反馈自动修复运行
流程中断，未产出补丁。
演练运行运行失败
总耗时
5m 52s
2026-08-11 21:19
Token
24,015
3 次模型调用
代码改动
无补丁
工具调用
7
沙箱与外部操作
- RUN REF
- ad5e12d9d045
- 类别
- docx_structure
- Provider
- openai_compatible
- 模型
- deepseek-ai/DeepSeek-V4-Flash
- Graph
- agent-graph-v3
- Policy
- repair-policy-v2
- 扩展版本
- 0.3.0
- 基线 commit
- 89c9943f59c2
阶段 4 / 7
沙箱复现5m 31s重试 2 次
在隔离沙箱里生成一个会失败的测试，用来证明缺陷真实存在。断言只能从已登记的 Oracle 列表中挑选，模型不能提交可执行表达式。测试在基线上必须以预期方式失败，否则不算复现。
本阶段调用14
deepseek-ai/DeepSeek-V4-Flash
输入 686 · 输出 3,062 tokens
- messages
- [{"role":"system","bytes":1735,"sha256":"a5103335efe46486f248bc53f705fc580c4a3ff727f2ab7cdc976ad3dbe9db40"},{"role":"user","bytes":722,"sha256":"84ec6c1231767fc58b2a216a19f144e457fba63aead6d1f513bf679a09243b66"}]
- tool_count
- 0
- response_schema
- ReproductionPlan
- oracle
- {"kind":"docx_xpath","parameters":{"parameter_names":["minimum","style","text","validator"]}}
- hypothesis
- [REDACTED_SUMMARY]
- files_to_read
- [{"path":"backend/app/pandoc_runner.py","end_line":80,"start_line":1},{"path":"backend/app/normalizer.py","end_line":80,"start_line":1},{"path":"backend/tests/test_pandoc_runner.py","end_line":100,"start_line":1},{"path":"backend/tests/test_normalizer.py","end_line":100,"start_line":1},{"path":"backend/tests/test_convert_api.py","end_line":100,"start_line":1}]
- target_test_selector
- test_feedback_baabbc0e_mermaid_not_rendered
- expected_failure_kind
- assertion
- extension_sync_possible
- false
- path
- backend/app/pandoc_runner.py
- end_line
- 80
- start_line
- 1
- path
- backend/app/pandoc_runner.py
- end_line
- 80
- start_line
- 1
- content_bytes
- 2995
- path
- backend/app/normalizer.py
- end_line
- 80
- start_line
- 1
- path
- backend/app/normalizer.py
- end_line
- 80
- start_line
- 1
- content_bytes
- 2738
- path
- backend/tests/test_pandoc_runner.py
- end_line
- 100
- start_line
- 1
- path
- backend/tests/test_pandoc_runner.py
- end_line
- 100
- start_line
- 1
- content_bytes
- 3775
- path
- backend/tests/test_normalizer.py
- end_line
- 100
- start_line
- 1
- path
- backend/tests/test_normalizer.py
- end_line
- 100
- start_line
- 1
- content_bytes
- 2947
- path
- backend/tests/test_convert_api.py
- end_line
- 100
- start_line
- 1
- path
- backend/tests/test_convert_api.py
- end_line
- 92
- start_line
- 1
- content_bytes
- 2446
deepseek-ai/DeepSeek-V4-Flash
输入 6,157 · 输出 12,865 tokens
- messages
- [{"role":"system","bytes":1402,"sha256":"0ff456bc4a7eee6deafbff6e663aaae[REDACTED_PHONE]aea894736b0b422f346db0"},{"role":"user","bytes":18100,"sha256":"831935a6be4be0a23e9056ddb1c00590d66451cf1620cdded8a7c9b2e4276754"}]
- tool_count
- 0
- response_schema
- TestGenerationResult
- edits
- [{"mode":"full_file","path":"backend/tests/fixtures/feedback/mermaid_graph.md"},{"mode":"search_replace","path":"backend/tests/test_feedback_regressions.py"}]
- oracle
- {"kind":"docx_xpath","parameters":{"parameter_names":["minimum","style","text","validator"]}}
- reason
- [REDACTED_SUMMARY]
- files_needed_for_fix
- []
- target_test_selector
- test_feedback_baabbc0e_mermaid_not_rendered
- expected_failure_kind
- assertion
- extension_sync_required
- false
source_access_denied
- edit_count
- 2
- error_code
- source_access_denied
- error_type
- SourceAccessError
- path
- backend/app/pandoc_runner.py
- end_line
- 80
- start_line
- 1
- path
- backend/app/pandoc_runner.py
- end_line
- 80
- start_line
- 1
- content_bytes
- 2995
- path
- backend/app/normalizer.py
- end_line
- 80
- start_line
- 1
- path
- backend/app/normalizer.py
- end_line
- 80
- start_line
- 1
- content_bytes
- 2738
- path
- backend/tests/test_pandoc_runner.py
- end_line
- 100
- start_line
- 1
- path
- backend/tests/test_pandoc_runner.py
- end_line
- 100
- start_line
- 1
- content_bytes
- 3775
- path
- backend/tests/test_normalizer.py
- end_line
- 100
- start_line
- 1
- path
- backend/tests/test_normalizer.py
- end_line
- 100
- start_line
- 1
- content_bytes
- 2947
- path
- backend/tests/test_convert_api.py
- end_line
- 100
- start_line
- 1
- path
- backend/tests/test_convert_api.py
- end_line
- 92
- start_line
- 1
- content_bytes
- 2446
已收到模型响应，但未通过严格 Schema 或本地策略校验。
deepseek-ai/DeepSeek-V4-Flash
- messages
- [{"role":"system","bytes":1402,"sha256":"0ff456bc4a7eee6deafbff6e663aaae[REDACTED_PHONE]aea894736b0b422f346db0"},{"role":"user","bytes":18355,"sha256":"f727e56a11c6bde34f9ba19be7b1516d8ce0124a95e3c1e72beeb92aa95c10ec"}]
- tool_count
- 0
- response_schema
- TestGenerationResult
- error_code
- invalid_response
- error_type
- InvalidModelResponseError
独立验证
该运行未进入验证阶段。
已收到模型响应，但未通过严格 Schema 或本地策略校验。
补丁策略检查
patch-policy-v2
该运行未产生补丁。