Add 9 curator ground truths + harden runner timeouts (148/160 SDLC)

New GT: qutebrowser-darkmode, qutebrowser-url, bustub-hyperloglog,
tensorrt-mxfp4, element-web-unread, django-repo-scoped, grpcurl-vuln,
calcom-code-review, kafka-build-orient.

Runner fixes:
- Python-level thread timeout on exec_cmd (shutdown(wait=False))
- OS-level timeout wrapper on git clone and curator CLI
- kernel.org → github.com/torvalds/linux mirror for curator clones
- Skip 4 linux kernel tasks (repo too large for Daytona sandbox)
- Bump SANDBOX_TIMEOUT_SEC 900→1500, SIGALRM 840→1440

12 tasks remain: 4 kernel (need manual GT), 8 curator timeout
(need parallel=1 or dedicated account).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: sjarmak

benchmarks/csb_sdlc_debug/qutebrowser-darkmode-threshold-regression-prove-001/tests/ground_truth_agent.json

@@ -0,0 +1,49 @@
+{
+  "files": [
+    "qutebrowser/browser/webengine/darkmode.py",
+    "tests/unit/browser/webengine/test_darkmode.py",
+    "qutebrowser/utils/version.py"
+  ],
+  "symbols": [
+    {
+      "file": "qutebrowser/browser/webengine/darkmode.py",
+      "symbol": "Variant",
+      "repo": null
+    },
+    {
+      "file": "qutebrowser/browser/webengine/darkmode.py",
+      "symbol": "_variant",
+      "repo": null
+    },
+    {
+      "file": "qutebrowser/browser/webengine/darkmode.py",
+      "symbol": "_DEFINITIONS",
+      "repo": null
+    },
+    {
+      "file": "qutebrowser/browser/webengine/darkmode.py",
+      "symbol": "_Definition",
+      "repo": null
+    },
+    {
+      "file": "qutebrowser/browser/webengine/darkmode.py",
+      "symbol": "_Setting",
+      "repo": null
+    },
+    {
+      "file": "qutebrowser/browser/webengine/darkmode.py",
+      "symbol": "copy_replace_setting",
+      "repo": null
+    },
+    {
+      "file": "qutebrowser/browser/webengine/darkmode.py",
+      "symbol": "settings",
+      "repo": null
+    },
+    {
+      "file": "qutebrowser/utils/version.py",
+      "symbol": "WebEngineVersions",
+      "repo": null
+    }
+  ]
+}

benchmarks/csb_sdlc_debug/qutebrowser-darkmode-threshold-regression-prove-001/tests/ground_truth_meta.json

@@ -7,12 +7,12 @@
   "curator_agent_version": "2.0",
   "model": "claude-opus-4-6",
   "backend": "hybrid",
-  "timestamp": "2026-03-03T21:34:52Z",
-  "files_count": 4,
+  "timestamp": "2026-03-06T14:49:26Z",
+  "files_count": 3,
   "edit_files_count": 0,
   "chunks_count": 0,
-  "symbols_count": 9,
-  "cost_usd": 3.1905410000000005,
-  "elapsed_sec": 625.6,
-  "exploration_notes": "The bug is in qutebrowser/browser/webengine/darkmode.py. Qt 6.4's Chromium renamed 'TextBrightnessThreshold' to 'ForegroundBrightnessThreshold'. The fix (commit 50efac08f) added: (1) a Variant.qt_64 enum member, (2) a copy_replace_setting method on _Definition, (3) a Qt 6.4 entry in _DEFINITIONS that maps threshold.text to ForegroundBrightnessThreshold, and (4) version detection in _variant() for Qt >= 6.4. Before the fix, Qt 6.4 was treated as Qt 6.3 and used the old TextBrightnessThreshold key"
+  "symbols_count": 8,
+  "cost_usd": 1.6713492500000002,
+  "elapsed_sec": 361.8,
+  "exploration_notes": "The bug is in darkmode.py. Before the fix (commit 50efac08f), there was no Variant.qt_64 enum value. The _variant() function mapped Qt 6.4 to Variant.qt_63 (since 6.4 >= 6.3). The Qt 6.3 definition uses 'TextBrightnessThreshold' as the Chromium key for the threshold.text setting. However, Qt 6.4's Chromium (99.0.4785.0) renamed this to 'ForegroundBrightnessThreshold'. Since the old key name is not recognized by Qt 6.4's Chromium, the threshold.text setting was silently ignored. The fix adds a ne"
 }

benchmarks/csb_sdlc_debug/qutebrowser-url-regression-prove-001/tests/ground_truth_agent.json

@@ -0,0 +1,29 @@
+{
+  "files": [
+    "qutebrowser/utils/urlutils.py",
+    "tests/unit/utils/test_urlutils.py",
+    "qutebrowser/browser/navigate.py"
+  ],
+  "symbols": [
+    {
+      "file": "qutebrowser/utils/urlutils.py",
+      "symbol": "incdec_number",
+      "repo": null
+    },
+    {
+      "file": "qutebrowser/utils/urlutils.py",
+      "symbol": "_get_incdec_value",
+      "repo": null
+    },
+    {
+      "file": "qutebrowser/utils/urlutils.py",
+      "symbol": "_URL_SEGMENTS",
+      "repo": null
+    },
+    {
+      "file": "qutebrowser/utils/urlutils.py",
+      "symbol": "IncDecError",
+      "repo": null
+    }
+  ]
+}

benchmarks/csb_sdlc_debug/qutebrowser-url-regression-prove-001/tests/ground_truth_meta.json

@@ -7,12 +7,12 @@
   "curator_agent_version": "2.0",
   "model": "claude-opus-4-6",
   "backend": "hybrid",
-  "timestamp": "2026-03-03T18:06:21Z",
+  "timestamp": "2026-03-06T16:00:54Z",
   "files_count": 3,
   "edit_files_count": 0,
   "chunks_count": 0,
   "symbols_count": 4,
-  "cost_usd": 2.5924995,
-  "elapsed_sec": 592.7,
-  "exploration_notes": "The root cause is in urlutils.py lines 559-624. The _URL_SEGMENTS table uses QUrl.PrettyDecoded for getters and QUrl.StrictMode for setters. PrettyDecoded decodes percent-encoded characters that don't need encoding in their context (e.g., %20\u2192space). When the modified string is written back via StrictMode, the decoded characters (like literal spaces) make the URL invalid. The regression test at /workspace/regression_test.py demonstrates: (1) %20 in path/query causes URLs to become invalid after "
+  "cost_usd": 0.97326375,
+  "elapsed_sec": 325.9,
+  "exploration_notes": "The bugs are in qutebrowser/utils/urlutils.py. The regex at line 618 (r'(.*\\D|^)(?<!%)(?<!%.)(0*)(\\d+)(.*)') has an over-aggressive lookbehind that blocks legitimate digits after percent-encoded sequences (e.g., '5' in '%235' which is '#5'). Additionally, the _URL_SEGMENTS getters at lines 559-580 use QUrl.PrettyDecoded which decodes certain percent-encoded chars (like %3A\u2192':'), and the setters use QUrl.StrictMode which doesn't re-encode them, causing encoding loss. The regression test at /works"
 }

benchmarks/csb_sdlc_feature/bustub-hyperloglog-impl-001/tests/ground_truth_agent.json

@@ -0,0 +1,67 @@
+{
+  "files": [
+    "src/include/primer/hyperloglog.h",
+    "src/primer/hyperloglog.cpp",
+    "src/include/primer/hyperloglog_presto.h",
+    "src/primer/hyperloglog_presto.cpp",
+    "src/include/common/util/hash_util.h",
+    "test/primer/hyperloglog_test.cpp"
+  ],
+  "symbols": [
+    {
+      "file": "src/include/primer/hyperloglog.h",
+      "symbol": "HyperLogLog",
+      "repo": null
+    },
+    {
+      "file": "src/include/primer/hyperloglog.h",
+      "symbol": "AddElem",
+      "repo": null
+    },
+    {
+      "file": "src/include/primer/hyperloglog.h",
+      "symbol": "ComputeCardinality",
+      "repo": null
+    },
+    {
+      "file": "src/include/primer/hyperloglog.h",
+      "symbol": "ComputeBinary",
+      "repo": null
+    },
+    {
+      "file": "src/include/primer/hyperloglog.h",
+      "symbol": "PositionOfLeftmostOne",
+      "repo": null
+    },
+    {
+      "file": "src/include/primer/hyperloglog.h",
+      "symbol": "CalculateHash",
+      "repo": null
+    },
+    {
+      "file": "src/include/primer/hyperloglog_presto.h",
+      "symbol": "HyperLogLogPresto",
+      "repo": null
+    },
+    {
+      "file": "src/include/primer/hyperloglog_presto.h",
+      "symbol": "GetDenseBucket",
+      "repo": null
+    },
+    {
+      "file": "src/include/primer/hyperloglog_presto.h",
+      "symbol": "GetOverflowBucketofIndex",
+      "repo": null
+    },
+    {
+      "file": "src/include/common/util/hash_util.h",
+      "symbol": "HashBytes",
+      "repo": null
+    },
+    {
+      "file": "src/include/common/util/hash_util.h",
+      "symbol": "HashValue",
+      "repo": null
+    }
+  ]
+}

benchmarks/csb_sdlc_feature/bustub-hyperloglog-impl-001/tests/ground_truth_meta.json

@@ -1,10 +1,18 @@
 {
-  "model": "manual",
-  "backend": "instruction-extraction",
-  "prompt_version": "manual",
-  "cost_usd": 0.0,
-  "elapsed_sec": 0,
-  "timestamp": "2026-03-03T22:22:15Z",
-  "tool_calls": 0,
-  "generator": "manual_instruction_extraction"
+  "has_ground_truth": true,
+  "has_chunk_ground_truth": false,
+  "ground_truth_source": "curator_agent",
+  "ground_truth_confidence": "medium",
+  "task_name": "bustub-hyperloglog-impl-001",
+  "curator_agent_version": "2.0",
+  "model": "claude-opus-4-6",
+  "backend": "hybrid",
+  "timestamp": "2026-03-06T14:25:14Z",
+  "files_count": 6,
+  "edit_files_count": 0,
+  "chunks_count": 0,
+  "symbols_count": 11,
+  "cost_usd": 1.1280160000000001,
+  "elapsed_sec": 442.5,
+  "exploration_notes": "The task requires implementing HyperLogLog cardinality estimation across 4 files plus a fix to hash_util.h. The test file reveals two variants:\n\n**Basic HyperLogLog** (hyperloglog.h/.cpp): Uses `n_bits` top bits of a 64-bit hash for bucket indexing (m=2^n_bits buckets). `ComputeBinary` converts hash to bitset<64>. `PositionOfLeftmostOne` finds the 1-indexed position of the highest set bit scanning from MSB (bit 63). `AddElem` hashes the value, extracts bucket index from top n_bits, zeros those b"
 }

benchmarks/csb_sdlc_feature/tensorrt-mxfp4-quant-feat-001/tests/ground_truth_agent.json

@@ -0,0 +1,99 @@
+{
+  "files": [
+    "tensorrt_llm/quantization/mode.py",
+    "cpp/include/tensorrt_llm/common/quantization.h",
+    "tensorrt_llm/quantization/utils/fp4_utils.py",
+    "tensorrt_llm/_torch/model_config.py",
+    "cpp/tensorrt_llm/thop/fp4Gemm.cpp",
+    "cpp/tensorrt_llm/thop/moeOp.cpp",
+    "cpp/tensorrt_llm/kernels/cutlass_kernels/include/fp4_gemm.h",
+    "cpp/tensorrt_llm/kernels/internal_cutlass_kernels/include/fp4_gemm.h",
+    "cpp/tensorrt_llm/kernels/cutlass_kernels/fp4_gemm/fp4_gemm_template.h",
+    "cpp/tensorrt_llm/kernels/cutlass_kernels/fp4_gemm/fp4_gemm_fp32.cu",
+    "cpp/tensorrt_llm/kernels/cutlass_kernels/fp4_gemm/fp4_gemm_fp16.cu",
+    "cpp/tensorrt_llm/kernels/cutlass_kernels/fp4_gemm/fp4_gemm_bf16.cu",
+    "tensorrt_llm/_torch/modules/fused_moe/fused_moe_cutlass.py",
+    "tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.py",
+    "tensorrt_llm/_torch/modules/fused_moe/fused_moe_triton.py",
+    "tensorrt_llm/_torch/modules/fused_moe/quantization.py",
+    "tensorrt_llm/_torch/modules/linear.py",
+    "tensorrt_llm/_torch/modules/triton_linear.py",
+    "tensorrt_llm/_torch/custom_ops/torch_custom_ops.py",
+    "tests/unittest/_torch/modules/test_fused_moe.py",
+    "tests/unittest/_torch/modules/test_triton_linear.py",
+    "tests/unittest/_torch/modules/moe/test_moe_backend.py",
+    "tests/unittest/_torch/modules/moe/quantize_utils.py"
+  ],
+  "symbols": [
+    {
+      "file": "tensorrt_llm/quantization/mode.py",
+      "symbol": "QuantAlgo",
+      "repo": null
+    },
+    {
+      "file": "tensorrt_llm/quantization/mode.py",
+      "symbol": "QuantMode",
+      "repo": null
+    },
+    {
+      "file": "cpp/include/tensorrt_llm/common/quantization.h",
+      "symbol": "QuantMode",
+      "repo": null
+    },
+    {
+      "file": "cpp/tensorrt_llm/kernels/cutlass_kernels/include/fp4_gemm.h",
+      "symbol": "FP4GemmType",
+      "repo": null
+    },
+    {
+      "file": "cpp/tensorrt_llm/kernels/internal_cutlass_kernels/include/fp4_gemm.h",
+      "symbol": "FP4GemmType",
+      "repo": null
+    },
+    {
+      "file": "tensorrt_llm/quantization/utils/fp4_utils.py",
+      "symbol": "FP4GemmType",
+      "repo": null
+    },
+    {
+      "file": "tensorrt_llm/_torch/model_config.py",
+      "symbol": "ModelConfig.get_mxfp4_quant_algo",
+      "repo": null
+    },
+    {
+      "file": "tensorrt_llm/_torch/model_config.py",
+      "symbol": "ModelConfig.override_quant_algo",
+      "repo": null
+    },
+    {
+      "file": "tensorrt_llm/_torch/modules/fused_moe/quantization.py",
+      "symbol": "W4A8MXFP4FP8CutlassFusedMoEMethod",
+      "repo": null
+    },
+    {
+      "file": "tensorrt_llm/_torch/modules/fused_moe/quantization.py",
+      "symbol": "W4A8MXFP4MXFP8CutlassFusedMoEMethod",
+      "repo": null
+    },
+    {
+      "file": "tensorrt_llm/_torch/modules/fused_moe/quantization.py",
+      "symbol": "W4A8MXFP4FP8TRTLLMGenFusedMoEMethod",
+      "repo": null
+    },
+    {
+      "file": "tensorrt_llm/_torch/modules/fused_moe/quantization.py",
+      "symbol": "W4A8MXFP4MXFP8TRTLLMGenFusedMoEMethod",
+      "repo": null
+    },
+    {
+      "file": "tensorrt_llm/_torch/modules/fused_moe/quantization.py",
+      "symbol": "FusedMoEQuantScalesW4A8MXFP4FP8",
+      "repo": null
+    },
+    {
+      "file": "tensorrt_llm/_torch/modules/fused_moe/quantization.py",
+      "symbol": "FusedMoEQuantScalesW4A8MXFP4MXFP8",
+      "repo": null
+    }
+  ]
+}

benchmarks/csb_sdlc_feature/tensorrt-mxfp4-quant-feat-001/tests/ground_truth_meta.json

@@ -0,0 +1,18 @@
+{
+  "has_ground_truth": true,
+  "has_chunk_ground_truth": false,
+  "ground_truth_source": "curator_agent",
+  "ground_truth_confidence": "medium",
+  "task_name": "tensorrt-mxfp4-quant-feat-001",
+  "curator_agent_version": "2.0",
+  "model": "claude-opus-4-6",
+  "backend": "hybrid",
+  "timestamp": "2026-03-06T14:19:49Z",
+  "files_count": 23,
+  "edit_files_count": 0,
+  "chunks_count": 0,
+  "symbols_count": 14,
+  "cost_usd": 1.0432487499999998,
+  "elapsed_sec": 115.5,
+  "exploration_notes": "The W4A8_MXFP4_INT8 mode needs to be added following the pattern of W4A8_MXFP4_FP8 and W4A8_MXFP4_MXFP8. Key changes span:\n\n**Python enums & mode mapping**: `tensorrt_llm/quantization/mode.py` (QuantAlgo enum + QuantMode IntFlag + from_quant_algo/from_description). The Python FP4GemmType in `tensorrt_llm/quantization/utils/fp4_utils.py` may need a new variant.\n\n**C++ enums & mode mapping**: `cpp/include/tensorrt_llm/common/quantization.h` (QuantMode bit flag + hasW4a8Mxfp4Int8 + fromDescription "
+}

benchmarks/csb_sdlc_fix/element-web-unread-indicators-diverge-fix-001/tests/ground_truth_agent.json

@@ -0,0 +1,51 @@
+{
+  "files": [
+    "apps/web/src/Unread.ts",
+    "apps/web/test/unit-tests/Unread-test.ts",
+    "apps/web/src/RoomNotifs.ts",
+    "apps/web/src/shouldHideEvent.ts",
+    "apps/web/src/events/EventTileFactory.tsx"
+  ],
+  "symbols": [
+    {
+      "file": "apps/web/src/Unread.ts",
+      "symbol": "doesTimelineHaveUnreadMessages",
+      "repo": null
+    },
+    {
+      "file": "apps/web/src/Unread.ts",
+      "symbol": "doesRoomHaveUnreadMessages",
+      "repo": null
+    },
+    {
+      "file": "apps/web/src/Unread.ts",
+      "symbol": "doesRoomHaveUnreadThreads",
+      "repo": null
+    },
+    {
+      "file": "apps/web/src/Unread.ts",
+      "symbol": "doesRoomOrThreadHaveUnreadMessages",
+      "repo": null
+    },
+    {
+      "file": "apps/web/src/Unread.ts",
+      "symbol": "eventTriggersUnreadCount",
+      "repo": null
+    },
+    {
+      "file": "apps/web/src/Unread.ts",
+      "symbol": "isRelevantEvent",
+      "repo": null
+    },
+    {
+      "file": "apps/web/src/Unread.ts",
+      "symbol": "findLatestRelevantEvent",
+      "repo": null
+    },
+    {
+      "file": "apps/web/src/RoomNotifs.ts",
+      "symbol": "determineUnreadState",
+      "repo": null
+    }
+  ]
+}