW,EAAE;QAC/B,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,WAAW,EAAE;QAChC,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,WAAW,EAAE;QACnC,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,WAAW,EAAE;QACpC,MAAM,EAAE,CAAC,CAAC,GAAG,EAAE;QACf,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,IAAI,EAAE;KACnB,CAAC;SACD,MAAM,EAAE;SACR,QAAQ,EAAE;CACd,CAAC;KACD,MAAM,EAAE;KACR,WAAW,CAAC,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE;IAChC,IAAI,OAAO,CAAC,SAAS,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,UAAU,CAAC,SAAS,CAAC,EAAE,CAAC;QAChF,OAAO,CAAC,QAAQ,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,gCAAgC,EAAE,CAAC,CAAC;IAClF,CAAC;IACD,IAAI,OAAO,CAAC,SAAS,CAAC,QAAQ,CAAC,aAAa,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,UAAU,CAAC,YAAY,CAAC,EAAE,CAAC;QACzF,OAAO,CAAC,QAAQ,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,yCAAyC,EAAE,CAAC,CAAC;IAC3F,CAAC;AACH,CAAC,CAAC,CAAC;AACL,MAAM,UAAU,gBAAgB,CAAC,MAAyB,OAAO,CAAC,GAAG;IACnE,MAAM,QAAQ,GAAG,CAAC;SACf,KAAK,CAAC,aAAa,CAAC;SACpB,GAAG,CAAC,EAAE,CAAC;SACP,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,qBAAqB,IAAI,IAAI,CAAC,CAAC,CAAC;IACxD,IAAI,IAAI,GAAG,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,QAAQ,CAAC,MAAM,EAAE,CAAC;QAC5E,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IACD,MAAM,SAAS,GAAG,CAAC;SAChB,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,mBAAmB,CAAC,CAAC;SAClC,GAAG,CAAC,CAAC,CAAC;SACN,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,wBAAwB,IAAI,IAAI,CAAC,CAAC,CAAC;IAC3D,OAAO,EAAE,QAAQ,EAAE,SAAS,EAAE,CAAC;AACjC,CAAC;AACD,MAAM,UAAU,iBAAiB,CAC/B,MAAc,EACd,MAAyB,OAAO,CAAC,GAAG;IAEpC,MAAM,MAAM,GAAG,gBAAgB,CAAC,GAAG,CAAC,CAAC;IACrC,OAAO;QACL,aAAa,EAAE,cAAc;QAC7B,SAAS,EAAE,CAAC,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;QACzC,MAAM,EAAE,MAAM,CAAC,QAAQ;aACpB,MAAM,CAAC,CAAC,OAAO,EAAE,EAAE,CAClB,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,WAAW,EAAE,KAAK,MAAM,CAAC,WAAW,EAAE,CAAC,CAClF;aACA,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,SAAS,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC,CAAC;YAClD,EAAE;YACF,KAAK;YACL,KAAK;YACL,SAAS,EAAE,CAAC,GAAG,IAAI,GAAG,CAAC,SAAS,CAAC,CAAC;YAClC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAChC,CAAC,CAAC;KACN,CAAC;AACJ,CAAC;AACD,MAAM,UAAU,iBAAiB,CAC/B,KAAsB,EACtB,IAAY,EACZ,MAAyB,OAAO,CAAC,GAAG;IAEpC,MAAM,MAAM,GAAG,gBAAgB,CAAC,GAAG,CAAC,CAAC;IACrC,MAAM,OAAO,GAAG,MAAM,CAAC,QAAQ,CAAC,IAAI,CAClC,CAAC,SAAS,EAAE,EAAE,CACZ,SAAS,CAAC,EAAE,KAAK,KAAK,CAAC,KAAK;QAC5B,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,WAAW,EAAE,KAAK,KAAK,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC,CAC1F,CAAC;IACF,IACE,CAAC,OAAO;QACR,OAAO,CAAC,KAAK,KAAK,KAAK,CAAC,SAAS;QACjC,CAAC,MAAM,CAAC,SAAS,CAAC,QAAQ,CAAC,KAAK,CAAC,OAAO,CAAC;QACzC,KAAK,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,SAAS,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,EACvE,CAAC;QACD,MAAM,IAAI,KAAK,CAAC,iDAAiD,CAAC,CAAC;IACrE,CAAC;IACD,MAAM,MAAM,GAAG,GAAG,CAAC,OAAO,CAAC,aAAa,CAAC,CAAC;IAC1C,IAAI,CAAC,MAAM;QAAE,MAAM,IAAI,KAAK,CAAC,6DAA6D,CAAC,CAAC;IAC5F,MAAM,KAAK,GAAsB;QAC/B,IAAI,EAAE,GAAG,CAAC,IAAI;QACd,IAAI,EAAE,IAAI;QACV,MAAM,EAAE,IAAI;QACZ,IAAI,EAAE,SAAS;QACf,gBAAgB,EAAE,GAAG;KACtB,CAAC;IACF,MAAM,QAAQ,GAAG,aAAa,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;IAC9C,IAAI,CAAC,QAAQ;QAAE,MAAM,IAAI,KAAK,CAAC,4BAA4B,CAAC,CAAC;IAC7D,KAAK,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,UAAU,CAAC,GAAG,MAAM,CAAC;IAC/C,IAAI,KAAK,CAAC,OAAO,KAAK,QAAQ,IAAI,GAAG,CAAC,WAAW;QAAE,KAAK,CAAC,WAAW,GAAG,GAAG,CAAC,WAAW,CAAC;IACvF,IAAI,KAAK,CAAC,OAAO,KAAK,OAAO,EAAE,CAAC;QAC9B,IACE,CAAC,GAAG,CAAC,oCAAoC;YACzC,CAAC,GAAG,CAAC,wCAAwC,EAC7C,CAAC;YACD,MAAM,IAAI,KAAK,CAAC,+DAA+D,CAAC,CAAC;QACnF,CAAC;QACD,KAAK,CAAC,cAAc,GAAG,GAAG,CAAC,oCAAoC,CAAC;QAChE,KAAK,CAAC,kBAAkB,GAAG,GAAG,CAAC,wCAAwC,CAAC;IAC1E,CAAC;IACD,KAAK,MAAM,CAAC,OAAO,EAAE,IAAI,CAAC,IAAI;QAC5B,CAAC,KAAK,EAAE,aAAa,CAAC;QACtB,CAAC,SAAS,EAAE,iBAAiB,CAAC;KACtB,EAAE,CAAC;QACX,IAAI,KAAK,CAAC,OAAO,KAAK,OAAO;YAAE,SAAS;QACxC,MAAM,KAAK,GAAG,GAAG,CAAC,yBAAyB,IAAI,EAAE,CAAC,CAAC;QACnD,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,cAAc,OAAO,qBAAqB,CAAC,CAAC;QACxE,KAAK,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC;IACtB,CAAC;IACD,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;AAC5B,CAAC"}
|
|
@@ -0,0 +1,3 @@
|
|
|
1
|
+
import type { EvaluationRun, EvaluationTrial } from "./types.js";
|
|
2
|
+
export declare function trialCost(run: EvaluationRun, harness: string, files: Map<string, string>, result: unknown): Pick<EvaluationTrial, "modelVerified" | "apiCostUsd" | "tokenUsage" | "costSource">;
|
|
3
|
+
//# sourceMappingURL=cost.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"cost.d.ts","sourceRoot":"","sources":["../../src/evaluation/cost.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,aAAa,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AAIjE,wBAAgB,SAAS,CACvB,GAAG,EAAE,aAAa,EAClB,OAAO,EAAE,MAAM,EACf,KAAK,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,EAC1B,MAAM,EAAE,OAAO,GACd,IAAI,CAAC,eAAe,EAAE,eAAe,GAAG,YAAY,GAAG,YAAY,GAAG,YAAY,CAAC,CAuFrF"}
|
|
@@ -0,0 +1,95 @@
|
|
|
1
|
+
import { harborCost } from "./harbor-cost.js";
|
|
2
|
+
import { record } from "./output.js";
|
|
3
|
+
const count = (value) => typeof value === "number" && Number.isSafeInteger(value) && value >= 0;
|
|
4
|
+
export function trialCost(run, harness, files, result) {
|
|
5
|
+
const pricing = run.pricing;
|
|
6
|
+
const model = run.modelName.split("/").slice(1).join("/");
|
|
7
|
+
const matches = (value) => value === model || value === run.modelName;
|
|
8
|
+
let usage;
|
|
9
|
+
let verified = false;
|
|
10
|
+
let reportedCost;
|
|
11
|
+
if (harness === "pi") {
|
|
12
|
+
const text = [...files].find(([name]) => name.endsWith("/pi.txt"))?.[1];
|
|
13
|
+
if (!text)
|
|
14
|
+
return {};
|
|
15
|
+
const messages = [];
|
|
16
|
+
for (const line of text.split("\n").filter(Boolean)) {
|
|
17
|
+
let event;
|
|
18
|
+
try {
|
|
19
|
+
event = record(JSON.parse(line));
|
|
20
|
+
}
|
|
21
|
+
catch {
|
|
22
|
+
return {};
|
|
23
|
+
}
|
|
24
|
+
const message = record(event.message);
|
|
25
|
+
if (event.type === "message_end" && message.role === "assistant")
|
|
26
|
+
messages.push(message);
|
|
27
|
+
}
|
|
28
|
+
verified =
|
|
29
|
+
messages.length > 0 &&
|
|
30
|
+
messages.every((message) => `${message.provider}/${message.model}` === run.modelName);
|
|
31
|
+
const totals = { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 };
|
|
32
|
+
for (const message of messages) {
|
|
33
|
+
const tokens = record(message.usage);
|
|
34
|
+
for (const key of ["input", "output", "cacheRead", "cacheWrite"]) {
|
|
35
|
+
const value = tokens[key];
|
|
36
|
+
if (!count(value))
|
|
37
|
+
return { modelVerified: verified };
|
|
38
|
+
totals[key] += value;
|
|
39
|
+
}
|
|
40
|
+
}
|
|
41
|
+
usage = totals;
|
|
42
|
+
}
|
|
43
|
+
else if (harness === "codex") {
|
|
44
|
+
const text = [...files].find(([name]) => name.endsWith("/trajectory.json"))?.[1];
|
|
45
|
+
if (!text)
|
|
46
|
+
return {};
|
|
47
|
+
let trajectory;
|
|
48
|
+
try {
|
|
49
|
+
trajectory = record(JSON.parse(text));
|
|
50
|
+
}
|
|
51
|
+
catch {
|
|
52
|
+
return {};
|
|
53
|
+
}
|
|
54
|
+
const steps = Array.isArray(trajectory.steps)
|
|
55
|
+
? trajectory.steps.map(record).filter((step) => step.source === "agent")
|
|
56
|
+
: [];
|
|
57
|
+
verified =
|
|
58
|
+
steps.length > 0 &&
|
|
59
|
+
steps.every((step) => matches(step.model_name ?? record(trajectory.agent).model_name));
|
|
60
|
+
const tokens = record(record(result).agent_result);
|
|
61
|
+
const cacheWrite = record(record(trajectory.final_metrics).extra).total_cache_write_input_tokens ?? 0;
|
|
62
|
+
if (count(tokens.n_input_tokens) &&
|
|
63
|
+
count(tokens.n_output_tokens) &&
|
|
64
|
+
count(tokens.n_cache_tokens) &&
|
|
65
|
+
count(cacheWrite) &&
|
|
66
|
+
tokens.n_cache_tokens + cacheWrite <= tokens.n_input_tokens) {
|
|
67
|
+
usage = {
|
|
68
|
+
input: tokens.n_input_tokens - tokens.n_cache_tokens - cacheWrite,
|
|
69
|
+
output: tokens.n_output_tokens,
|
|
70
|
+
cacheRead: tokens.n_cache_tokens,
|
|
71
|
+
cacheWrite,
|
|
72
|
+
};
|
|
73
|
+
reportedCost = harborCost(trajectory, usage, tokens.cost_usd);
|
|
74
|
+
}
|
|
75
|
+
}
|
|
76
|
+
const measured = { modelVerified: verified, ...(usage ? { tokenUsage: usage } : {}) };
|
|
77
|
+
if (!verified || !usage)
|
|
78
|
+
return measured;
|
|
79
|
+
if (reportedCost !== undefined)
|
|
80
|
+
return { ...measured, apiCostUsd: reportedCost, costSource: "harbor" };
|
|
81
|
+
if (!pricing)
|
|
82
|
+
return measured;
|
|
83
|
+
if (pricing.maxInputTokens &&
|
|
84
|
+
usage.input + usage.cacheRead + usage.cacheWrite > pricing.maxInputTokens)
|
|
85
|
+
return measured;
|
|
86
|
+
const apiCostUsd = (usage.input * pricing.input +
|
|
87
|
+
usage.output * pricing.output +
|
|
88
|
+
usage.cacheRead * pricing.cacheRead +
|
|
89
|
+
usage.cacheWrite * pricing.cacheWrite) /
|
|
90
|
+
1_000_000;
|
|
91
|
+
return Number.isFinite(apiCostUsd)
|
|
92
|
+
? { ...measured, apiCostUsd, costSource: "reference-rates" }
|
|
93
|
+
: measured;
|
|
94
|
+
}
|
|
95
|
+
//# sourceMappingURL=cost.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"cost.js","sourceRoot":"","sources":["../../src/evaluation/cost.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAC9C,OAAO,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC;AAGrC,MAAM,KAAK,GAAG,CAAC,KAAc,EAAmB,EAAE,CAChD,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC;AACzE,MAAM,UAAU,SAAS,CACvB,GAAkB,EAClB,OAAe,EACf,KAA0B,EAC1B,MAAe;IAEf,MAAM,OAAO,GAAG,GAAG,CAAC,OAAO,CAAC;IAC5B,MAAM,KAAK,GAAG,GAAG,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAC1D,MAAM,OAAO,GAAG,CAAC,KAAc,EAAE,EAAE,CAAC,KAAK,KAAK,KAAK,IAAI,KAAK,KAAK,GAAG,CAAC,SAAS,CAAC;IAC/E,IAAI,KAA2F,CAAC;IAChG,IAAI,QAAQ,GAAG,KAAK,CAAC;IACrB,IAAI,YAAgC,CAAC;IACrC,IAAI,OAAO,KAAK,IAAI,EAAE,CAAC;QACrB,MAAM,IAAI,GAAG,CAAC,GAAG,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;QACxE,IAAI,CAAC,IAAI;YAAE,OAAO,EAAE,CAAC;QACrB,MAAM,QAAQ,GAA8B,EAAE,CAAC;QAC/C,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC;YACpD,IAAI,KAA8B,CAAC;YACnC,IAAI,CAAC;gBACH,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;YACnC,CAAC;YAAC,MAAM,CAAC;gBACP,OAAO,EAAE,CAAC;YACZ,CAAC;YACD,MAAM,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;YACtC,IAAI,KAAK,CAAC,IAAI,KAAK,aAAa,IAAI,OAAO,CAAC,IAAI,KAAK,WAAW;gBAAE,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAC3F,CAAC;QACD,QAAQ;YACN,QAAQ,CAAC,MAAM,GAAG,CAAC;gBACnB,QAAQ,CAAC,KAAK,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,GAAG,OAAO,CAAC,QAAQ,IAAI,OAAO,CAAC,KAAK,EAAE,KAAK,GAAG,CAAC,SAAS,CAAC,CAAC;QACxF,MAAM,MAAM,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,CAAC;QACpE,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;YAC/B,MAAM,MAAM,GAAG,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;YACrC,KAAK,MAAM,GAAG,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,WAAW,EAAE,YAAY,CAAU,EAAE,CAAC;gBAC1E,MAAM,KAAK,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC;gBAC1B,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC;oBAAE,OAAO,EAAE,aAAa,EAAE,QAAQ,EAAE,CAAC;gBACtD,MAAM,CAAC,GAAG,CAAC,IAAI,KAAK,CAAC;YACvB,CAAC;QACH,CAAC;QACD,KAAK,GAAG,MAAM,CAAC;IACjB,CAAC;SAAM,IAAI,OAAO,KAAK,OAAO,EAAE,CAAC;QAC/B,MAAM,IAAI,GAAG,CAAC,GAAG,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC,kBAAkB,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;QACjF,IAAI,CAAC,IAAI;YAAE,OAAO,EAAE,CAAC;QACrB,IAAI,UAAmC,CAAC;QACxC,IAAI,CAAC;YACH,UAAU,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;QACxC,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,EAAE,CAAC;QACZ,CAAC;QACD,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,UAAU,CAAC,KAAK,CAAC;YAC3C,CAAC,CAAC,UAAU,CAAC,KAAK,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,KAAK,OAAO,CAAC;YACxE,CAAC,CAAC,EAAE,CAAC;QACP,QAAQ;YACN,KAAK,CAAC,MAAM,GAAG,CAAC;gBAChB,KAAK,CAAC,KAAK,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,UAAU,IAAI,MAAM,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC;QACzF,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,YAAY,CAAC,CAAC;QACnD,MAAM,UAAU,GACd,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,aAAa,CAAC,CAAC,KAAK,CAAC,CAAC,8BAA8B,IAAI,CAAC,CAAC;QACrF,IACE,KAAK,CAAC,MAAM,CAAC,cAAc,CAAC;YAC5B,KAAK,CAAC,MAAM,CAAC,eAAe,CAAC;YAC7B,KAAK,CAAC,MAAM,CAAC,cAAc,CAAC;YAC5B,KAAK,CAAC,UAAU,CAAC;YACjB,MAAM,CAAC,cAAc,GAAG,UAAU,IAAI,MAAM,CAAC,cAAc,EAC3D,CAAC;YACD,KAAK,GAAG;gBACN,KAAK,EAAE,MAAM,CAAC,cAAc,GAAG,MAAM,CAAC,cAAc,GAAG,UAAU;gBACjE,MAAM,EAAE,MAAM,CAAC,eAAe;gBAC9B,SAAS,EAAE,MAAM,CAAC,cAAc;gBAChC,UAAU;aACX,CAAC;YACF,YAAY,GAAG,UAAU,CAAC,UAAU,EAAE,KAAK,EAAE,MAAM,CAAC,QAAQ,CAAC,CAAC;QAChE,CAAC;IACH,CAAC;IACD,MAAM,QAAQ,GAAG,EAAE,aAAa,EAAE,QAAQ,EAAE,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC;IACtF,IAAI,CAAC,QAAQ,IAAI,CAAC,KAAK;QAAE,OAAO,QAAQ,CAAC;IACzC,IAAI,YAAY,KAAK,SAAS;QAC5B,OAAO,EAAE,GAAG,QAAQ,EAAE,UAAU,EAAE,YAAY,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC;IACzE,IAAI,CAAC,OAAO;QAAE,OAAO,QAAQ,CAAC;IAC9B,IACE,OAAO,CAAC,cAAc;QACtB,KAAK,CAAC,KAAK,GAAG,KAAK,CAAC,SAAS,GAAG,KAAK,CAAC,UAAU,GAAG,OAAO,CAAC,cAAc;QAEzE,OAAO,QAAQ,CAAC;IAClB,MAAM,UAAU,GACd,CAAC,KAAK,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK;QAC1B,KAAK,CAAC,MAAM,GAAG,OAAO,CAAC,MAAM;QAC7B,KAAK,CAAC,SAAS,GAAG,OAAO,CAAC,SAAS;QACnC,KAAK,CAAC,UAAU,GAAG,OAAO,CAAC,UAAU,CAAC;QACxC,SAAS,CAAC;IACZ,OAAO,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC;QAChC,CAAC,CAAC,EAAE,GAAG,QAAQ,EAAE,UAAU,EAAE,UAAU,EAAE,iBAAiB,EAAE;QAC5D,CAAC,CAAC,QAAQ,CAAC;AACf,CAAC"}
|
|
@@ -0,0 +1,10 @@
|
|
|
1
|
+
import type { EncryptedRecordStore } from "./encrypted-records.js";
|
|
2
|
+
import type { EvaluationInput } from "./types.js";
|
|
3
|
+
export declare function credentialExecution(input: EvaluationInput, home: string, env: NodeJS.ProcessEnv, records: EncryptedRecordStore): Promise<{
|
|
4
|
+
profile: {
|
|
5
|
+
model: string;
|
|
6
|
+
};
|
|
7
|
+
child: NodeJS.ProcessEnv;
|
|
8
|
+
secrets: string[];
|
|
9
|
+
}>;
|
|
10
|
+
//# sourceMappingURL=credential-execution.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"credential-execution.d.ts","sourceRoot":"","sources":["../../src/evaluation/credential-execution.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,wBAAwB,CAAC;AAEnE,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AAElD,wBAAsB,mBAAmB,CACvC,KAAK,EAAE,eAAe,EACtB,IAAI,EAAE,MAAM,EACZ,GAAG,EAAE,MAAM,CAAC,UAAU,EACtB,OAAO,EAAE,oBAAoB;;;;;;GAsE9B"}
|
|
@@ -0,0 +1,65 @@
|
|
|
1
|
+
import { writeFile } from "node:fs/promises";
|
|
2
|
+
import { join } from "node:path";
|
|
3
|
+
import { isDeepStrictEqual } from "node:util";
|
|
4
|
+
import { readAccount, secretPath } from "./account.js";
|
|
5
|
+
import { validateEndpoint } from "./credentials.js";
|
|
6
|
+
import { orgRecords } from "./org-records.js";
|
|
7
|
+
export async function credentialExecution(input, home, env, records) {
|
|
8
|
+
if (!input.credentials || !input.credentialOwnerId || !input.comparisonId)
|
|
9
|
+
throw new Error("Credential references missing");
|
|
10
|
+
records = orgRecords(records, input.credentialOrgId);
|
|
11
|
+
const account = await readAccount(records, input.credentialOwnerId);
|
|
12
|
+
const saved = account.comparisons
|
|
13
|
+
.find((entry) => entry.id === input.comparisonId && entry.repoId === input.repoId)
|
|
14
|
+
?.inputs.find((entry) => entry.id === input.id);
|
|
15
|
+
if (!saved || !isDeepStrictEqual(saved, input))
|
|
16
|
+
throw new Error("Execution does not match the reserved comparison");
|
|
17
|
+
const info = account.credentials.find((entry) => entry.id === input.credentials?.modelCredentialId && !entry.deleted);
|
|
18
|
+
const sandbox = account.credentials.find((entry) => entry.id === input.credentials?.sandboxCredentialId && !entry.deleted);
|
|
19
|
+
if (!info ||
|
|
20
|
+
!sandbox ||
|
|
21
|
+
sandbox.kind !== input.sandbox ||
|
|
22
|
+
info.kind !== input.credentials.provider)
|
|
23
|
+
throw new Error("Credential unavailable");
|
|
24
|
+
const modelSecret = (await records.read(secretPath(input.credentialOwnerId, info.id)))?.value;
|
|
25
|
+
const sandboxSecret = (await records.read(secretPath(input.credentialOwnerId, sandbox.id)))?.value;
|
|
26
|
+
if (!modelSecret || !sandboxSecret)
|
|
27
|
+
throw new Error("Credential unavailable");
|
|
28
|
+
const child = {
|
|
29
|
+
PATH: env.PATH,
|
|
30
|
+
HOME: home,
|
|
31
|
+
TMPDIR: home,
|
|
32
|
+
LANG: "C.UTF-8",
|
|
33
|
+
PYTHONUNBUFFERED: "1",
|
|
34
|
+
};
|
|
35
|
+
const secrets = [modelSecret.value, sandboxSecret.value, sandboxSecret.tokenId ?? ""].filter(Boolean);
|
|
36
|
+
if (info.auth === "codex-login") {
|
|
37
|
+
if (input.harnesses.some((harness) => harness !== "codex"))
|
|
38
|
+
throw new Error("Codex login requires Codex harness");
|
|
39
|
+
const authPath = join(home, "codex-auth.json");
|
|
40
|
+
await writeFile(authPath, modelSecret.value, { mode: 0o600 });
|
|
41
|
+
child.CODEX_AUTH_JSON_PATH = authPath;
|
|
42
|
+
const auth = JSON.parse(modelSecret.value);
|
|
43
|
+
secrets.push(...Object.values(auth.tokens).filter((value) => typeof value === "string"));
|
|
44
|
+
}
|
|
45
|
+
else {
|
|
46
|
+
const name = info.kind === "anthropic"
|
|
47
|
+
? "ANTHROPIC_API_KEY"
|
|
48
|
+
: info.kind === "openrouter"
|
|
49
|
+
? "OPENROUTER_API_KEY"
|
|
50
|
+
: "OPENAI_API_KEY";
|
|
51
|
+
child[name] = modelSecret.value;
|
|
52
|
+
}
|
|
53
|
+
if (info.kind === "custom")
|
|
54
|
+
child.OPENAI_BASE_URL = await validateEndpoint(info.endpoint ?? "", env);
|
|
55
|
+
if (sandbox.kind === "modal") {
|
|
56
|
+
child.MODAL_TOKEN_ID = sandboxSecret.tokenId;
|
|
57
|
+
child.MODAL_TOKEN_SECRET = sandboxSecret.value;
|
|
58
|
+
}
|
|
59
|
+
else if (sandbox.kind === "e2b")
|
|
60
|
+
child.E2B_API_KEY = sandboxSecret.value;
|
|
61
|
+
else if (sandbox.kind === "daytona")
|
|
62
|
+
child.DAYTONA_API_KEY = sandboxSecret.value;
|
|
63
|
+
return { profile: { model: input.modelName }, child, secrets };
|
|
64
|
+
}
|
|
65
|
+
//# sourceMappingURL=credential-execution.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"credential-execution.js","sourceRoot":"","sources":["../../src/evaluation/credential-execution.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAC7C,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,iBAAiB,EAAE,MAAM,WAAW,CAAC;AAC9C,OAAO,EAAE,WAAW,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AACvD,OAAO,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AAEpD,OAAO,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAG9C,MAAM,CAAC,KAAK,UAAU,mBAAmB,CACvC,KAAsB,EACtB,IAAY,EACZ,GAAsB,EACtB,OAA6B;IAE7B,IAAI,CAAC,KAAK,CAAC,WAAW,IAAI,CAAC,KAAK,CAAC,iBAAiB,IAAI,CAAC,KAAK,CAAC,YAAY;QACvE,MAAM,IAAI,KAAK,CAAC,+BAA+B,CAAC,CAAC;IACnD,OAAO,GAAG,UAAU,CAAC,OAAO,EAAE,KAAK,CAAC,eAAe,CAAC,CAAC;IACrD,MAAM,OAAO,GAAG,MAAM,WAAW,CAAC,OAAO,EAAE,KAAK,CAAC,iBAAiB,CAAC,CAAC;IACpE,MAAM,KAAK,GAAG,OAAO,CAAC,WAAW;SAC9B,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,KAAK,CAAC,YAAY,IAAI,KAAK,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,CAAC;QAClF,EAAE,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,KAAK,CAAC,EAAE,CAAC,CAAC;IAClD,IAAI,CAAC,KAAK,IAAI,CAAC,iBAAiB,CAAC,KAAK,EAAE,KAAK,CAAC;QAC5C,MAAM,IAAI,KAAK,CAAC,kDAAkD,CAAC,CAAC;IACtE,MAAM,IAAI,GAAG,OAAO,CAAC,WAAW,CAAC,IAAI,CACnC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,KAAK,CAAC,WAAW,EAAE,iBAAiB,IAAI,CAAC,KAAK,CAAC,OAAO,CAC/E,CAAC;IACF,MAAM,OAAO,GAAG,OAAO,CAAC,WAAW,CAAC,IAAI,CACtC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,KAAK,CAAC,WAAW,EAAE,mBAAmB,IAAI,CAAC,KAAK,CAAC,OAAO,CACjF,CAAC;IACF,IACE,CAAC,IAAI;QACL,CAAC,OAAO;QACR,OAAO,CAAC,IAAI,KAAK,KAAK,CAAC,OAAO;QAC9B,IAAI,CAAC,IAAI,KAAK,KAAK,CAAC,WAAW,CAAC,QAAQ;QAExC,MAAM,IAAI,KAAK,CAAC,wBAAwB,CAAC,CAAC;IAC5C,MAAM,WAAW,GAAG,CAClB,MAAM,OAAO,CAAC,IAAI,CAAoB,UAAU,CAAC,KAAK,CAAC,iBAAiB,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC,CACpF,EAAE,KAAK,CAAC;IACT,MAAM,aAAa,GAAG,CACpB,MAAM,OAAO,CAAC,IAAI,CAChB,UAAU,CAAC,KAAK,CAAC,iBAAiB,EAAE,OAAO,CAAC,EAAE,CAAC,CAChD,CACF,EAAE,KAAK,CAAC;IACT,IAAI,CAAC,WAAW,IAAI,CAAC,aAAa;QAAE,MAAM,IAAI,KAAK,CAAC,wBAAwB,CAAC,CAAC;IAC9E,MAAM,KAAK,GAAsB;QAC/B,IAAI,EAAE,GAAG,CAAC,IAAI;QACd,IAAI,EAAE,IAAI;QACV,MAAM,EAAE,IAAI;QACZ,IAAI,EAAE,SAAS;QACf,gBAAgB,EAAE,GAAG;KACtB,CAAC;IACF,MAAM,OAAO,GAAG,CAAC,WAAW,CAAC,KAAK,EAAE,aAAa,CAAC,KAAK,EAAE,aAAa,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,MAAM,CAC1F,OAAO,CACR,CAAC;IACF,IAAI,IAAI,CAAC,IAAI,KAAK,aAAa,EAAE,CAAC;QAChC,IAAI,KAAK,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,KAAK,OAAO,CAAC;YACxD,MAAM,IAAI,KAAK,CAAC,oCAAoC,CAAC,CAAC;QACxD,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,EAAE,iBAAiB,CAAC,CAAC;QAC/C,MAAM,SAAS,CAAC,QAAQ,EAAE,WAAW,CAAC,KAAK,EAAE,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC;QAC9D,KAAK,CAAC,oBAAoB,GAAG,QAAQ,CAAC;QACtC,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC;QAC3C,OAAO,CAAC,IAAI,CACV,GAAG,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,CAAC,KAAK,EAAmB,EAAE,CAAC,OAAO,KAAK,KAAK,QAAQ,CAAC,CAC5F,CAAC;IACJ,CAAC;SAAM,CAAC;QACN,MAAM,IAAI,GACR,IAAI,CAAC,IAAI,KAAK,WAAW;YACvB,CAAC,CAAC,mBAAmB;YACrB,CAAC,CAAC,IAAI,CAAC,IAAI,KAAK,YAAY;gBAC1B,CAAC,CAAC,oBAAoB;gBACtB,CAAC,CAAC,gBAAgB,CAAC;QACzB,KAAK,CAAC,IAAI,CAAC,GAAG,WAAW,CAAC,KAAK,CAAC;IAClC,CAAC;IACD,IAAI,IAAI,CAAC,IAAI,KAAK,QAAQ;QACxB,KAAK,CAAC,eAAe,GAAG,MAAM,gBAAgB,CAAC,IAAI,CAAC,QAAQ,IAAI,EAAE,EAAE,GAAG,CAAC,CAAC;IAC3E,IAAI,OAAO,CAAC,IAAI,KAAK,OAAO,EAAE,CAAC;QAC7B,KAAK,CAAC,cAAc,GAAG,aAAa,CAAC,OAAO,CAAC;QAC7C,KAAK,CAAC,kBAAkB,GAAG,aAAa,CAAC,KAAK,CAAC;IACjD,CAAC;SAAM,IAAI,OAAO,CAAC,IAAI,KAAK,KAAK;QAAE,KAAK,CAAC,WAAW,GAAG,aAAa,CAAC,KAAK,CAAC;SACtE,IAAI,OAAO,CAAC,IAAI,KAAK,SAAS;QAAE,KAAK,CAAC,eAAe,GAAG,aAAa,CAAC,KAAK,CAAC;IACjF,OAAO,EAAE,OAAO,EAAE,EAAE,KAAK,EAAE,KAAK,CAAC,SAAS,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC;AACjE,CAAC"}
|
|
@@ -0,0 +1,40 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
import type { ArtifactStore } from "../artifacts.js";
|
|
3
|
+
import { type CredentialInfo } from "./account.js";
|
|
4
|
+
import type { EncryptedRecordStore } from "./encrypted-records.js";
|
|
5
|
+
export declare const credentialSchema: z.ZodObject<{
|
|
6
|
+
name: z.ZodString;
|
|
7
|
+
kind: z.ZodEnum<{
|
|
8
|
+
modal: "modal";
|
|
9
|
+
vercel: "vercel";
|
|
10
|
+
e2b: "e2b";
|
|
11
|
+
daytona: "daytona";
|
|
12
|
+
custom: "custom";
|
|
13
|
+
openai: "openai";
|
|
14
|
+
anthropic: "anthropic";
|
|
15
|
+
openrouter: "openrouter";
|
|
16
|
+
}>;
|
|
17
|
+
auth: z.ZodDefault<z.ZodEnum<{
|
|
18
|
+
"api-key": "api-key";
|
|
19
|
+
"codex-login": "codex-login";
|
|
20
|
+
}>>;
|
|
21
|
+
value: z.ZodString;
|
|
22
|
+
tokenId: z.ZodOptional<z.ZodString>;
|
|
23
|
+
teamId: z.ZodOptional<z.ZodString>;
|
|
24
|
+
projectId: z.ZodOptional<z.ZodString>;
|
|
25
|
+
endpoint: z.ZodOptional<z.ZodURL>;
|
|
26
|
+
}, z.core.$strict>;
|
|
27
|
+
export type CredentialDraft = z.infer<typeof credentialSchema>;
|
|
28
|
+
export declare function validateEndpoint(endpoint: string, env: NodeJS.ProcessEnv): Promise<string>;
|
|
29
|
+
export declare function saveCredential(records: EncryptedRecordStore, ownerId: number, draft: CredentialDraft, env: NodeJS.ProcessEnv, migratedFrom?: string): Promise<CredentialInfo>;
|
|
30
|
+
export declare function deleteCredential(records: EncryptedRecordStore, store: ArtifactStore, ownerId: number, id: string): Promise<void>;
|
|
31
|
+
export declare function listCredentials(records: EncryptedRecordStore, ownerId: number): Promise<{
|
|
32
|
+
id: string;
|
|
33
|
+
name: string;
|
|
34
|
+
kind: import("./catalog.js").CatalogProvider | import("./catalog.js").HostedSandbox | "vercel";
|
|
35
|
+
auth: "api-key" | "codex-login";
|
|
36
|
+
createdAt: string;
|
|
37
|
+
endpoint?: string;
|
|
38
|
+
deleted?: boolean;
|
|
39
|
+
}[]>;
|
|
40
|
+
//# sourceMappingURL=credentials.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"credentials.d.ts","sourceRoot":"","sources":["../../src/evaluation/credentials.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AACrD,OAAO,EAAE,KAAK,cAAc,EAA0C,MAAM,cAAc,CAAC;AAC3F,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,wBAAwB,CAAC;AAInE,eAAO,MAAM,gBAAgB;;;;;;;;;;;;;;;;;;;;;kBA+DzB,CAAC;AACL,MAAM,MAAM,eAAe,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,gBAAgB,CAAC,CAAC;AAC/D,wBAAsB,gBAAgB,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,EAAE,MAAM,CAAC,UAAU,mBAc9E;AACD,wBAAsB,cAAc,CAClC,OAAO,EAAE,oBAAoB,EAC7B,OAAO,EAAE,MAAM,EACf,KAAK,EAAE,eAAe,EACtB,GAAG,EAAE,MAAM,CAAC,UAAU,EACtB,YAAY,CAAC,EAAE,MAAM,GACpB,OAAO,CAAC,cAAc,CAAC,CAyCzB;AACD,wBAAsB,gBAAgB,CACpC,OAAO,EAAE,oBAAoB,EAC7B,KAAK,EAAE,aAAa,EACpB,OAAO,EAAE,MAAM,EACf,EAAE,EAAE,MAAM,iBAoBX;AACD,wBAAsB,eAAe,CAAC,OAAO,EAAE,oBAAoB,EAAE,OAAO,EAAE,MAAM;;;;;;;;KAInF"}
|