Jannik/PaperClipAI
1
0 Fork 0
mirror of https://github.com/paperclipai/paperclip.git synced 2026-11-07 07:42:19 +01:00
Code Issues Packages Projects Releases Wiki Activity

test(e2e): reconcile finite catalog authoring checkpoint

Browse Source
This commit is contained in:
Dotta committed 2026-10-26 19:16:40 -05:00
1 parent 50775edc48
commit a1859e0e67
4 files changed
+1061 -161
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

tests/runner-e2e/chat-campaign-manifest.json
+114 -106
View File
@@ -25,8 +25,8 @@
"sourceSha256": "14324aa63a93c0ac919e7b8650bed87d8ae24ca7ebdbcbc68f5a441b925bf9e8",
"scenarioSha256": "41527a821ba793d6dbb1053a7ec68a5b9787b56cf856c2d6b7aa50f6653b107b",
"owner": "eval_identity",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.access.cli-authorization",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -126,8 +126,8 @@
"sourceSha256": "14324aa63a93c0ac919e7b8650bed87d8ae24ca7ebdbcbc68f5a441b925bf9e8",
"scenarioSha256": "e33310e9624ea49cc21e83dae75e4cf08e571fb645bb8c66156b32a1f28ee964",
"owner": "eval_identity",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.access.sign-in",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -146,8 +146,8 @@
"sourceSha256": "a9b43a3c0500b8628031ed6178dec40c80b1dc0b6df116a32597ba74c9e0de6f",
"scenarioSha256": "e0d80d373468c2487f2c29bdd9809bb2fbb98557c64c68b594f4104c0758f0b0",
"owner": "eval_domains",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.activity.agent-feedback",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -206,8 +206,8 @@
"sourceSha256": "a9b43a3c0500b8628031ed6178dec40c80b1dc0b6df116a32597ba74c9e0de6f",
"scenarioSha256": "a340cf61cb2fda86d420ca576e98c26296f4131a59c3d978d1ecb164475b3d3c",
"owner": "eval_domains",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.activity.find-runs",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -330,8 +330,8 @@
"sourceSha256": "19defd25f25ebd32749e069a1db15af779ba08b482e136a15473ab6897053f4a",
"scenarioSha256": "7bb7a05011d7e278bc3060ac9fd04ad37be3cdec7ab0e99679ee45ada4a40f4c",
"owner": "eval_runtime",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.agent-chat.remember-and-follow-up",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -372,8 +372,8 @@
"sourceSha256": "19defd25f25ebd32749e069a1db15af779ba08b482e136a15473ab6897053f4a",
"scenarioSha256": "08d5cb833e45393a84cde5225b558c708546ef33be2273372a851331c6350ffd",
"owner": "eval_runtime",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.agent-chat.track-delegated-work",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -600,8 +600,8 @@
"sourceSha256": "a5aa89ff1a55f7a2f580230c72f20d05420d272dbad365c4d7dd963016332bf3",
"scenarioSha256": "35244f0953539274eea8d2bf250d9eb0b1bc404c62d64875bae6e96f40833557",
"owner": "eval_runtime",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.agents.hire-needs-approval",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -700,8 +700,8 @@
"sourceSha256": "a5aa89ff1a55f7a2f580230c72f20d05420d272dbad365c4d7dd963016332bf3",
"scenarioSha256": "38b4b159d89acf62c5257edc93045c91bc93e926e0366a6ddc5402df3fc7d31e",
"owner": "eval_runtime",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.agents.retire-agent",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -842,9 +842,9 @@
"sourceSha256": "b87c34edd99aa5cce96cf2196327b66366ef24bbcea96dc8c86d404030372901",
"scenarioSha256": "8e1b5269d92d5b144054e7ea43f08aed14f516dbe1bff0db73e971bb3bdb5f96",
"owner": "eval_external",
"readiness": "implemented",
"readiness": "partial",
"executionId": "chat-features.runner-codex-subscription.local.assistant-connections.connect-external-assistant",
"blocker": null,
"blocker": "named-chatgpt-human-setup",
"attempts": [],
"qualifyingDefinitionDigest": null,
"qualifyingGraderDigest": null
@@ -1412,8 +1412,8 @@
"sourceSha256": "d3d89e58aeb7e36198c9a1109c31ff501e927621ce68f36d78a10b5d00839565",
"scenarioSha256": "1c29be9a9e8351e00602d37ca1cf670c0dc83cf8474739c7983209764065670c",
"owner": "eval_campaign",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.cli-operations.choose-cli-context",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -1433,8 +1433,8 @@
"sourceSha256": "d3d89e58aeb7e36198c9a1109c31ff501e927621ce68f36d78a10b5d00839565",
"scenarioSha256": "6eb82a4a3f0dd7a7c197063dc685fdb7cdaa59e07532157e40983643e5bebfe9",
"owner": "eval_campaign",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.cli-operations.create-from-cli",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -1454,8 +1454,8 @@
"sourceSha256": "d3d89e58aeb7e36198c9a1109c31ff501e927621ce68f36d78a10b5d00839565",
"scenarioSha256": "6c82c34c435e528504b136f65d46062011c0c1a094047592bdd76c197d99e9bd",
"owner": "eval_campaign",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.cli-operations.export-document-cli",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -1475,8 +1475,8 @@
"sourceSha256": "d3d89e58aeb7e36198c9a1109c31ff501e927621ce68f36d78a10b5d00839565",
"scenarioSha256": "3496eafcb5cf5aebd86a1efaf1db636073e9ea3254d43421e3c09af8e45928e3",
"owner": "eval_campaign",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.cli-operations.feedback-bundle",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -1496,8 +1496,8 @@
"sourceSha256": "d3d89e58aeb7e36198c9a1109c31ff501e927621ce68f36d78a10b5d00839565",
"scenarioSha256": "788d495f8d426b07844edb19a81d4d0cd8bb8ee1db611307f175708c9a6f09a7",
"owner": "eval_campaign",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.cli-operations.inspect-run-cli",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -1517,8 +1517,8 @@
"sourceSha256": "d3d89e58aeb7e36198c9a1109c31ff501e927621ce68f36d78a10b5d00839565",
"scenarioSha256": "3036873f567ce80ea64c7ee3d0753afb04686314b02a6d775ea13bd0f2a7f6ee",
"owner": "eval_campaign",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.cli-operations.owned-dev-instance",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -1557,8 +1557,8 @@
"sourceSha256": "395472a5531e93259672c2c07f1819992f6f14ccf11513e1990aa3c576f3fa27",
"scenarioSha256": "4b53ad7ca9885c939ff6b69d685078be0ff1756e79f7f132b7253d2959796abb",
"owner": "eval_identity",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.companies.archive-company",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -1699,8 +1699,8 @@
"sourceSha256": "395472a5531e93259672c2c07f1819992f6f14ccf11513e1990aa3c576f3fa27",
"scenarioSha256": "04c42aee139857a9158fffdbc57f062a5ead44b06d438f9aea9926fff5e1ea66",
"owner": "eval_identity",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.companies.switch-company",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -1990,8 +1990,8 @@
"sourceSha256": "c09b9eeb0b76dd462dc2a42678d97859c4289e6265cf76097bde8dd7c8322cf1",
"scenarioSha256": "1381c6828ab6059e280782897ac9cc1d7a47cc6b4cbb317ae3c4716040c04890",
"owner": "eval_campaign",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.developer-labs.component-reference",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2011,8 +2011,8 @@
"sourceSha256": "c09b9eeb0b76dd462dc2a42678d97859c4289e6265cf76097bde8dd7c8322cf1",
"scenarioSha256": "995508eb028d95beffdb28b2c546a1606ddb903911ba0d5b824c07b0260c73b3",
"owner": "eval_campaign",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.developer-labs.gated-lab",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2031,8 +2031,8 @@
"sourceSha256": "c09b9eeb0b76dd462dc2a42678d97859c4289e6265cf76097bde8dd7c8322cf1",
"scenarioSha256": "a586a03c729abbb3ab305fe789c3d0e00d93d13da2d0e5bc216ea3c43b6e0f29",
"owner": "eval_campaign",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.developer-labs.interaction-fixture",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2051,8 +2051,8 @@
"sourceSha256": "c09b9eeb0b76dd462dc2a42678d97859c4289e6265cf76097bde8dd7c8322cf1",
"scenarioSha256": "7a36242f134f6783deb659f0cc38f40598776c03e4ba90bd88e681a134892c88",
"owner": "eval_campaign",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.developer-labs.long-thread-performance",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2072,8 +2072,8 @@
"sourceSha256": "766da628af9419eaf0b112ec635cddb8e9d1fca903d1dc4602f69f31b0c539ca",
"scenarioSha256": "48f3a32f0e2a1aa4878b22c658ee0987056dfd798819e938e59210ba51678706",
"owner": "eval_domains",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.documents-artifacts.find-artifact",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2093,8 +2093,8 @@
"sourceSha256": "766da628af9419eaf0b112ec635cddb8e9d1fca903d1dc4602f69f31b0c539ca",
"scenarioSha256": "887b384828ded827e052a068f2d50513677eb82be1cf0dbe4ca204d0963519ce",
"owner": "eval_domains",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.documents-artifacts.private-download",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2114,8 +2114,8 @@
"sourceSha256": "766da628af9419eaf0b112ec635cddb8e9d1fca903d1dc4602f69f31b0c539ca",
"scenarioSha256": "400e48d0b70407f718712718f3252747339152473d2d7fc8078e26e88080996a",
"owner": "eval_domains",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.documents-artifacts.produce-download",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2175,8 +2175,8 @@
"sourceSha256": "766da628af9419eaf0b112ec635cddb8e9d1fca903d1dc4602f69f31b0c539ca",
"scenarioSha256": "b4d273782ebdb6ab66ab8c10e1af3a9fd7249a9b896a2a723dd8a01a6b2606ff",
"owner": "eval_domains",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.documents-artifacts.revise-delivery",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2621,8 +2621,8 @@
"sourceSha256": "afc939f4eec821edd38b8e819274ede302c10550aca68b2ff72c5ebf870aed05",
"scenarioSha256": "a02544496a0f5a71bd58a5c00f5df5ac4cbe3199b624c5324fa4defa5d53b856",
"owner": "eval_domains",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.inbox-search.archive-done",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2681,8 +2681,8 @@
"sourceSha256": "afc939f4eec821edd38b8e819274ede302c10550aca68b2ff72c5ebf870aed05",
"scenarioSha256": "703eb37ed81af1af5c2b3739409d85e4062125d8ea286bf8d93a7f15c056337d",
"owner": "eval_domains",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.inbox-search.find-design",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2702,8 +2702,8 @@
"sourceSha256": "afc939f4eec821edd38b8e819274ede302c10550aca68b2ff72c5ebf870aed05",
"scenarioSha256": "58051c51f23a3bc27e7333ae6cd0afa6f66de08513f50beae58eaf0cab0dcec9",
"owner": "eval_domains",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.inbox-search.notification-audience",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2722,8 +2722,8 @@
"sourceSha256": "afc939f4eec821edd38b8e819274ede302c10550aca68b2ff72c5ebf870aed05",
"scenarioSha256": "855fc33bc106cfb11caea791ccb7b52e9fb00dcd20f960bfcda90f4dd36bcc8f",
"owner": "eval_domains",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.inbox-search.restore-archive",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2743,8 +2743,8 @@
"sourceSha256": "afc939f4eec821edd38b8e819274ede302c10550aca68b2ff72c5ebf870aed05",
"scenarioSha256": "b27eaf08a3f7a9602b4d5d08569ada924c5c5bcf03c1d5d6399b2188177f328c",
"owner": "eval_domains",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.inbox-search.show-decisions",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2783,8 +2783,8 @@
"sourceSha256": "7cf3dbf8bcddd40a9b5d7ca9430231c15dc2d10346eadad96ab81617c20c6aa4",
"scenarioSha256": "211416b7d059a27afa2120548496a3bcdb1cac14b3a9623148e3a061a2841306",
"owner": "eval_campaign",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.instance-operations.backup-and-restore",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2803,8 +2803,8 @@
"sourceSha256": "7cf3dbf8bcddd40a9b5d7ca9430231c15dc2d10346eadad96ab81617c20c6aa4",
"scenarioSha256": "486dc1d42513db88f52e90a3b91cb6ba644e62ee739e126140c7cc3c243a4b94",
"owner": "eval_campaign",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.instance-operations.diagnose-health",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2823,8 +2823,8 @@
"sourceSha256": "7cf3dbf8bcddd40a9b5d7ca9430231c15dc2d10346eadad96ab81617c20c6aa4",
"scenarioSha256": "778883565814847de78d8c1cba156e9d659f4ed536f422e8435a852d35b464f6",
"owner": "eval_campaign",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.instance-operations.instance-settings",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2843,9 +2843,9 @@
"sourceSha256": "7cf3dbf8bcddd40a9b5d7ca9430231c15dc2d10346eadad96ab81617c20c6aa4",
"scenarioSha256": "286887bd9a36e5ba926b5420db411cab8ac05ddb0bffa41e6b014594f87d369b",
"owner": "eval_campaign",
"readiness": "unimplemented",
"executionId": null,
"blocker": null,
"readiness": "partial",
"executionId": "chat-features.runner-codex-subscription.local.instance-operations.provision-runtime",
"blocker": "installed-source-activation-and-artifact-fingerprint-lane",
"attempts": [],
"qualifyingDefinitionDigest": null,
"qualifyingGraderDigest": null
@@ -2863,8 +2863,8 @@
"sourceSha256": "7cf3dbf8bcddd40a9b5d7ca9430231c15dc2d10346eadad96ab81617c20c6aa4",
"scenarioSha256": "fbfe0844b0b859ea1a2ee5d0368df09dcabe6bbbd04820f9f4eba9dfdf3e2448",
"owner": "eval_campaign",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.instance-operations.restart-service",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2883,9 +2883,9 @@
"sourceSha256": "7cf3dbf8bcddd40a9b5d7ca9430231c15dc2d10346eadad96ab81617c20c6aa4",
"scenarioSha256": "e1dae8554a765745a6e05392a5e42d3532716904b30c2c64c1f9c4fd789866d7",
"owner": "eval_campaign",
"readiness": "unimplemented",
"executionId": null,
"blocker": null,
"readiness": "partial",
"executionId": "chat-features.runner-codex-subscription.local.instance-operations.update-instance",
"blocker": "installed-source-activation-and-artifact-fingerprint-lane",
"attempts": [],
"qualifyingDefinitionDigest": null,
"qualifyingGraderDigest": null
@@ -2944,8 +2944,8 @@
"sourceSha256": "097af7c681a618cf3b1b543a6218ebb732f94b0873e388d9795b88d33894b3ab",
"scenarioSha256": "49dbb46581e775b36016577113475f9555a996e7c1868c511f1d7c544260aec8",
"owner": "eval_identity",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.navigation-preferences.edit-profile",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -2986,9 +2986,9 @@
"sourceSha256": "097af7c681a618cf3b1b543a6218ebb732f94b0873e388d9795b88d33894b3ab",
"scenarioSha256": "7bba464b3edbdb96b9181c6b4f89cdfdab38d94a5bcb4e3f5711ead767dd8c91",
"owner": "eval_identity",
"readiness": "partial",
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.navigation-preferences.quick-search",
"blocker": "filter-preservation; authenticated-search-denial",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
"qualifyingGraderDigest": null
@@ -3006,8 +3006,8 @@
"sourceSha256": "097af7c681a618cf3b1b543a6218ebb732f94b0873e388d9795b88d33894b3ab",
"scenarioSha256": "c9aa7fdd8e6be91bcf411c81cb6151419fd6d1580b1cabd70a527562cc5ad75e",
"owner": "eval_identity",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.navigation-preferences.sidebar-order",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -3397,9 +3397,9 @@
"sourceSha256": "f72b3d7e4ba0b4bb52b2912e3bce58c5f75fa0561411cb86330f2f5ba6d2df7f",
"scenarioSha256": "d2647b32f3939c2b08738340d02cf5c71b82326f4e3299accbb0944cf0f0f1e1",
"owner": "eval_identity",
"readiness": "partial",
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.questions-and-approvals.ambiguous-yes",
"blocker": "stale-and-expired-proposal-rejection",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
"qualifyingGraderDigest": null
@@ -3418,9 +3418,9 @@
"sourceSha256": "f72b3d7e4ba0b4bb52b2912e3bce58c5f75fa0561411cb86330f2f5ba6d2df7f",
"scenarioSha256": "3db91caac72e32867b80c1006fc54864936053a285b52582722345ec6b60aea2",
"owner": "eval_identity",
"readiness": "partial",
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.questions-and-approvals.approve-tool-once",
"blocker": "second-action-requires-distinct-approval; approved-service-failure-retained",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
"qualifyingGraderDigest": null
@@ -3439,9 +3439,9 @@
"sourceSha256": "f72b3d7e4ba0b4bb52b2912e3bce58c5f75fa0561411cb86330f2f5ba6d2df7f",
"scenarioSha256": "2c4bda5093eb3f29bdd60b16351a4f98ebc985e41cd6fb763aa2637bacb357d6",
"owner": "eval_identity",
"readiness": "partial",
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.questions-and-approvals.decline-tool",
"blocker": "decline-request-must-already-be-pending-before-original-query",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
"qualifyingGraderDigest": null
@@ -3460,9 +3460,9 @@
"sourceSha256": "f72b3d7e4ba0b4bb52b2912e3bce58c5f75fa0561411cb86330f2f5ba6d2df7f",
"scenarioSha256": "67251ef988b8bd96ea0bf1b237d793495ed09370b35df82f7f67dae435d95a5d",
"owner": "eval_identity",
"readiness": "partial",
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.questions-and-approvals.defer-and-reopen",
"blocker": "draft-state-retention; protected-gate-not-waived; later-answer-continuation",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
"qualifyingGraderDigest": null
@@ -3521,9 +3521,9 @@
"sourceSha256": "f72b3d7e4ba0b4bb52b2912e3bce58c5f75fa0561411cb86330f2f5ba6d2df7f",
"scenarioSha256": "ff3b9bca8076608a758954d05b3a1c73d96f72e8be94c67720e11db55e03f9e2",
"owner": "eval_identity",
"readiness": "partial",
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.questions-and-approvals.multi-part-question",
"blocker": "submitted-set-resumes-work; accurate-agent-summary",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
"qualifyingGraderDigest": null
@@ -3542,9 +3542,9 @@
"sourceSha256": "f72b3d7e4ba0b4bb52b2912e3bce58c5f75fa0561411cb86330f2f5ba6d2df7f",
"scenarioSha256": "9ca36d3ec76ce296651c2cd403e41960125af1fbc7702802cc3b36549adf6855",
"owner": "eval_identity",
"readiness": "partial",
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.questions-and-approvals.necessary-question",
"blocker": "provider-native-question-correlation; incidental-settings-exclusion; task-brief-version-provenance",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
"qualifyingGraderDigest": null
@@ -3626,9 +3626,9 @@
"sourceSha256": "f74fb387ab717d656e4ce11b8abc03960946f0cba83ccf76e83f4ac77c520a3a",
"scenarioSha256": "1a24782c0adce5db07abd005849c5b7770936902d43d18ba51ff57e117e6a4cf",
"owner": "eval_runtime",
"readiness": "unimplemented",
"executionId": null,
"blocker": null,
"readiness": "partial",
"executionId": "chat-features.runner-codex-subscription.local.recovery.browser-reconnect",
"blocker": "Reconnect after a genuinely failed native run remains unexercised; successful and still-running native phases plus failed-evidence negative controls are authored.",
"attempts": [],
"qualifyingDefinitionDigest": null,
"qualifyingGraderDigest": null
@@ -3976,8 +3976,8 @@
"sourceSha256": "15d3741baab816afe14d24043d8229010fc204146ccb4bb2f525eb0f8d3037fa",
"scenarioSha256": "dbcbce44978644b3a10b1e8ad2cea9449d23eeb0cdf7bf6511450a6f3d08a7ed",
"owner": "eval_runtime",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.runs-adapters.reset-session",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -4695,8 +4695,8 @@
"sourceSha256": "31fca6be61221bb970b7c67f1b9234aa0705162571b46c99592926a286b240f0",
"scenarioSha256": "6255dc9a8c2fa505c544d6c46d07c6e92380025897095bfabb4def986b802746",
"owner": "eval_domains",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.tasks.assignment-conflict",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -4863,8 +4863,8 @@
"sourceSha256": "31fca6be61221bb970b7c67f1b9234aa0705162571b46c99592926a286b240f0",
"scenarioSha256": "1b7a53fb395d169e0c40bb258b38ff658c39dfff4c9eede4810d912689f5917b",
"owner": "eval_domains",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.tasks.private-task",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -5029,8 +5029,8 @@
"sourceSha256": "6b4be6151f155981ae7a68ac69c3a31420d54fe33f1ac567de0db73b651b3ae7",
"scenarioSha256": "71548aa07a1e910b82535691a66f9e8d4f9dafc911191b43e9460a32fce1bb98",
"owner": "eval_runtime",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.workspaces.close-and-reopen",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -5049,8 +5049,8 @@
"sourceSha256": "6b4be6151f155981ae7a68ac69c3a31420d54fe33f1ac567de0db73b651b3ae7",
"scenarioSha256": "db6dd1bf35167ccc383cc913aef0110e41cca2c553d3ff9ff338fde2681bc33d",
"owner": "eval_runtime",
"readiness": "unimplemented",
"executionId": null,
"readiness": "implemented",
"executionId": "chat-features.runner-codex-subscription.local.workspaces.inspect-git",
"blocker": null,
"attempts": [],
"qualifyingDefinitionDigest": null,
@@ -5160,5 +5160,13 @@
"qualifyingDefinitionDigest": null,
"qualifyingGraderDigest": null
}
]
],
"implementationCheckpoint": {
"sourceRevision": "b33f3b8b649cc597ffe9a4a66fc834e47fd93e0e",
"descriptorSha256": "6273f9afea18717074abe13bda4bc3132485c2c73245a42b9bee7f25c07a3c73",
"authored": 110,
"complete": 103,
"partial": 7,
"unimplemented": 134
}
}
tests/runner-e2e/chat-campaign.ts
+2
View File
@@ -48,6 +48,8 @@ export interface ChatCampaign {
denominator: number; areas: number; capabilities: number;
requiredPasses: number;
corePrs: string[]; evalPr: string | null;
/** Authored metadata only; never passing execution evidence. */
implementationCheckpoint?: { sourceRevision: string; descriptorSha256: string; authored: number; complete: number; partial: number; unimplemented: number };
scenarios: CampaignScenario[];
}
const digest = (value: string) => createHash("sha256").update(value).digest("hex");
tests/runner-e2e/chat-catalog-descriptors.json
+921 -52
View File
File diff suppressed because it is too large. Load diff