Datasets:
archive_bytes int64 | archive_sha256 string | archive_url string | base_commit string | candidate_revision string | cli_framework string | completed bool | deepseek_cli_exit_code int64 | deepseek_model string | deepseek_reward float64 | deepseek_session_id string | dependency_index_path string | evidence_path string | f2p_count int64 | fix_commit string | glm_cli_exit_code int64 | glm_model string | glm_reward float64 | glm_session_id string | hf_filename string | hf_repo_id string | hf_repo_type string | hf_revision string | image_id string | independent_acceptance_recorded bool | instruction string | instruction_path string | instruction_sha256 string | language string | original_p2p_count int64 | package_sha256 string | publication_kind string | quality_note string | quality_status string | schema string | scope string | source_repo string | source_repo_url string | summary_zh string | task_id string | title string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
424,482,987 | 9dd4d09ddeb0bfd83be9822d63f31afac7c2c331a790ba456de899186778ea94 | https://huggingface.co/datasets/Zeng-Weijun/taskforge-rack-harbor-v2/resolve/386d389810111b83a9e1dda7c23243fd79bb87f8/bundles/9dd4d09ddeb0bfd83be9822d63f31afac7c2c331a790ba456de899186778ea94/delivery.tar.gz | 6ee0826dcae55ed1e06b2c5701981221e979e1e6 | sha256:dcdc28154eeaf8a063b954a3762d3370b5bd8be902dd89187d2ba24f5e42dc9f | claude-code-cli | true | 0 | DeepSeek-V4-Flash | 1 | 4560da4a-0ec9-49f7-b83e-43a411f1dc3e | dependencies/catchorg__Catch2-mc69e0473f6e98/INDEX.json | evidence/backfill-20261012/catchorg__Catch2-mc69e0473f6e98/SUMMARY.json | 1 | 69e0473f6e98d47c93518424c08ee69ee632c0f0 | 0 | GLM-5.3-Flash-A | 1 | 97b1bec0-41f0-43a6-8e67-0b77b4bd18da | bundles/9dd4d09ddeb0bfd83be9822d63f31afac7c2c331a790ba456de899186778ea94/delivery.tar.gz | Zeng-Weijun/taskforge-rack-harbor-v2 | dataset | 386d389810111b83a9e1dda7c23243fd79bb87f8 | sha256:3518e43e96bdda89589aa98f8ab368d96c38d1bd4cc7af03348cc1b2606516aa | true | # Underflow when iterating backward over a string starting with a UTF-8 continuation byte
## Scenario
A user passes text containing invalid UTF-8 (e.g., a string whose first byte is a UTF-8 continuation byte) into Catch2's text-flow machinery. When the machinery iterates backward over the string (for example, while sk... | instructions/catchorg__Catch2-mc69e0473f6e98.md | ab98fd5e7029630949e8c419e61289a378d7920fcd2b071b929fb60acc66025a | cpp | 2 | b1178e9b5385e25eece64930eabad574f1ccac10c0840e1bce95b449e199ce54 | original-v2 | 本批 7 题共有的评分包装器缺陷已独立复现:判分器冻结隐藏测试但未冻结装载链(Ruby test/helper.rb、pytest conftest.py、顶层 CMakeLists.txt),可不修目标缺陷取得 reward=1.0;当前不推荐用作评分。详见 QUALITY-NOTES.md。 | load_path_defect_reproduced | taskforge-completed-cli-preview.v2 | Published complete task and recorded outcomes; metadata backfill performs no new test or model run | catchorg/Catch2 | https://github.com/catchorg/Catch2 | 修复以 UTF-8 continuation byte 开头的字符串在反向迭代时越过起点的问题。 | catchorg__Catch2-mc69e0473f6e98 | Underflow when iterating backward over a string starting with a UTF-8 continuation byte |
2,040,542,852 | b4d752c9931e88fd8a24467f1806112a477a5f7d08dcafbf89de463a9382c211 | https://huggingface.co/datasets/Zeng-Weijun/taskforge-rack-harbor-v2/resolve/386d389810111b83a9e1dda7c23243fd79bb87f8/bundles/b4d752c9931e88fd8a24467f1806112a477a5f7d08dcafbf89de463a9382c211/delivery.tar.gz | 22b02e26f9e8c7e32bd8266e2b0ebe8bb3a0db2b | sha256:6fe89a270a30df97bd8f6d519e4a88caea765c45299bdc0e4ebe8a3317bdc655 | claude-code-cli | true | 0 | DeepSeek-V4-Flash | 1 | 1613e0dd-9952-46a5-a74b-d92f333178d3 | dependencies/fastapi__fastapi-mc063b5bf582d3/INDEX.json | evidence/backfill-20261012/fastapi__fastapi-mc063b5bf582d3/SUMMARY.json | 2 | 063b5bf582d31fb155cc6bc6f88cf512329d0fd5 | 0 | GLM-5.3-Flash-A | 1 | cca9c571-fa1c-464a-911e-af85197f0a3e | bundles/b4d752c9931e88fd8a24467f1806112a477a5f7d08dcafbf89de463a9382c211/delivery.tar.gz | Zeng-Weijun/taskforge-rack-harbor-v2 | dataset | 386d389810111b83a9e1dda7c23243fd79bb87f8 | sha256:a329435ed1f8429ed189c84b96a3afa59531f4b508b146d2cfeb972bdbdb7b9e | true | # Underscore headers are incorrectly accepted when convert_underscores=True is default
## Scenario
A client sends an HTTP header with underscores (e.g., `x_user_id`) to an endpoint that uses a Pydantic model with `Header()` and the default `convert_underscores=True`. Both the underscored and hyphenated versions of the... | instructions/fastapi__fastapi-mc063b5bf582d3.md | 919499d58c0640fc7dcbb79aa90b72efd838b222101d4f44fea9e9375d0baf8e | python | 128 | dda823d7ba6f9e5502cce771389be7e300ba34245809217079ab278bba2416da | cli-recovery-001 | 本批 7 题共有的评分包装器缺陷已独立复现:判分器冻结隐藏测试但未冻结装载链(Ruby test/helper.rb、pytest conftest.py、顶层 CMakeLists.txt),可不修目标缺陷取得 reward=1.0;当前不推荐用作评分。详见 QUALITY-NOTES.md。 | load_path_defect_reproduced | taskforge-completed-cli-preview.v2 | Published complete task and recorded outcomes; metadata backfill performs no new test or model run | fastapi/fastapi | https://github.com/fastapi/fastapi | 修复 Header 模型默认下划线转换:只绑定连字符头名,单独下划线头应返回422。 | fastapi__fastapi-mc063b5bf582d3 | Underscore headers are incorrectly accepted when convert_underscores=True is default |
2,040,781,634 | c071e4eff54cb3975709d41b79581d60d6e0d14160aedcc7027e4c4993562c20 | https://huggingface.co/datasets/Zeng-Weijun/taskforge-rack-harbor-v2/resolve/386d389810111b83a9e1dda7c23243fd79bb87f8/bundles/c071e4eff54cb3975709d41b79581d60d6e0d14160aedcc7027e4c4993562c20/delivery.tar.gz | ef1c927b0558d414e199a666833942a6fabb3a51 | sha256:d7780d8161fa590f7e0a936ad9807e7fcc1c0d84ea6de4768db7716f0ed4da62 | claude-code-cli | true | 0 | DeepSeek-V4-Flash | 1 | a9058b38-8dec-463a-8de1-ff97346251af | dependencies/fastapi__fastapi-mc7815a32f2ed1/INDEX.json | evidence/backfill-20261012/fastapi__fastapi-mc7815a32f2ed1/SUMMARY.json | 1 | 7815a32f2ed177b8b786a48b3e0712c05b5c644f | 0 | GLM-5.3-Flash-A | 1 | f3df6599-397e-4680-88d9-18acacb54897 | bundles/c071e4eff54cb3975709d41b79581d60d6e0d14160aedcc7027e4c4993562c20/delivery.tar.gz | Zeng-Weijun/taskforge-rack-harbor-v2 | dataset | 386d389810111b83a9e1dda7c23243fd79bb87f8 | sha256:6f65a3dbdbbf88cce45fa692595e3313ec1fc3b9095dee7116f720b0aa8427ca | true | # Support pydantic_extra_types.color.Color in jsonable_encoder and update forward reference
## Scenario
A developer uses the Color type from pydantic_extra_types (the replacement for the deprecated pydantic.color.Color) in data passed through FastAPI's jsonable_encoder, for example when returning a model containing a ... | instructions/fastapi__fastapi-mc7815a32f2ed1.md | f17ed55ee5b017f8c11dfc12673118647483dc1f8943024e552406e1ed245167 | python | 9 | a521b1eeadde6216b893d19f1241dc949c3f102d48d8ce20cdcd1608cc870b66 | cli-recovery-001 | 本批 7 题共有的评分包装器缺陷已独立复现:判分器冻结隐藏测试但未冻结装载链(Ruby test/helper.rb、pytest conftest.py、顶层 CMakeLists.txt),可不修目标缺陷取得 reward=1.0;当前不推荐用作评分。详见 QUALITY-NOTES.md。 | load_path_defect_reproduced | taskforge-completed-cli-preview.v2 | Published complete task and recorded outcomes; metadata backfill performs no new test or model run | fastapi/fastapi | https://github.com/fastapi/fastapi | 支持 pydantic_extra_types.color.Color 的 JSON 编码,并更新前向引用兼容逻辑。 | fastapi__fastapi-mc7815a32f2ed1 | Support pydantic_extra_types.color.Color in jsonable_encoder and update forward reference |
2,038,096,380 | 1cc7bc4f15d1dc98872365d49854fef2d0cb0709b9c1d883886349fa541f975d | https://huggingface.co/datasets/Zeng-Weijun/taskforge-rack-harbor-v2/resolve/386d389810111b83a9e1dda7c23243fd79bb87f8/bundles/1cc7bc4f15d1dc98872365d49854fef2d0cb0709b9c1d883886349fa541f975d/delivery.tar.gz | 9653034b8d459607a4b7f761e6c23a6f287fa5fa | sha256:13ed72717a6a559bae290b166e95d9409d0bf8103977b6f53f4261da9f6c714c | claude-code-cli | true | 0 | DeepSeek-V4-Flash | 1 | 9fd746f4-0413-48a2-bd00-afc6cbcc5f3b | dependencies/fastapi__fastapi-mcae4e45c5cc20/INDEX.json | evidence/backfill-20261012/fastapi__fastapi-mcae4e45c5cc20/SUMMARY.json | 1 | ae4e45c5cc20a1e1503fbcab2369821d188feb09 | 0 | GLM-5.3-Flash-A | 1 | d4a6abfb-7102-4575-85ef-8fd2c989bf91 | bundles/1cc7bc4f15d1dc98872365d49854fef2d0cb0709b9c1d883886349fa541f975d/delivery.tar.gz | Zeng-Weijun/taskforge-rack-harbor-v2 | dataset | 386d389810111b83a9e1dda7c23243fd79bb87f8 | sha256:a100e259f9e1ccf4228263d2adfc91a0928f02b388d56f1d2f0fe251515bdf07 | true | # Remove April Fool's @app.vibe() decorator
## Scenario
A FastAPI user decorates a path operation with @app.vibe(...) or inspects the FastAPI class's public API surface (e.g., checking if a `vibe` attribute exists on the app instance). This decorator was an April Fool's joke that temporarily existed in the codebase.
... | instructions/fastapi__fastapi-mcae4e45c5cc20.md | 85190b18233d56203740ca5a22a3456ad983010e25ee15929e6f824e9b86f36c | python | 4 | 44e61b1c18e2f00ec26989a19bdf5612e1ecdf95a06f74713b171ff42f851a50 | original-v2 | 本批 7 题共有的评分包装器缺陷已独立复现:判分器冻结隐藏测试但未冻结装载链(Ruby test/helper.rb、pytest conftest.py、顶层 CMakeLists.txt),可不修目标缺陷取得 reward=1.0;当前不推荐用作评分。详见 QUALITY-NOTES.md。 | load_path_defect_reproduced | taskforge-completed-cli-preview.v2 | Published complete task and recorded outcomes; metadata backfill performs no new test or model run | fastapi/fastapi | https://github.com/fastapi/fastapi | 移除愚人节的 app.vibe() 装饰器及其示例与测试。 | fastapi__fastapi-mcae4e45c5cc20 | Remove April Fool's @app.vibe() decorator |
542,872,276 | 127f6821b388b421dd22fbb9d38560e57c69081edfe6cf3cc8c67a1ad0b84a2b | https://huggingface.co/datasets/Zeng-Weijun/taskforge-rack-harbor-v2/resolve/26394dce6809a4c8d694120d67d49cba6c015f57/bundles/127f6821b388b421dd22fbb9d38560e57c69081edfe6cf3cc8c67a1ad0b84a2b/delivery.tar.gz | 68fbb71d88e22c138d964bb8d06fd4ac299da5c4 | sha256:aecf6f9d5d7f8a39ad1becd12bfb2ec7c22052886f9995c64e41f7432fb64521 | claude-code-cli | true | 0 | DeepSeek-V4-Flash | 1 | be754167-ca03-43bc-9189-abecb5ebd8d6 | dependencies/INDEX.json | reviews/VERSION_CHAIN.json | 1 | 3628cea311718b9f2350bd641abfb0f8c5c7292e | 0 | GLM-5.3-Flash-A | 1 | e809e5f8-819e-4320-acce-e46da1fd49b4 | bundles/127f6821b388b421dd22fbb9d38560e57c69081edfe6cf3cc8c67a1ad0b84a2b/delivery.tar.gz | Zeng-Weijun/taskforge-rack-harbor-v2 | dataset | 26394dce6809a4c8d694120d67d49cba6c015f57 | sha256:2bcc74356180187480200f5ff5f867092a19ca7738c04e65cfb4cd2f32720106 | true | # Fix MIME type for .pem files
## Scenario
An application uses Rack to determine the MIME type of a .pem file, for example when serving assets or classifying file extensions through Rack::Mime's public lookup method.
## Observed behaviour
Rack currently returns 'application/x-x509-ca-cert' for the .pem extension, whi... | instructions/pem.md | cb72ea119447d97e0f21fe8fd25cf7a3971f660ce26d56a5290c11271ab34eec | ruby | 8 | 5f0aac6c578c60192108208ce7a472313a5084a01c469b1abe04dadf82e99c0a | original-v2 | 本批 7 题共有的评分包装器缺陷已独立复现:判分器冻结隐藏测试但未冻结装载链(Ruby test/helper.rb、pytest conftest.py、顶层 CMakeLists.txt),可不修目标缺陷取得 reward=1.0;当前不推荐用作评分。详见 QUALITY-NOTES.md。 | load_path_defect_reproduced | taskforge-completed-cli-preview.v2 | Published complete task and recorded outcomes; metadata backfill performs no new test or model run | rack/rack | https://github.com/rack/rack | 将 .pem 文件的 MIME 类型改为 application/x-pem-file。 | rack__rack-mc3628cea31171 | Fix MIME type for .pem files |
543,970,650 | 435e240b6237de818f6c3b14ad8042df027658f990a7a0015a19f30783eafb4f | https://huggingface.co/datasets/Zeng-Weijun/taskforge-rack-harbor-v2/resolve/386d389810111b83a9e1dda7c23243fd79bb87f8/bundles/435e240b6237de818f6c3b14ad8042df027658f990a7a0015a19f30783eafb4f/delivery.tar.gz | 939f2f81b635331f50b3c933f150c8152aa2950f | sha256:2fcf9ddc86a4b8e413329ee1c7152753afdfba29a1ce1c241b472019a5eba991 | claude-code-cli | true | 0 | DeepSeek-V4-Flash | 1 | 4005d003-1e5a-4ea3-a801-38cb44355c89 | dependencies/rack__rack-mc83fff58206e6/INDEX.json | evidence/backfill-20261012/rack__rack-mc83fff58206e6/SUMMARY.json | 2 | 83fff58206e6fec46815322e0783b6d7cb50f979 | 0 | GLM-5.3-Flash-A | 1 | 1f640539-2b6d-4b20-ac7b-7ccbfc49ff0b | bundles/435e240b6237de818f6c3b14ad8042df027658f990a7a0015a19f30783eafb4f/delivery.tar.gz | Zeng-Weijun/taskforge-rack-harbor-v2 | dataset | 386d389810111b83a9e1dda7c23243fd79bb87f8 | sha256:61341eb788e89dd02d513e211e593a504ac34334f8cdd80c7325bd1ca3f42e7d | true | # Multipart parser crashes on missing semicolon or equals in header parameters
## Scenario
A client sends a multipart/form-data request where one part's Content-Disposition header has no parameters (missing semicolon), e.g. `Content-Disposition: form-data` with no parameters, or another part's Content-Type header cont... | instructions/rack__rack-mc83fff58206e6.md | 7024e05ab518bd9bfc30a77e25f847605a1e9cb2bcff401df4b067c9ce3b3fa2 | ruby | 30 | c7fe15adeddefa22f9f1c6ddfc5aa870eb74d4fafd0ec5e1bb57160530fcbe48 | original-v2 | 本批 7 题共有的评分包装器缺陷已独立复现:判分器冻结隐藏测试但未冻结装载链(Ruby test/helper.rb、pytest conftest.py、顶层 CMakeLists.txt),可不修目标缺陷取得 reward=1.0;当前不推荐用作评分。详见 QUALITY-NOTES.md。 | load_path_defect_reproduced | taskforge-completed-cli-preview.v2 | Published complete task and recorded outcomes; metadata backfill performs no new test or model run | rack/rack | https://github.com/rack/rack | 修复 multipart 请求头参数缺少分号或等号时触发的 NoMethodError。 | rack__rack-mc83fff58206e6 | Multipart parser crashes on missing semicolon or equals in header parameters |
542,907,270 | 04a7ffbcb53e41e5a8a307a2f5bfec667ff1dddbf213ba266e2401d259284c06 | https://huggingface.co/datasets/Zeng-Weijun/taskforge-rack-harbor-v2/resolve/26394dce6809a4c8d694120d67d49cba6c015f57/bundles/04a7ffbcb53e41e5a8a307a2f5bfec667ff1dddbf213ba266e2401d259284c06/delivery.tar.gz | d3fd39964156cb34fdf0f35f57c10bda0b66dc2e | sha256:c0a765e666591dd27a93f981f6998faf562f3601f857703f69635ce98ae67a0c | claude-code-cli | true | 0 | DeepSeek-V4-Flash | 1 | 522a62c1-2851-4157-ae7a-092c71e2516c | dependencies/INDEX.json | reviews/VERSION_CHAIN.json | 1 | 8979a1689b65a02f0363fe81e7290ae5d6c45724 | 0 | GLM-5.3-Flash-A | 1 | bcba13b1-7916-460a-9073-3f0a8f5da5b7 | bundles/04a7ffbcb53e41e5a8a307a2f5bfec667ff1dddbf213ba266e2401d259284c06/delivery.tar.gz | Zeng-Weijun/taskforge-rack-harbor-v2 | dataset | 26394dce6809a4c8d694120d67d49cba6c015f57 | sha256:916514ea4e7502eb04ed4b743cd44972df27376cf0fec7247f7fb0dc83050fad | true | # Fix Content-Length header in 404 responses for paths with multi-byte characters
## Scenario
A client requests a static file through Rack::Files using a URL whose decoded path contains multi-byte UTF-8 characters (e.g., `/cgi/caf%C3%A9` decoding to 'café'). The file does not exist, so the server generates a 404 error... | instructions/404.md | 0f9569549e8953568f5d3911a8913b4e3e028539dba3dd74a54cc5dd5387d566 | ruby | 33 | 3eb835f3051fc6879ecee669ad5a7b0a5a6a17a3d0c7c7669104ac9afbfdd934 | original-v2 | 本批 7 题共有的评分包装器缺陷已独立复现:判分器冻结隐藏测试但未冻结装载链(Ruby test/helper.rb、pytest conftest.py、顶层 CMakeLists.txt),可不修目标缺陷取得 reward=1.0;当前不推荐用作评分。详见 QUALITY-NOTES.md。 | load_path_defect_reproduced | taskforge-completed-cli-preview.v2 | Published complete task and recorded outcomes; metadata backfill performs no new test or model run | rack/rack | https://github.com/rack/rack | 修复多字节路径触发 404 时 Content-Length 按字符数而非字节数计算的问题。 | rack__rack-mc8979a1689b65 | Fix Content-Length header in 404 responses for paths with multi-byte characters |
7 个完整 Harbor Task 与 14 条双 Flash Claude Code CLI 轨迹
当前包含 7 题、3 个仓库:Rack 3、FastAPI 3、Catch2 1。全部来自 Claude Code CLI(2.1.296)在真实 Harbor 环境中的解题轨迹,每题两条:DeepSeek-V4-Flash 与 GLM-5.3-Flash-A 各一条。
质量提示(2026-10-13):本批 7 题的评分包装器存在同一类缺陷——判分器冻结了隐藏测试,但未冻结测试的装载链(Ruby
test/helper.rb、pytestconftest.py、顶层CMakeLists.txt),因此存在不修复目标缺陷即可拿到 reward=1.0 的路径。独立复现已覆盖三种语言。当前不推荐将本批用作评分或 RL reward 来源,详见 QUALITY-NOTES。原包、原题面与历史成绩均按原字节保留,未做任何改写。
本仓库只收录 Claude Code CLI 交付。 早期由自研 Bash agent 产出的 8 个包已于 2026-10-13 移除;其字节仍保存在原始对象存储中,需要审计可另行索取。保留的 7 个包字节与移除前完全一致。
七题预览 使用 v2 扁平 schema,默认 dataset 配置为 completed_cli_deliveries_v2。预览包含原题面、固定的 base/fix commit、包 SHA、镜像 ID、依赖入口与两条 CLI 原始回执摘要。
| Task | 仓库 | 语言 | 题面摘要 | DeepSeek / GLM reward |
|---|---|---|---|---|
catchorg__Catch2-mc69e0473f6e98 |
catchorg/Catch2 | cpp | 修复以 UTF-8 continuation byte 开头的字符串在反向迭代时越过起点的问题。 | 1 / 1 |
fastapi__fastapi-mc063b5bf582d3 |
fastapi/fastapi | python | 修复 Header 模型默认下划线转换:只绑定连字符头名,单独下划线头应返回 422。 | 1 / 1 |
fastapi__fastapi-mc7815a32f2ed1 |
fastapi/fastapi | python | 支持 pydantic_extra_types.color.Color 的 JSON 编码,并更新前向引用兼容逻辑。 |
1 / 1 |
fastapi__fastapi-mcae4e45c5cc20 |
fastapi/fastapi | python | 移除愚人节的 app.vibe() 装饰器及其示例与测试。 |
1 / 1 |
rack__rack-mc3628cea31171 |
rack/rack | ruby | 将 .pem 文件的 MIME 类型改为 application/x-pem-file。 |
1 / 1 |
rack__rack-mc83fff58206e6 |
rack/rack | ruby | 修复 multipart 请求头参数缺少分号或等号时触发的 NoMethodError。 |
1 / 1 |
rack__rack-mc8979a1689b65 |
rack/rack | ruby | 修复多字节路径触发 404 时 Content-Length 按字符数而非字节数计算的问题。 |
1 / 1 |
每个包包含:完整 Linux/amd64 的 Docker-save 镜像、Harbor task 定义、两个模型各自的原始 Claude Code CLI 轨迹、独立验收与 Harbor 验收记录、runtime 与默认验收入口。
上述 14 个成绩均为已封存的正式回执,本次整理未重新判断任务语义,也未重跑真实仓库测试套件。FastAPI 两个包保留各自独立版本与 lineage:mc7815a32f2ed1 的 GLM 轨迹为原成功轨迹复用、DeepSeek 为恢复新跑;mc063b5bf582d3 两条均为恢复新跑,原失败证据未被改写。
下载与验收
请按 QUICKSTART 下载。资产位于 bundles/<archive SHA256>/delivery.tar.gz;ASSETS.json 固定每个包的完整 SHA256、字节数与资产 commit,DOWNLOAD_COMMANDS.json 提供全部下载命令与质量标记。
每个包解包后自带验收入口:
sha256sum -c SHA256SUMS
docker load -i image.tar
python3 verify.py --task task --image-id <manifest 中的 image_id> --output verification
harbor run -p task -a nop # 期望 reward 0.0
harbor run -p task -a oracle # 期望 reward 1.0
难度
7 题均为真实上游 bug 的精确小补丁,按改动规模与实测工具调用数评估为偏易:
| Task | gold patch | 实测工具调用(DS / GLM) |
|---|---|---|
fastapi__fastapi-mc7815a32f2ed1 |
4 文件 +37/-7 | 54 / 70 |
fastapi__fastapi-mc063b5bf582d3 |
1 文件 +4/-0 | 48 / 33 |
catchorg__Catch2-mc69e0473f6e98 |
1 文件 +2/-2 | 35 / 30 |
rack__rack-mc83fff58206e6 |
1 文件 +6/-2 | 27 / 25 |
fastapi__fastapi-mcae4e45c5cc20 |
2 文件 +1/-71 | 21 / 13 |
rack__rack-mc8979a1689b65 |
1 文件 +1/-1 | 14 / 12 |
rack__rack-mc3628cea31171 |
1 文件 +1/-1 | 8 / 6 |
14 条轨迹全部 reward=1.0,两个模型均一次通过;工具调用数的跨度主要反映仓库体积与依赖摩擦,而非概念难度。本批区分度有限,适合作管线冒烟与教学样例,不适合作能力分层。
质量与已知问题
- QUALITY-NOTES.md:本批 7 题的评分包装器缺陷说明与当前状态。
- KNOWN_ISSUES.md:题面与测试前提的已知偏差(PEM 题面措辞、404 encoding 前提、reporter 错误压缩、环境权限等)。
- 其余审阅证据见 reviews/ 与 quality/。
来源与证据
证据索引 提供原题面、package/plan、原始 CLI 回执、独立与 Harbor 验收记录及来源 SHA。完整原始轨迹存放在各自包内,本 metadata 不重复导出大型 raw trace。base/fix 来自封存 source-selection 或受保护的原 parent-ready 记录。
依赖入口提供原交付源快照中的配置或锁文件字节及 snapshot SHA。FastAPI 的 pyproject.toml 与 Catch2 的 CMake 配置不能当作完整已安装依赖锁;完整运行环境在各自的 image.tar 内。题面为原正式题面,中文摘要不替换或修订原题面。
English
7 Completed Harbor Tasks with 14 Dual-Flash Claude Code CLI Rollouts
7 tasks across 3 repositories (Rack ×3, FastAPI ×3, Catch2 ×1). Every rollout is a real
Claude Code CLI (2.1.296) session in a real Harbor environment — one per model, per task:
DeepSeek-V4-Flash and GLM-5.3-Flash-A.
Quality notice (2026-10-13): the grading wrapper shipped with these 7 tasks shares a single defect class — the verifier freezes the hidden tests but not their load path (Ruby
test/helper.rb, pytestconftest.py, top-levelCMakeLists.txt). A reward of 1.0 is therefore reachable without fixing the target bug. Independently reproduced in all three languages. This batch is not recommended as a scoring or RL-reward source in its current form. See QUALITY-NOTES. Original bundles, statements and historical scores are kept byte-for-byte; nothing has been rewritten.
This repository now carries Claude Code CLI deliveries only. The 8 earlier bundles produced by a custom Bash agent were removed on 2026-10-13; their bytes remain in the original object storage and can be supplied for audit. The 7 retained bundles are byte-identical to before the removal.
Each bundle contains the full Linux/amd64 Docker-save image, the Harbor task, both models' raw CLI
trajectories, independent and Harbor acceptance records, and the runtime plus default acceptance
entrypoint. Assets live at bundles/<archive SHA256>/delivery.tar.gz; ASSETS.json
pins each archive's SHA256, byte size and asset commit.
The gold patch for all 7 tasks is a small, precise fix to a real upstream bug; measured tool-call counts range from 6 to 70 and all 14 rollouts scored 1.0. The spread reflects repository size and dependency friction rather than conceptual difficulty, so treat this batch as a pipeline smoke test and teaching sample rather than a capability benchmark.
See QUICKSTART.md to download, and KNOWN_ISSUES.md for statement-level caveats that ship with the original packages.
- Downloads last month
- 52