Datasets:

Modalities:
Text
Formats:
parquet
ArXiv:
License:
JerroldH commited on
Commit
d59aabb
·
verified ·
1 Parent(s): 0fb314f

Release community v0.1 Markdown and package datasets

Browse files
.gitattributes CHANGED
@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
 
 
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
61
+ records.jsonl filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,182 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ pretty_name: SkillScreen Data
3
+ license: other
4
+ license_name: mixed-source-terms
5
+ license_link: https://huggingface.co/datasets/SkillScreen/SkillScreen-Data#sources-and-terms
6
+ task_categories:
7
+ - text-classification
8
+ configs:
9
+ - config_name: markdown_random
10
+ default: true
11
+ data_files:
12
+ - split: train
13
+ path: data/markdown_random/train.parquet
14
+ - split: validation
15
+ path: data/markdown_random/validation.parquet
16
+ - split: test
17
+ path: data/markdown_random/test.parquet
18
+ - config_name: markdown_sd
19
+ data_files:
20
+ - split: train
21
+ path: data/markdown_sd/train.parquet
22
+ - split: validation
23
+ path: data/markdown_sd/validation.parquet
24
+ - split: test
25
+ path: data/markdown_sd/test.parquet
26
+ - config_name: package_random
27
+ data_files:
28
+ - split: train
29
+ path: data/package_random/train.parquet
30
+ - split: validation
31
+ path: data/package_random/validation.parquet
32
+ - split: test
33
+ path: data/package_random/test.parquet
34
+ - config_name: package_sd
35
+ data_files:
36
+ - split: train
37
+ path: data/package_sd/train.parquet
38
+ - split: validation
39
+ path: data/package_sd/validation.parquet
40
+ - split: test
41
+ path: data/package_sd/test.parquet
42
+ ---
43
+
44
+ # SkillScreen Data
45
+
46
+ Agent-skill text, labels and source records for training decision models.
47
+ Choose Markdown alone or Markdown with companion files, with the same case IDs.
48
+
49
+ **v0.1** provides four full-text configurations in
50
+ [SkillScreen/SkillScreen-Data](https://huggingface.co/datasets/SkillScreen/SkillScreen-Data).
51
+ [GitHub](https://github.com/JerroldH/SkillScreen/tree/community-v0.1) keeps the
52
+ index, export scripts and contribution workflow.
53
+
54
+ ## Load the data
55
+
56
+ Install `datasets` and choose an input view and split protocol. No upstream
57
+ recovery, GPU, API key or custom dataset code is needed.
58
+
59
+ ```python
60
+ from datasets import load_dataset
61
+
62
+ ds = load_dataset("SkillScreen/SkillScreen-Data", "package_sd")
63
+ sample = ds["train"][0]
64
+ skill_text = sample["skill_text"]
65
+ companion_files = sample["files"] # [{"path": ..., "text": ..., "sha256": ...}]
66
+ label = sample["label"] # 0 = benign, 1 = malicious
67
+ ```
68
+
69
+ For a downloaded copy, replace the repository name with the local dataset
70
+ folder. Each configuration contains `train`, `validation` and `test` sets.
71
+
72
+ | Configuration | Input | Train | Validation | Test |
73
+ | --- | --- | ---: | ---: | ---: |
74
+ | `markdown_random` | SKILL.md | 9,461 | 1,350 | 2,703 |
75
+ | `markdown_sd` | SKILL.md | 10,607 | 1,516 | 1,364 |
76
+ | `package_random` | SKILL.md + eligible companion text | 8,290 | 1,178 | 2,363 |
77
+ | `package_sd` | SKILL.md + eligible companion text | 9,563 | 1,369 | 874 |
78
+
79
+ The pool has **13,514 records**, comprising 7,500 malicious and 6,014 benign
80
+ skills. **11,831 packages** have verified complete inventories. Of those,
81
+ 7,036 have eligible companion text and 4,795 have no additional eligible text.
82
+ Files are stored as data inside Parquet, never installed or executed.
83
+ Treat their instructions as untrusted input.
84
+
85
+ ## Inputs and provenance
86
+
87
+ | Column | Meaning |
88
+ | --- | --- |
89
+ | `id`, `group_id` | Stable normalized-Markdown content group |
90
+ | `label` | ClassLabel, 0 benign / 1 malicious |
91
+ | `skill_text`, `skill_sha256` | Exact canonical Markdown and its UTF-8 hash |
92
+ | `files` | Companion paths, decoded text and hashes, sorted by path; empty in Markdown configs |
93
+ | `coverage` | Original package availability, independent of the selected input mode |
94
+ | `source_id`, `source_ids`, `source_name` | Primary source collection and all public source aliases |
95
+ | `source_url`, `revision`, `member_ids` | Fixed source version and original public record IDs |
96
+ | `label_origin`, `evidence` | Origin and basis of the reference label |
97
+ | `upstream_license`, `package_source` | Source terms and the selected package's URL, revision and terms |
98
+
99
+ Use `skill_text` and, for package models, `files` as model inputs. Labels,
100
+ source names and evidence are metadata. Canonical SKILL.md is identical across
101
+ the two input views. Package text includes eligible scripts, manifests and
102
+ documents. Binary files and benchmark annotations are excluded; runtime
103
+ dependencies are not bundled. Models choose their own context limit.
104
+
105
+ The accompanying `records.jsonl` contains full inventories, original byte
106
+ hashes, public ID mappings and recovery locations. Its text fields stay null
107
+ because text lives in Parquet. `sources.json` pins the download assets.
108
+ `splits/` preserves the original assignment CSVs, including exclusions.
109
+
110
+ ## Splits and labels
111
+
112
+ Grouped random is source/label-stratified with seed 42 and keeps normalized
113
+ duplicates together. SD holds out SkillHarm (SRC009), ATR (SRC011) and
114
+ SkillFortifyBench (SRC012). Train and validation may share source collections.
115
+ These groups refer to benchmark collections, not individual repositories.
116
+
117
+ **The existing SD assignment is unchanged.** Its Markdown test has 838 malicious
118
+ and 526 benign cases. Twenty-seven excluded IDs remain in the index and split
119
+ CSV, outside the SD Parquet sets. Package configs select complete packages
120
+ within each existing partition without reshuffling. The package SD test has
121
+ 784 malicious and 90 benign cases, reflecting incomplete ATR package coverage.
122
+ For paired comparisons, filter Markdown to the IDs in the corresponding package
123
+ partition before comparing models.
124
+
125
+ These community splits are separate from the paper's SBR/MSD/SD protocols.
126
+ Existing checkpoints used earlier protocols, so these partitions are not fresh
127
+ held-out tests for those checkpoints. The original paper artifacts are unchanged.
128
+
129
+ Labels are inherited upstream references. Identical normalized Markdown shares
130
+ one ID. Six content groups with unresolved label questions are withheld.
131
+ The pool contains no private labels or weak positives. Additional coverage is
132
+ 1,634 Markdown-only, 25 partial and 24 ambiguous records. Ambiguous package
133
+ versions are not combined.
134
+
135
+ ## Build or audit a local copy
136
+
137
+ From the code repository with Python 3.12+, restore the fixed source data once,
138
+ then build and verify all four configs. Use a new output directory.
139
+
140
+ ```sh
141
+ python -m pip install -r requirements/data.txt
142
+ python scripts/restore_community.py --mode markdown --output outputs/community-markdown.jsonl
143
+ python scripts/restore_community.py --mode package --output outputs/community-packages.jsonl
144
+ python scripts/export_community.py build --markdown-text outputs/community-markdown.jsonl --package-text outputs/community-packages.jsonl --output outputs/hf-fulltext
145
+ ```
146
+
147
+ `build` also runs `load_dataset` for every config and verifies every ID, label,
148
+ eligible text hash and split membership. It writes `verification.json` in the
149
+ export folder and never uploads. Verify an existing export with:
150
+
151
+ ```sh
152
+ python scripts/export_community.py verify --output outputs/hf-fulltext
153
+ ```
154
+
155
+ Recovery is an optional provenance check for maintainers, not a step required
156
+ by users loading a generated release. Downloads are cached as inert hash blobs.
157
+ No source file is executed. The index-only export remains available through
158
+ `python scripts/community_data.py export --output outputs/hf-index`.
159
+
160
+ ## Contribute
161
+
162
+ Start with a public source URL, fixed revision and evidence through the
163
+ [case form](https://github.com/JerroldH/SkillScreen/issues/new?template=case.yml).
164
+ Maintainers check labels, duplicates, coverage and terms. Unknown cases stay out
165
+ of supervised sets. New cases remain excluded from frozen protocols until the
166
+ next release. Markdown and package views are generated from one shared index.
167
+
168
+ ## Sources and terms
169
+
170
+ - [MaliciousSkillBench](https://huggingface.co/datasets/ProtectSkills/MaliciousSkillBench/tree/d4b42ce5766a6e0359c987cf59c1007cb3795a90) supplies benchmark labels and Markdown. Its metadata and splits use CC BY 4.0; third-party skill contents retain their original terms.
171
+ - [MalSkillBench](https://github.com/lxyeternal/MalSkillBench/tree/3548e2837adf58b96497f2c688f96beff26485da) supplies additional benign packages and SRC001 cases. Its notice states academic research use only.
172
+ - [SkillTrustBench](https://huggingface.co/datasets/cuhk-zhuque/SkillTrustBench/tree/f90517b7058fdcfea89af114c069fbf973f42bc7) supplies SRC010 packages under its CC BY-NC-SA 4.0 notice and retained upstream terms.
173
+ - [SkillHarm](https://huggingface.co/datasets/osunlp/SkillHarm/tree/cc21f90982ea49de43bc30a08f338788a92916f4) supplies SRC009 cases; its dataset card lists CC BY 4.0.
174
+ - [SkillFortifyBench](https://github.com/qualixar/skillfortifybench/tree/eb9d5a9cbfa13b82ec3392d26dbae82fc6454b39) supplies native malicious and benign cases under MIT.
175
+ - [Official package archives](https://github.com/protectskills/MaliciousSkillBench/tree/893afd609be22a2dfac5a2b94bbebb4eeb99515e/packages) provide fixed package snapshots without relicensing their contents.
176
+
177
+ Our original documentation and annotations use CC BY 4.0. Imported bodies
178
+ retain their source terms; the project code's Apache-2.0 license does not apply
179
+ to them. `unknown` means a record's original content license has not been
180
+ resolved. The dataset does not grant additional rights over imported text.
181
+
182
+ Please cite the source datasets you use and the [SkillScreen study](https://arxiv.org/abs/2610.17727).
data/markdown_random/test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5ea2359b6dae03d68be1e5e03d77ec512fa2e44965f8f77b4d426e73a090c423
3
+ size 8235638
data/markdown_random/train.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2f8bbd2bede1b1197e7fbdee4b6b646f6677223ce69e17328db6c65768c4ce31
3
+ size 28788652
data/markdown_random/validation.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:402322a9c0f8814b353b704977dd01e1c019c225c190f45e45de840b2452249f
3
+ size 4192335
data/markdown_sd/test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a67969615f9905b4c52dc9c2723e48954425a42452c53f56ca945e04f50025ad
3
+ size 2708901
data/markdown_sd/train.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:525fe6ca2ec0966836aa20c1166e7b6f9152844018ee32eb3e0962531d69d52e
3
+ size 32897411
data/markdown_sd/validation.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ddab6d842e0d43812e22688bc461b3ea4267ce38ed5dc1a1fc3f30409c13fc44
3
+ size 4907736
data/package_random/test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ce4a39b108c7fd944beb9f96c54dd4483dbe79714f4fdd5faf9a185b7861a564
3
+ size 19673901
data/package_random/train.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ca16e7d09884c14250bd7f0063ae1525b2bf60b423bf98f271c0260b60b5ce39
3
+ size 66081019
data/package_random/validation.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:34e9b189e2f7d65f71a1f7fef17314644e506ab0cabf3f67c2ef206017d10191
3
+ size 9545259
data/package_sd/test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:79e1f77bd417e18bfd41812e0c24e71dffef5edc35505092273f2233d8925d4b
3
+ size 3089365
data/package_sd/train.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0acfc9a366e33d336212444bbc1888c43767a2746d14e141f8e03a94a63f7c3f
3
+ size 76824894
data/package_sd/validation.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:70f333ea18a765ee825e01dc48bd23aba28b78b706b4b9ff89654574358d9c98
3
+ size 10970583
records.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:26f05a8a82573e6e7641e93e1b00a60be726ee605ea07a8185f4e8021b7f71a3
3
+ size 32024296
sources.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "primary": {
3
+ "url": "https://huggingface.co/datasets/ProtectSkills/MaliciousSkillBench/resolve/d4b42ce5766a6e0359c987cf59c1007cb3795a90/primary.parquet",
4
+ "sha256": "1d6382017f259bde0b699d402318f3c03f16b2e8ed2363a346b77a5465f06a06",
5
+ "format": "parquet"
6
+ },
7
+ "skilltrust": {
8
+ "url": "https://huggingface.co/datasets/cuhk-zhuque/SkillTrustBench/resolve/f90517b7058fdcfea89af114c069fbf973f42bc7/benchmark_full_v1.0.zip",
9
+ "sha256": "e1d8950ef01c3b24fa80e32101844abc8c5ab3a0a38525427e8b16f00a414ae4",
10
+ "format": "zip"
11
+ },
12
+ "SRC001": {
13
+ "url": "https://raw.githubusercontent.com/protectskills/MaliciousSkillBench/893afd609be22a2dfac5a2b94bbebb4eeb99515e/packages/SRC001_packages.tar.gz",
14
+ "sha256": "878498b42d9c67d0b084ef416153d69742ef75e57c938d7a38614f9bb914c3db",
15
+ "format": "tar.gz"
16
+ },
17
+ "SRC009": {
18
+ "url": "https://raw.githubusercontent.com/protectskills/MaliciousSkillBench/893afd609be22a2dfac5a2b94bbebb4eeb99515e/packages/SRC009_packages.tar.gz",
19
+ "sha256": "10bf2de3d35b50a3c8888356831d43b4963f61f726cab931aeab33ecd60c882f",
20
+ "format": "tar.gz"
21
+ },
22
+ "SRC012": {
23
+ "url": "https://raw.githubusercontent.com/protectskills/MaliciousSkillBench/893afd609be22a2dfac5a2b94bbebb4eeb99515e/packages/SRC012_artifacts.tar.gz",
24
+ "sha256": "b57c3663d9d943802bdb341659ba8e0cd0ffa4befcac407e268f4a8ad9a6398d",
25
+ "format": "tar.gz"
26
+ }
27
+ }
splits/grouped_random.csv ADDED
The diff for this file is too large to render. See raw diff
 
splits/source_disjoint.csv ADDED
The diff for this file is too large to render. See raw diff
 
verification.json ADDED
@@ -0,0 +1,251 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "passed": true,
3
+ "checked_at": "2026-11-09T16:07:36.950523+00:00",
4
+ "configs": {
5
+ "markdown_random": {
6
+ "train": {
7
+ "rows": 9461,
8
+ "labels": {
9
+ "benign": 4211,
10
+ "malicious": 5250
11
+ },
12
+ "primary_sources": {
13
+ "msb:SRC010": 3097,
14
+ "msb:SRC009": 509,
15
+ "msb:SRC011": 333,
16
+ "msb:SRC001": 5043,
17
+ "msb:SRC004": 23,
18
+ "msb:SRC012": 126,
19
+ "msb:SRC013": 108,
20
+ "msb:SRC006": 91,
21
+ "msb:SRC002": 107,
22
+ "msb:SRC005": 21,
23
+ "msb:SRC008": 3
24
+ },
25
+ "companion_files": 0,
26
+ "ids_sha256": "0833dd07078eb8f08807d25d4aab48e60219120ac7d9848fde8ee042b849d4f7"
27
+ },
28
+ "validation": {
29
+ "rows": 1350,
30
+ "labels": {
31
+ "malicious": 749,
32
+ "benign": 601
33
+ },
34
+ "primary_sources": {
35
+ "msb:SRC001": 720,
36
+ "msb:SRC010": 442,
37
+ "msb:SRC011": 47,
38
+ "msb:SRC009": 73,
39
+ "msb:SRC013": 15,
40
+ "msb:SRC002": 15,
41
+ "msb:SRC004": 3,
42
+ "msb:SRC006": 14,
43
+ "msb:SRC012": 18,
44
+ "msb:SRC005": 3
45
+ },
46
+ "companion_files": 0,
47
+ "ids_sha256": "e7b43d8f0972461febaabcbcd0a1f93409af65d8da3c0b54a473a695bf7fe9bc"
48
+ },
49
+ "test": {
50
+ "rows": 2703,
51
+ "labels": {
52
+ "benign": 1202,
53
+ "malicious": 1501
54
+ },
55
+ "primary_sources": {
56
+ "msb:SRC010": 884,
57
+ "msb:SRC001": 1441,
58
+ "msb:SRC002": 31,
59
+ "msb:SRC012": 36,
60
+ "msb:SRC006": 26,
61
+ "msb:SRC011": 95,
62
+ "msb:SRC013": 31,
63
+ "msb:SRC009": 146,
64
+ "msb:SRC005": 6,
65
+ "msb:SRC004": 7
66
+ },
67
+ "companion_files": 0,
68
+ "ids_sha256": "a9af1ae8f22c31f1b8387c2f5ea169f2598f8db56a47951246673dcd038f1338"
69
+ }
70
+ },
71
+ "markdown_sd": {
72
+ "train": {
73
+ "rows": 10607,
74
+ "labels": {
75
+ "benign": 4786,
76
+ "malicious": 5821
77
+ },
78
+ "primary_sources": {
79
+ "msb:SRC010": 3870,
80
+ "msb:SRC001": 6296,
81
+ "msb:SRC004": 29,
82
+ "msb:SRC002": 134,
83
+ "msb:SRC013": 135,
84
+ "msb:SRC006": 114,
85
+ "msb:SRC005": 26,
86
+ "msb:SRC008": 3
87
+ },
88
+ "companion_files": 0,
89
+ "ids_sha256": "fd228df9b468298140db92f16e6888a2521751b8ce7a4cd670afedaa5c4685d7"
90
+ },
91
+ "validation": {
92
+ "rows": 1516,
93
+ "labels": {
94
+ "benign": 684,
95
+ "malicious": 832
96
+ },
97
+ "primary_sources": {
98
+ "msb:SRC001": 900,
99
+ "msb:SRC010": 553,
100
+ "msb:SRC013": 19,
101
+ "msb:SRC002": 19,
102
+ "msb:SRC004": 4,
103
+ "msb:SRC006": 17,
104
+ "msb:SRC005": 4
105
+ },
106
+ "companion_files": 0,
107
+ "ids_sha256": "974543026ff38978bf621bc5a4ca44d43e42c44dfb9e1fcbe2c2d015679f9c8e"
108
+ },
109
+ "test": {
110
+ "rows": 1364,
111
+ "labels": {
112
+ "malicious": 838,
113
+ "benign": 526
114
+ },
115
+ "primary_sources": {
116
+ "msb:SRC009": 728,
117
+ "msb:SRC011": 456,
118
+ "msb:SRC012": 180
119
+ },
120
+ "companion_files": 0,
121
+ "ids_sha256": "a36f1f63755621d82188098882faa71869aa5eeeb3915ab2346cdd5338a66bba"
122
+ }
123
+ },
124
+ "package_random": {
125
+ "train": {
126
+ "rows": 8290,
127
+ "labels": {
128
+ "benign": 3779,
129
+ "malicious": 4511
130
+ },
131
+ "primary_sources": {
132
+ "msb:SRC010": 2860,
133
+ "msb:SRC009": 486,
134
+ "msb:SRC001": 4808,
135
+ "msb:SRC012": 126,
136
+ "msb:SRC011": 10
137
+ },
138
+ "companion_files": 24504,
139
+ "ids_sha256": "ff4461d4f5178eac703abc2659b9b47d511e0111313e75b7299222a5cb162a20"
140
+ },
141
+ "validation": {
142
+ "rows": 1178,
143
+ "labels": {
144
+ "malicious": 637,
145
+ "benign": 541
146
+ },
147
+ "primary_sources": {
148
+ "msb:SRC001": 678,
149
+ "msb:SRC010": 409,
150
+ "msb:SRC009": 71,
151
+ "msb:SRC012": 18,
152
+ "msb:SRC011": 2
153
+ },
154
+ "companion_files": 3559,
155
+ "ids_sha256": "52dae4e84d8ddcf336d8a564745aa89d975a7d1f6001ac7f644114a6513fb184"
156
+ },
157
+ "test": {
158
+ "rows": 2363,
159
+ "labels": {
160
+ "benign": 1075,
161
+ "malicious": 1288
162
+ },
163
+ "primary_sources": {
164
+ "msb:SRC010": 812,
165
+ "msb:SRC001": 1373,
166
+ "msb:SRC012": 36,
167
+ "msb:SRC009": 137,
168
+ "msb:SRC011": 5
169
+ },
170
+ "companion_files": 6556,
171
+ "ids_sha256": "3562073cabbffbffe8131f0a8fac5049157383d7f8989b5c17c3e2ffeb9c873e"
172
+ }
173
+ },
174
+ "package_sd": {
175
+ "train": {
176
+ "rows": 9563,
177
+ "labels": {
178
+ "benign": 4625,
179
+ "malicious": 4938
180
+ },
181
+ "primary_sources": {
182
+ "msb:SRC010": 3567,
183
+ "msb:SRC001": 5996
184
+ },
185
+ "companion_files": 27098,
186
+ "ids_sha256": "191fc1d67c1afe9e9752d2b4cd891d35dde10bc854e922536e1e751a2058a000"
187
+ },
188
+ "validation": {
189
+ "rows": 1369,
190
+ "labels": {
191
+ "benign": 663,
192
+ "malicious": 706
193
+ },
194
+ "primary_sources": {
195
+ "msb:SRC001": 855,
196
+ "msb:SRC010": 514
197
+ },
198
+ "companion_files": 4065,
199
+ "ids_sha256": "e107432eefa0d94f104b9ceeb89d2992b2308b3341ca73fe53e1c420a694984c"
200
+ },
201
+ "test": {
202
+ "rows": 874,
203
+ "labels": {
204
+ "malicious": 784,
205
+ "benign": 90
206
+ },
207
+ "primary_sources": {
208
+ "msb:SRC009": 694,
209
+ "msb:SRC012": 180
210
+ },
211
+ "companion_files": 3370,
212
+ "ids_sha256": "2a33173739cab10ed108bad5038beaac07026099c5298dba6a0f0d7a22a6c86d"
213
+ }
214
+ }
215
+ },
216
+ "versions": {
217
+ "datasets": "5.0.1",
218
+ "pyarrow": "25.0.1"
219
+ },
220
+ "split_sha256": {
221
+ "grouped_random": "05736cffc3c3d770d8aea66c023ae5971559cf784192f99a95847e2b8c236748",
222
+ "source_disjoint": "639ff609afd54781aecbf6272374b39fd1431feada2bd16f75256e8c9f04f3c9"
223
+ },
224
+ "checks": [
225
+ "All four configs loaded by name with train/validation/test",
226
+ "Every ID, label, source field and eligible text hash matches the index",
227
+ "Original split files copied byte-for-byte, including SD exclusions",
228
+ "Package cohorts preserve Markdown memberships and canonical SKILL.md",
229
+ "Groups and exact Markdown hashes do not cross partitions",
230
+ "SD test sources do not appear in train or validation"
231
+ ],
232
+ "files_sha256": {
233
+ "data/markdown_random/test.parquet": "5ea2359b6dae03d68be1e5e03d77ec512fa2e44965f8f77b4d426e73a090c423",
234
+ "data/markdown_random/train.parquet": "2f8bbd2bede1b1197e7fbdee4b6b646f6677223ce69e17328db6c65768c4ce31",
235
+ "data/markdown_random/validation.parquet": "402322a9c0f8814b353b704977dd01e1c019c225c190f45e45de840b2452249f",
236
+ "data/markdown_sd/test.parquet": "a67969615f9905b4c52dc9c2723e48954425a42452c53f56ca945e04f50025ad",
237
+ "data/markdown_sd/train.parquet": "525fe6ca2ec0966836aa20c1166e7b6f9152844018ee32eb3e0962531d69d52e",
238
+ "data/markdown_sd/validation.parquet": "ddab6d842e0d43812e22688bc461b3ea4267ce38ed5dc1a1fc3f30409c13fc44",
239
+ "data/package_random/test.parquet": "ce4a39b108c7fd944beb9f96c54dd4483dbe79714f4fdd5faf9a185b7861a564",
240
+ "data/package_random/train.parquet": "ca16e7d09884c14250bd7f0063ae1525b2bf60b423bf98f271c0260b60b5ce39",
241
+ "data/package_random/validation.parquet": "34e9b189e2f7d65f71a1f7fef17314644e506ab0cabf3f67c2ef206017d10191",
242
+ "data/package_sd/test.parquet": "79e1f77bd417e18bfd41812e0c24e71dffef5edc35505092273f2233d8925d4b",
243
+ "data/package_sd/train.parquet": "0acfc9a366e33d336212444bbc1888c43767a2746d14e141f8e03a94a63f7c3f",
244
+ "data/package_sd/validation.parquet": "70f333ea18a765ee825e01dc48bd23aba28b78b706b4b9ff89654574358d9c98",
245
+ "README.md": "da23c67950e47cd1d443fd5f7d4d66a6f632f1c579cc07e7713a05bd6f9373be",
246
+ "records.jsonl": "26f05a8a82573e6e7641e93e1b00a60be726ee605ea07a8185f4e8021b7f71a3",
247
+ "sources.json": "aa19b6f33983fae2f0b7f425a632bf4690bf7e6b227c064a6f8dd275bde83732",
248
+ "splits/grouped_random.csv": "05736cffc3c3d770d8aea66c023ae5971559cf784192f99a95847e2b8c236748",
249
+ "splits/source_disjoint.csv": "639ff609afd54781aecbf6272374b39fd1431feada2bd16f75256e8c9f04f3c9"
250
+ }
251
+ }