JerroldH commited on
Commit
b60475f
·
verified ·
1 Parent(s): 08ffbe3

Release v0.1.0: full-data deployment model, inference entry point and OOF thresholds

Browse files
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
CITATION.cff ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ cff-version: 1.2.0
2
+ message: "If you use this work in research, please cite the paper in preferred-citation."
3
+ type: software
4
+ title: "SkillScreen"
5
+ authors:
6
+ - family-names: Huang
7
+ given-names: Jerrold
8
+ - family-names: Quan
9
+ given-names: Junwei
10
+ - family-names: Lyu
11
+ given-names: Saiyue
12
+ - family-names: Guo
13
+ given-names: Zihan
14
+ - family-names: Ge
15
+ given-names: Yingqiang
16
+ url: "https://github.com/JerroldH/SkillScreen"
17
+ license: Apache-2.0
18
+ preferred-citation:
19
+ type: article
20
+ license: CC-BY-4.0
21
+ title: "SkillScreen: High-Recall Screening of Malicious Agent Skills with Decision Models"
22
+ authors:
23
+ - family-names: Huang
24
+ given-names: Jerrold
25
+ - family-names: Quan
26
+ given-names: Junwei
27
+ - family-names: Lyu
28
+ given-names: Saiyue
29
+ - family-names: Guo
30
+ given-names: Zihan
31
+ - family-names: Ge
32
+ given-names: Yingqiang
33
+ journal: "arXiv"
34
+ year: 2026
35
+ doi: "10.48550/arXiv.2610.17727"
36
+ url: "https://arxiv.org/abs/2610.17727"
LICENSE ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ Apache License
3
+ Version 2.0, January 2004
4
+ http://www.apache.org/licenses/
5
+
6
+ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
7
+
8
+ 1. Definitions.
9
+
10
+ "License" shall mean the terms and conditions for use, reproduction,
11
+ and distribution as defined by Sections 1 through 9 of this document.
12
+
13
+ "Licensor" shall mean the copyright owner or entity authorized by
14
+ the copyright owner that is granting the License.
15
+
16
+ "Legal Entity" shall mean the union of the acting entity and all
17
+ other entities that control, are controlled by, or are under common
18
+ control with that entity. For the purposes of this definition,
19
+ "control" means (i) the power, direct or indirect, to cause the
20
+ direction or management of such entity, whether by contract or
21
+ otherwise, or (ii) ownership of fifty percent (50%) or more of the
22
+ outstanding shares, or (iii) beneficial ownership of such entity.
23
+
24
+ "You" (or "Your") shall mean an individual or Legal Entity
25
+ exercising permissions granted by this License.
26
+
27
+ "Source" form shall mean the preferred form for making modifications,
28
+ including but not limited to software source code, documentation
29
+ source, and configuration files.
30
+
31
+ "Object" form shall mean any form resulting from mechanical
32
+ transformation or translation of a Source form, including but
33
+ not limited to compiled object code, generated documentation,
34
+ and conversions to other media types.
35
+
36
+ "Work" shall mean the work of authorship, whether in Source or
37
+ Object form, made available under the License, as indicated by a
38
+ copyright notice that is included in or attached to the work
39
+ (an example is provided in the Appendix below).
40
+
41
+ "Derivative Works" shall mean any work, whether in Source or Object
42
+ form, that is based on (or derived from) the Work and for which the
43
+ editorial revisions, annotations, elaborations, or other modifications
44
+ represent, as a whole, an original work of authorship. For the purposes
45
+ of this License, Derivative Works shall not include works that remain
46
+ separable from, or merely link (or bind by name) to the interfaces of,
47
+ the Work and Derivative Works thereof.
48
+
49
+ "Contribution" shall mean any work of authorship, including
50
+ the original version of the Work and any modifications or additions
51
+ to that Work or Derivative Works thereof, that is intentionally
52
+ submitted to Licensor for inclusion in the Work by the copyright owner
53
+ or by an individual or Legal Entity authorized to submit on behalf of
54
+ the copyright owner. For the purposes of this definition, "submitted"
55
+ means any form of electronic, verbal, or written communication sent
56
+ to the Licensor or its representatives, including but not limited to
57
+ communication on electronic mailing lists, source code control systems,
58
+ and issue tracking systems that are managed by, or on behalf of, the
59
+ Licensor for the purpose of discussing and improving the Work, but
60
+ excluding communication that is conspicuously marked or otherwise
61
+ designated in writing by the copyright owner as "Not a Contribution."
62
+
63
+ "Contributor" shall mean Licensor and any individual or Legal Entity
64
+ on behalf of whom a Contribution has been received by Licensor and
65
+ subsequently incorporated within the Work.
66
+
67
+ 2. Grant of Copyright License. Subject to the terms and conditions of
68
+ this License, each Contributor hereby grants to You a perpetual,
69
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
70
+ copyright license to reproduce, prepare Derivative Works of,
71
+ publicly display, publicly perform, sublicense, and distribute the
72
+ Work and such Derivative Works in Source or Object form.
73
+
74
+ 3. Grant of Patent License. Subject to the terms and conditions of
75
+ this License, each Contributor hereby grants to You a perpetual,
76
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
77
+ (except as stated in this section) patent license to make, have made,
78
+ use, offer to sell, sell, import, and otherwise transfer the Work,
79
+ where such license applies only to those patent claims licensable
80
+ by such Contributor that are necessarily infringed by their
81
+ Contribution(s) alone or by combination of their Contribution(s)
82
+ with the Work to which such Contribution(s) was submitted. If You
83
+ institute patent litigation against any entity (including a
84
+ cross-claim or counterclaim in a lawsuit) alleging that the Work
85
+ or a Contribution incorporated within the Work constitutes direct
86
+ or contributory patent infringement, then any patent licenses
87
+ granted to You under this License for that Work shall terminate
88
+ as of the date such litigation is filed.
89
+
90
+ 4. Redistribution. You may reproduce and distribute copies of the
91
+ Work or Derivative Works thereof in any medium, with or without
92
+ modifications, and in Source or Object form, provided that You
93
+ meet the following conditions:
94
+
95
+ (a) You must give any other recipients of the Work or
96
+ Derivative Works a copy of this License; and
97
+
98
+ (b) You must cause any modified files to carry prominent notices
99
+ stating that You changed the files; and
100
+
101
+ (c) You must retain, in the Source form of any Derivative Works
102
+ that You distribute, all copyright, patent, trademark, and
103
+ attribution notices from the Source form of the Work,
104
+ excluding those notices that do not pertain to any part of
105
+ the Derivative Works; and
106
+
107
+ (d) If the Work includes a "NOTICE" text file as part of its
108
+ distribution, then any Derivative Works that You distribute must
109
+ include a readable copy of the attribution notices contained
110
+ within such NOTICE file, excluding those notices that do not
111
+ pertain to any part of the Derivative Works, in at least one
112
+ of the following places: within a NOTICE text file distributed
113
+ as part of the Derivative Works; within the Source form or
114
+ documentation, if provided along with the Derivative Works; or,
115
+ within a display generated by the Derivative Works, if and
116
+ wherever such third-party notices normally appear. The contents
117
+ of the NOTICE file are for informational purposes only and
118
+ do not modify the License. You may add Your own attribution
119
+ notices within Derivative Works that You distribute, alongside
120
+ or as an addendum to the NOTICE text from the Work, provided
121
+ that such additional attribution notices cannot be construed
122
+ as modifying the License.
123
+
124
+ You may add Your own copyright statement to Your modifications and
125
+ may provide additional or different license terms and conditions
126
+ for use, reproduction, or distribution of Your modifications, or
127
+ for any such Derivative Works as a whole, provided Your use,
128
+ reproduction, and distribution of the Work otherwise complies with
129
+ the conditions stated in this License.
130
+
131
+ 5. Submission of Contributions. Unless You explicitly state otherwise,
132
+ any Contribution intentionally submitted for inclusion in the Work
133
+ by You to the Licensor shall be under the terms and conditions of
134
+ this License, without any additional terms or conditions.
135
+ Notwithstanding the above, nothing herein shall supersede or modify
136
+ the terms of any separate license agreement you may have executed
137
+ with Licensor regarding such Contributions.
138
+
139
+ 6. Trademarks. This License does not grant permission to use the trade
140
+ names, trademarks, service marks, or product names of the Licensor,
141
+ except as required for reasonable and customary use in describing the
142
+ origin of the Work and reproducing the content of the NOTICE file.
143
+
144
+ 7. Disclaimer of Warranty. Unless required by applicable law or
145
+ agreed to in writing, Licensor provides the Work (and each
146
+ Contributor provides its Contributions) on an "AS IS" BASIS,
147
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
148
+ implied, including, without limitation, any warranties or conditions
149
+ of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
150
+ PARTICULAR PURPOSE. You are solely responsible for determining the
151
+ appropriateness of using or redistributing the Work and assume any
152
+ risks associated with Your exercise of permissions under this License.
153
+
154
+ 8. Limitation of Liability. In no event and under no legal theory,
155
+ whether in tort (including negligence), contract, or otherwise,
156
+ unless required by applicable law (such as deliberate and grossly
157
+ negligent acts) or agreed to in writing, shall any Contributor be
158
+ liable to You for damages, including any direct, indirect, special,
159
+ incidental, or consequential damages of any character arising as a
160
+ result of this License or out of the use or inability to use the
161
+ Work (including but not limited to damages for loss of goodwill,
162
+ work stoppage, computer failure or malfunction, or any and all
163
+ other commercial damages or losses), even if such Contributor
164
+ has been advised of the possibility of such damages.
165
+
166
+ 9. Accepting Warranty or Additional Liability. While redistributing
167
+ the Work or Derivative Works thereof, You may choose to offer,
168
+ and charge a fee for, acceptance of support, warranty, indemnity,
169
+ or other liability obligations and/or rights consistent with this
170
+ License. However, in accepting such obligations, You may act only
171
+ on Your own behalf and on Your sole responsibility, not on behalf
172
+ of any other Contributor, and only if You agree to indemnify,
173
+ defend, and hold each Contributor harmless for any liability
174
+ incurred by, or claims asserted against, such Contributor by reason
175
+ of your accepting any such warranty or additional liability.
176
+
177
+ END OF TERMS AND CONDITIONS
178
+
179
+ APPENDIX: How to apply the Apache License to your work.
180
+
181
+ To apply the Apache License to your work, attach the following
182
+ boilerplate notice, with the fields enclosed by brackets "[]"
183
+ replaced with your own identifying information. (Don't include
184
+ the brackets!) The text should be enclosed in the appropriate
185
+ comment syntax for the file format. We also recommend that a
186
+ file or class name and description of purpose be included on the
187
+ same "printed page" as the copyright notice for easier
188
+ identification within third-party archives.
189
+
190
+ Copyright [yyyy] [name of copyright owner]
191
+
192
+ Licensed under the Apache License, Version 2.0 (the "License");
193
+ you may not use this file except in compliance with the License.
194
+ You may obtain a copy of the License at
195
+
196
+ http://www.apache.org/licenses/LICENSE-2.0
197
+
198
+ Unless required by applicable law or agreed to in writing, software
199
+ distributed under the License is distributed on an "AS IS" BASIS,
200
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
201
+ See the License for the specific language governing permissions and
202
+ limitations under the License.
LICENSE-CC-BY-4.0 ADDED
@@ -0,0 +1,156 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Creative Commons Attribution 4.0 International
2
+
3
+ Creative Commons Corporation (“Creative Commons”) is not a law firm and does not provide legal services or legal advice. Distribution of Creative Commons public licenses does not create a lawyer-client or other relationship. Creative Commons makes its licenses and related information available on an “as-is” basis. Creative Commons gives no warranties regarding its licenses, any material licensed under their terms and conditions, or any related information. Creative Commons disclaims all liability for damages resulting from their use to the fullest extent possible.
4
+
5
+ Using Creative Commons Public Licenses
6
+
7
+ Creative Commons public licenses provide a standard set of terms and conditions that creators and other rights holders may use to share original works of authorship and other material subject to copyright and certain other rights specified in the public license below. The following considerations are for informational purposes only, are not exhaustive, and do not form part of our licenses.
8
+
9
+ Considerations for licensors: Our public licenses are intended for use by those authorized to give the public permission to use material in ways otherwise restricted by copyright and certain other rights. Our licenses are irrevocable. Licensors should read and understand the terms and conditions of the license they choose before applying it. Licensors should also secure all rights necessary before applying our licenses so that the public can reuse the material as expected. Licensors should clearly mark any material not subject to the license. This includes other CC-licensed material, or material used under an exception or limitation to copyright. More considerations for licensors.
10
+
11
+ Considerations for the public: By using one of our public licenses, a licensor grants the public permission to use the licensed material under specified terms and conditions. If the licensor’s permission is not necessary for any reason–for example, because of any applicable exception or limitation to copyright–then that use is not regulated by the license. Our licenses grant only permissions under copyright and certain other rights that a licensor has authority to grant. Use of the licensed material may still be restricted for other reasons, including because others have copyright or other rights in the material. A licensor may make special requests, such as asking that all changes be marked or described. Although not required by our licenses, you are encouraged to respect those requests where reasonable. More considerations for the public.
12
+
13
+ Creative Commons Attribution 4.0 International Public License
14
+
15
+ By exercising the Licensed Rights (defined below), You accept and agree to be bound by the terms and conditions of this Creative Commons Attribution 4.0 International Public License ("Public License"). To the extent this Public License may be interpreted as a contract, You are granted the Licensed Rights in consideration of Your acceptance of these terms and conditions, and the Licensor grants You such rights in consideration of benefits the Licensor receives from making the Licensed Material available under these terms and conditions.
16
+
17
+ Section 1 – Definitions.
18
+
19
+ a. Adapted Material means material subject to Copyright and Similar Rights that is derived from or based upon the Licensed Material and in which the Licensed Material is translated, altered, arranged, transformed, or otherwise modified in a manner requiring permission under the Copyright and Similar Rights held by the Licensor. For purposes of this Public License, where the Licensed Material is a musical work, performance, or sound recording, Adapted Material is always produced where the Licensed Material is synched in timed relation with a moving image.
20
+
21
+ b. Adapter's License means the license You apply to Your Copyright and Similar Rights in Your contributions to Adapted Material in accordance with the terms and conditions of this Public License.
22
+
23
+ c. Copyright and Similar Rights means copyright and/or similar rights closely related to copyright including, without limitation, performance, broadcast, sound recording, and Sui Generis Database Rights, without regard to how the rights are labeled or categorized. For purposes of this Public License, the rights specified in Section 2(b)(1)-(2) are not Copyright and Similar Rights.
24
+
25
+ d. Effective Technological Measures means those measures that, in the absence of proper authority, may not be circumvented under laws fulfilling obligations under Article 11 of the WIPO Copyright Treaty adopted on December 20, 1996, and/or similar international agreements.
26
+
27
+ e. Exceptions and Limitations means fair use, fair dealing, and/or any other exception or limitation to Copyright and Similar Rights that applies to Your use of the Licensed Material.
28
+
29
+ f. Licensed Material means the artistic or literary work, database, or other material to which the Licensor applied this Public License.
30
+
31
+ g. Licensed Rights means the rights granted to You subject to the terms and conditions of this Public License, which are limited to all Copyright and Similar Rights that apply to Your use of the Licensed Material and that the Licensor has authority to license.
32
+
33
+ h. Licensor means the individual(s) or entity(ies) granting rights under this Public License.
34
+
35
+ i. Share means to provide material to the public by any means or process that requires permission under the Licensed Rights, such as reproduction, public display, public performance, distribution, dissemination, communication, or importation, and to make material available to the public including in ways that members of the public may access the material from a place and at a time individually chosen by them.
36
+
37
+ j. Sui Generis Database Rights means rights other than copyright resulting from Directive 96/9/EC of the European Parliament and of the Council of 11 March 1996 on the legal protection of databases, as amended and/or succeeded, as well as other essentially equivalent rights anywhere in the world.
38
+
39
+ k. You means the individual or entity exercising the Licensed Rights under this Public License. Your has a corresponding meaning.
40
+
41
+ Section 2 – Scope.
42
+
43
+ a. License grant.
44
+
45
+ 1. Subject to the terms and conditions of this Public License, the Licensor hereby grants You a worldwide, royalty-free, non-sublicensable, non-exclusive, irrevocable license to exercise the Licensed Rights in the Licensed Material to:
46
+
47
+ A. reproduce and Share the Licensed Material, in whole or in part; and
48
+
49
+ B. produce, reproduce, and Share Adapted Material.
50
+
51
+ 2. Exceptions and Limitations. For the avoidance of doubt, where Exceptions and Limitations apply to Your use, this Public License does not apply, and You do not need to comply with its terms and conditions.
52
+
53
+ 3. Term. The term of this Public License is specified in Section 6(a).
54
+
55
+ 4. Media and formats; technical modifications allowed. The Licensor authorizes You to exercise the Licensed Rights in all media and formats whether now known or hereafter created, and to make technical modifications necessary to do so. The Licensor waives and/or agrees not to assert any right or authority to forbid You from making technical modifications necessary to exercise the Licensed Rights, including technical modifications necessary to circumvent Effective Technological Measures. For purposes of this Public License, simply making modifications authorized by this Section 2(a)(4) never produces Adapted Material.
56
+
57
+ 5. Downstream recipients.
58
+
59
+ A. Offer from the Licensor – Licensed Material. Every recipient of the Licensed Material automatically receives an offer from the Licensor to exercise the Licensed Rights under the terms and conditions of this Public License.
60
+
61
+ B. No downstream restrictions. You may not offer or impose any additional or different terms or conditions on, or apply any Effective Technological Measures to, the Licensed Material if doing so restricts exercise of the Licensed Rights by any recipient of the Licensed Material.
62
+
63
+ 6. No endorsement. Nothing in this Public License constitutes or may be construed as permission to assert or imply that You are, or that Your use of the Licensed Material is, connected with, or sponsored, endorsed, or granted official status by, the Licensor or others designated to receive attribution as provided in Section 3(a)(1)(A)(i).
64
+
65
+ b. Other rights.
66
+
67
+ 1. Moral rights, such as the right of integrity, are not licensed under this Public License, nor are publicity, privacy, and/or other similar personality rights; however, to the extent possible, the Licensor waives and/or agrees not to assert any such rights held by the Licensor to the limited extent necessary to allow You to exercise the Licensed Rights, but not otherwise.
68
+
69
+ 2. Patent and trademark rights are not licensed under this Public License.
70
+
71
+ 3. To the extent possible, the Licensor waives any right to collect royalties from You for the exercise of the Licensed Rights, whether directly or through a collecting society under any voluntary or waivable statutory or compulsory licensing scheme. In all other cases the Licensor expressly reserves any right to collect such royalties.
72
+
73
+ Section 3 – License Conditions.
74
+
75
+ Your exercise of the Licensed Rights is expressly made subject to the following conditions.
76
+
77
+ a. Attribution.
78
+
79
+ 1. If You Share the Licensed Material (including in modified form), You must:
80
+
81
+ A. retain the following if it is supplied by the Licensor with the Licensed Material:
82
+
83
+ i. identification of the creator(s) of the Licensed Material and any others designated to receive attribution, in any reasonable manner requested by the Licensor (including by pseudonym if designated);
84
+
85
+ ii. a copyright notice;
86
+
87
+ iii. a notice that refers to this Public License;
88
+
89
+ iv. a notice that refers to the disclaimer of warranties;
90
+
91
+ v. a URI or hyperlink to the Licensed Material to the extent reasonably practicable;
92
+
93
+ B. indicate if You modified the Licensed Material and retain an indication of any previous modifications; and
94
+
95
+ C. indicate the Licensed Material is licensed under this Public License, and include the text of, or the URI or hyperlink to, this Public License.
96
+
97
+ 2. You may satisfy the conditions in Section 3(a)(1) in any reasonable manner based on the medium, means, and context in which You Share the Licensed Material. For example, it may be reasonable to satisfy the conditions by providing a URI or hyperlink to a resource that includes the required information.
98
+
99
+ 3. If requested by the Licensor, You must remove any of the information required by Section 3(a)(1)(A) to the extent reasonably practicable.
100
+
101
+ 4. If You Share Adapted Material You produce, the Adapter's License You apply must not prevent recipients of the Adapted Material from complying with this Public License.
102
+
103
+ Section 4 – Sui Generis Database Rights.
104
+
105
+ Where the Licensed Rights include Sui Generis Database Rights that apply to Your use of the Licensed Material:
106
+
107
+ a. for the avoidance of doubt, Section 2(a)(1) grants You the right to extract, reuse, reproduce, and Share all or a substantial portion of the contents of the database;
108
+
109
+ b. if You include all or a substantial portion of the database contents in a database in which You have Sui Generis Database Rights, then the database in which You have Sui Generis Database Rights (but not its individual contents) is Adapted Material; and
110
+
111
+ c. You must comply with the conditions in Section 3(a) if You Share all or a substantial portion of the contents of the database.
112
+ For the avoidance of doubt, this Section 4 supplements and does not replace Your obligations under this Public License where the Licensed Rights include other Copyright and Similar Rights.
113
+
114
+ Section 5 – Disclaimer of Warranties and Limitation of Liability.
115
+
116
+ a. Unless otherwise separately undertaken by the Licensor, to the extent possible, the Licensor offers the Licensed Material as-is and as-available, and makes no representations or warranties of any kind concerning the Licensed Material, whether express, implied, statutory, or other. This includes, without limitation, warranties of title, merchantability, fitness for a particular purpose, non-infringement, absence of latent or other defects, accuracy, or the presence or absence of errors, whether or not known or discoverable. Where disclaimers of warranties are not allowed in full or in part, this disclaimer may not apply to You.
117
+
118
+ b. To the extent possible, in no event will the Licensor be liable to You on any legal theory (including, without limitation, negligence) or otherwise for any direct, special, indirect, incidental, consequential, punitive, exemplary, or other losses, costs, expenses, or damages arising out of this Public License or use of the Licensed Material, even if the Licensor has been advised of the possibility of such losses, costs, expenses, or damages. Where a limitation of liability is not allowed in full or in part, this limitation may not apply to You.
119
+
120
+ c. The disclaimer of warranties and limitation of liability provided above shall be interpreted in a manner that, to the extent possible, most closely approximates an absolute disclaimer and waiver of all liability.
121
+
122
+ Section 6 – Term and Termination.
123
+
124
+ a. This Public License applies for the term of the Copyright and Similar Rights licensed here. However, if You fail to comply with this Public License, then Your rights under this Public License terminate automatically.
125
+
126
+ b. Where Your right to use the Licensed Material has terminated under Section 6(a), it reinstates:
127
+
128
+ 1. automatically as of the date the violation is cured, provided it is cured within 30 days of Your discovery of the violation; or
129
+
130
+ 2. upon express reinstatement by the Licensor.
131
+
132
+ c. For the avoidance of doubt, this Section 6(b) does not affect any right the Licensor may have to seek remedies for Your violations of this Public License.
133
+
134
+ d. For the avoidance of doubt, the Licensor may also offer the Licensed Material under separate terms or conditions or stop distributing the Licensed Material at any time; however, doing so will not terminate this Public License.
135
+
136
+ e. Sections 1, 5, 6, 7, and 8 survive termination of this Public License.
137
+
138
+ Section 7 – Other Terms and Conditions.
139
+
140
+ a. The Licensor shall not be bound by any additional or different terms or conditions communicated by You unless expressly agreed.
141
+
142
+ b. Any arrangements, understandings, or agreements regarding the Licensed Material not stated herein are separate from and independent of the terms and conditions of this Public License.
143
+
144
+ Section 8 – Interpretation.
145
+
146
+ a. For the avoidance of doubt, this Public License does not, and shall not be interpreted to, reduce, limit, restrict, or impose conditions on any use of the Licensed Material that could lawfully be made without permission under this Public License.
147
+
148
+ b. To the extent possible, if any provision of this Public License is deemed unenforceable, it shall be automatically reformed to the minimum extent necessary to make it enforceable. If the provision cannot be reformed, it shall be severed from this Public License without affecting the enforceability of the remaining terms and conditions.
149
+
150
+ c. No term or condition of this Public License will be waived and no failure to comply consented to unless expressly agreed to by the Licensor.
151
+
152
+ d. Nothing in this Public License constitutes or may be interpreted as a limitation upon, or waiver of, any privileges and immunities that apply to the Licensor or You, including from the legal processes of any jurisdiction or authority.
153
+
154
+ Creative Commons is not a party to its public licenses. Notwithstanding, Creative Commons may elect to apply one of its public licenses to material it publishes and in those instances will be considered the “Licensor.” Except for the limited purpose of indicating that material is shared under a Creative Commons public license or as otherwise permitted by the Creative Commons policies published at creativecommons.org/policies, Creative Commons does not authorize the use of the trademark “Creative Commons” or any other trademark or logo of Creative Commons without its prior written consent including, without limitation, in connection with any unauthorized modifications to any of its public licenses or any other arrangements, understandings, or agreements concerning use of licensed material. For the avoidance of doubt, this paragraph does not form part of the public licenses.
155
+
156
+ Creative Commons may be contacted at creativecommons.org.
LICENSE-OPENJEV ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ OpenJev attribution
2
+
3
+ AlexWortega/openjev is designated MIT by the upstream model card.
4
+ Source revision a20448012c213128955ca0c693e7c943865cab77.
5
+ https://huggingface.co/AlexWortega/openjev
6
+
7
+ MIT License
8
+
9
+ Permission is hereby granted, free of charge, to any person obtaining a copy
10
+ of this software and associated documentation files (the "Software"), to deal
11
+ in the Software without restriction, including without limitation the rights
12
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
13
+ copies of the Software, and to permit persons to whom the Software is
14
+ furnished to do so, subject to the following conditions:
15
+
16
+ The above copyright notice and this permission notice shall be included in
17
+ all copies or substantial portions of the Software.
18
+
19
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
20
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
21
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
22
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
23
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
24
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25
+ THE SOFTWARE.
LICENSE-QWEN ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ Apache License
3
+ Version 2.0, January 2004
4
+ http://www.apache.org/licenses/
5
+
6
+ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
7
+
8
+ 1. Definitions.
9
+
10
+ "License" shall mean the terms and conditions for use, reproduction,
11
+ and distribution as defined by Sections 1 through 9 of this document.
12
+
13
+ "Licensor" shall mean the copyright owner or entity authorized by
14
+ the copyright owner that is granting the License.
15
+
16
+ "Legal Entity" shall mean the union of the acting entity and all
17
+ other entities that control, are controlled by, or are under common
18
+ control with that entity. For the purposes of this definition,
19
+ "control" means (i) the power, direct or indirect, to cause the
20
+ direction or management of such entity, whether by contract or
21
+ otherwise, or (ii) ownership of fifty percent (50%) or more of the
22
+ outstanding shares, or (iii) beneficial ownership of such entity.
23
+
24
+ "You" (or "Your") shall mean an individual or Legal Entity
25
+ exercising permissions granted by this License.
26
+
27
+ "Source" form shall mean the preferred form for making modifications,
28
+ including but not limited to software source code, documentation
29
+ source, and configuration files.
30
+
31
+ "Object" form shall mean any form resulting from mechanical
32
+ transformation or translation of a Source form, including but
33
+ not limited to compiled object code, generated documentation,
34
+ and conversions to other media types.
35
+
36
+ "Work" shall mean the work of authorship, whether in Source or
37
+ Object form, made available under the License, as indicated by a
38
+ copyright notice that is included in or attached to the work
39
+ (an example is provided in the Appendix below).
40
+
41
+ "Derivative Works" shall mean any work, whether in Source or Object
42
+ form, that is based on (or derived from) the Work and for which the
43
+ editorial revisions, annotations, elaborations, or other modifications
44
+ represent, as a whole, an original work of authorship. For the purposes
45
+ of this License, Derivative Works shall not include works that remain
46
+ separable from, or merely link (or bind by name) to the interfaces of,
47
+ the Work and Derivative Works thereof.
48
+
49
+ "Contribution" shall mean any work of authorship, including
50
+ the original version of the Work and any modifications or additions
51
+ to that Work or Derivative Works thereof, that is intentionally
52
+ submitted to Licensor for inclusion in the Work by the copyright owner
53
+ or by an individual or Legal Entity authorized to submit on behalf of
54
+ the copyright owner. For the purposes of this definition, "submitted"
55
+ means any form of electronic, verbal, or written communication sent
56
+ to the Licensor or its representatives, including but not limited to
57
+ communication on electronic mailing lists, source code control systems,
58
+ and issue tracking systems that are managed by, or on behalf of, the
59
+ Licensor for the purpose of discussing and improving the Work, but
60
+ excluding communication that is conspicuously marked or otherwise
61
+ designated in writing by the copyright owner as "Not a Contribution."
62
+
63
+ "Contributor" shall mean Licensor and any individual or Legal Entity
64
+ on behalf of whom a Contribution has been received by Licensor and
65
+ subsequently incorporated within the Work.
66
+
67
+ 2. Grant of Copyright License. Subject to the terms and conditions of
68
+ this License, each Contributor hereby grants to You a perpetual,
69
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
70
+ copyright license to reproduce, prepare Derivative Works of,
71
+ publicly display, publicly perform, sublicense, and distribute the
72
+ Work and such Derivative Works in Source or Object form.
73
+
74
+ 3. Grant of Patent License. Subject to the terms and conditions of
75
+ this License, each Contributor hereby grants to You a perpetual,
76
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
77
+ (except as stated in this section) patent license to make, have made,
78
+ use, offer to sell, sell, import, and otherwise transfer the Work,
79
+ where such license applies only to those patent claims licensable
80
+ by such Contributor that are necessarily infringed by their
81
+ Contribution(s) alone or by combination of their Contribution(s)
82
+ with the Work to which such Contribution(s) was submitted. If You
83
+ institute patent litigation against any entity (including a
84
+ cross-claim or counterclaim in a lawsuit) alleging that the Work
85
+ or a Contribution incorporated within the Work constitutes direct
86
+ or contributory patent infringement, then any patent licenses
87
+ granted to You under this License for that Work shall terminate
88
+ as of the date such litigation is filed.
89
+
90
+ 4. Redistribution. You may reproduce and distribute copies of the
91
+ Work or Derivative Works thereof in any medium, with or without
92
+ modifications, and in Source or Object form, provided that You
93
+ meet the following conditions:
94
+
95
+ (a) You must give any other recipients of the Work or
96
+ Derivative Works a copy of this License; and
97
+
98
+ (b) You must cause any modified files to carry prominent notices
99
+ stating that You changed the files; and
100
+
101
+ (c) You must retain, in the Source form of any Derivative Works
102
+ that You distribute, all copyright, patent, trademark, and
103
+ attribution notices from the Source form of the Work,
104
+ excluding those notices that do not pertain to any part of
105
+ the Derivative Works; and
106
+
107
+ (d) If the Work includes a "NOTICE" text file as part of its
108
+ distribution, then any Derivative Works that You distribute must
109
+ include a readable copy of the attribution notices contained
110
+ within such NOTICE file, excluding those notices that do not
111
+ pertain to any part of the Derivative Works, in at least one
112
+ of the following places: within a NOTICE text file distributed
113
+ as part of the Derivative Works; within the Source form or
114
+ documentation, if provided along with the Derivative Works; or,
115
+ within a display generated by the Derivative Works, if and
116
+ wherever such third-party notices normally appear. The contents
117
+ of the NOTICE file are for informational purposes only and
118
+ do not modify the License. You may add Your own attribution
119
+ notices within Derivative Works that You distribute, alongside
120
+ or as an addendum to the NOTICE text from the Work, provided
121
+ that such additional attribution notices cannot be construed
122
+ as modifying the License.
123
+
124
+ You may add Your own copyright statement to Your modifications and
125
+ may provide additional or different license terms and conditions
126
+ for use, reproduction, or distribution of Your modifications, or
127
+ for any such Derivative Works as a whole, provided Your use,
128
+ reproduction, and distribution of the Work otherwise complies with
129
+ the conditions stated in this License.
130
+
131
+ 5. Submission of Contributions. Unless You explicitly state otherwise,
132
+ any Contribution intentionally submitted for inclusion in the Work
133
+ by You to the Licensor shall be under the terms and conditions of
134
+ this License, without any additional terms or conditions.
135
+ Notwithstanding the above, nothing herein shall supersede or modify
136
+ the terms of any separate license agreement you may have executed
137
+ with Licensor regarding such Contributions.
138
+
139
+ 6. Trademarks. This License does not grant permission to use the trade
140
+ names, trademarks, service marks, or product names of the Licensor,
141
+ except as required for reasonable and customary use in describing the
142
+ origin of the Work and reproducing the content of the NOTICE file.
143
+
144
+ 7. Disclaimer of Warranty. Unless required by applicable law or
145
+ agreed to in writing, Licensor provides the Work (and each
146
+ Contributor provides its Contributions) on an "AS IS" BASIS,
147
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
148
+ implied, including, without limitation, any warranties or conditions
149
+ of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
150
+ PARTICULAR PURPOSE. You are solely responsible for determining the
151
+ appropriateness of using or redistributing the Work and assume any
152
+ risks associated with Your exercise of permissions under this License.
153
+
154
+ 8. Limitation of Liability. In no event and under no legal theory,
155
+ whether in tort (including negligence), contract, or otherwise,
156
+ unless required by applicable law (such as deliberate and grossly
157
+ negligent acts) or agreed to in writing, shall any Contributor be
158
+ liable to You for damages, including any direct, indirect, special,
159
+ incidental, or consequential damages of any character arising as a
160
+ result of this License or out of the use or inability to use the
161
+ Work (including but not limited to damages for loss of goodwill,
162
+ work stoppage, computer failure or malfunction, or any and all
163
+ other commercial damages or losses), even if such Contributor
164
+ has been advised of the possibility of such damages.
165
+
166
+ 9. Accepting Warranty or Additional Liability. While redistributing
167
+ the Work or Derivative Works thereof, You may choose to offer,
168
+ and charge a fee for, acceptance of support, warranty, indemnity,
169
+ or other liability obligations and/or rights consistent with this
170
+ License. However, in accepting such obligations, You may act only
171
+ on Your own behalf and on Your sole responsibility, not on behalf
172
+ of any other Contributor, and only if You agree to indemnify,
173
+ defend, and hold each Contributor harmless for any liability
174
+ incurred by, or claims asserted against, such Contributor by reason
175
+ of your accepting any such warranty or additional liability.
176
+
177
+ END OF TERMS AND CONDITIONS
178
+
179
+ APPENDIX: How to apply the Apache License to your work.
180
+
181
+ To apply the Apache License to your work, attach the following
182
+ boilerplate notice, with the fields enclosed by brackets "[]"
183
+ replaced with your own identifying information. (Don't include
184
+ the brackets!) The text should be enclosed in the appropriate
185
+ comment syntax for the file format. We also recommend that a
186
+ file or class name and description of purpose be included on the
187
+ same "printed page" as the copyright notice for easier
188
+ identification within third-party archives.
189
+
190
+ Copyright 2026 Alibaba Cloud
191
+
192
+ Licensed under the Apache License, Version 2.0 (the "License");
193
+ you may not use this file except in compliance with the License.
194
+ You may obtain a copy of the License at
195
+
196
+ http://www.apache.org/licenses/LICENSE-2.0
197
+
198
+ Unless required by applicable law or agreed to in writing, software
199
+ distributed under the License is distributed on an "AS IS" BASIS,
200
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
201
+ See the License for the specific language governing permissions and
202
+ limitations under the License.
NOTICE ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ SkillScreen
2
+ Copyright (c) 2026 the respective SkillScreen contributors.
3
+
4
+ Research paper:
5
+ SkillScreen: High-Recall Screening of Malicious Agent Skills with Decision Models
6
+ Jerrold Huang, Junwei Quan, Saiyue Lyu, Zihan Guo, Yingqiang Ge.
7
+ https://arxiv.org/abs/2610.17727
8
+
9
+ Project: https://jerroldh.github.io/SkillScreen-webpage/
10
+ This release: https://github.com/JerroldH/SkillScreen
11
+
12
+ See README.md for the material covered by Apache-2.0 and CC BY 4.0.
13
+ Third-party attribution and license notices are in THIRD_PARTY_NOTICES.md.
README.md ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: transformers
3
+ pipeline_tag: text-classification
4
+ license: apache-2.0
5
+ base_model: AlexWortega/openjev
6
+ base_model_relation: finetune
7
+ datasets:
8
+ - SkillScreen/SkillScreen-Data
9
+ tags:
10
+ - skillscreen
11
+ - agent-security
12
+ - decision-model
13
+ - nli
14
+ - arxiv:2610.17727
15
+ ---
16
+
17
+ # SkillScreen-2B-Markdown
18
+
19
+ **Deployment release v0.1.0 · 8K context**
20
+
21
+ SkillScreen-2B-Markdown screens AI-agent skills directly from `SKILL.md`. It returns a maliciousness score and a review decision in one forward pass. Complete merged weights, the original LoRA adapter and frozen operating points are included.
22
+
23
+ [Package model](https://huggingface.co/SkillScreen/SkillScreen-2B-Package) · [Markdown model](https://huggingface.co/SkillScreen/SkillScreen-2B-Markdown) · [Data](https://huggingface.co/datasets/SkillScreen/SkillScreen-Data) · [Code](https://github.com/JerroldH/SkillScreen) · [Paper](https://arxiv.org/abs/2610.17727)
24
+
25
+ ## Run locally
26
+
27
+ Use Linux, Python 3.12 and a CUDA GPU with BF16 support. The bundled runtime and full context limit are verified on an A100 80GB.
28
+
29
+ ```bash
30
+ python -m pip install huggingface_hub==1.33.0
31
+ hf download SkillScreen/SkillScreen-2B-Markdown --revision v0.1.0 --local-dir SkillScreen-2B-Markdown
32
+ python -m pip install -r SkillScreen-2B-Markdown/requirements/inference.txt
33
+ python SkillScreen-2B-Markdown/scan.py path/to/SKILL.md --output score.json
34
+ ```
35
+
36
+ For a bundled example, replace the input path with `SkillScreen-2B-Markdown/examples/local-inventory/SKILL.md`. The scanner reads files as text and never executes them. Scoring runs locally. `--inspect` lists input files without loading the model.
37
+
38
+ The default is the **native 0.5** threshold. Use `--operating-point high-recall` for the frozen 99% OOF recall threshold, approximately **0.011642**. `--operating-point native` selects 0.5. Full-precision thresholds are loaded from [deployment_settings.json](deployment_settings.json). Decisions compare raw entailment log-odds, avoiding rounded threshold values.
39
+
40
+ ## Out-of-fold results
41
+
42
+ Three-fold OOF evaluation over **13,514 skills**, including 7,500 malicious and 6,014 benign examples.
43
+
44
+ | Operating point | Score threshold | Recall | Benign FPR | Precision | F1 |
45
+ | --- | ---: | ---: | ---: | ---: | ---: |
46
+ | Native 0.5 | 0.500000 | 96.96% | 0.88% | 99.28% | 98.10% |
47
+ | 99% OOF recall | 0.011642 | 99.00% | 15.71% | 88.71% | 93.57% |
48
+
49
+ **AUPRC (average precision) 0.9966 · AUROC 0.9946.** [Per-skill OOF predictions](evaluation-oof.csv) and [full metrics](evaluation.json) are included.
50
+
51
+ Each OOF score comes from a fold model trained without that skill's content group. Thresholds were selected on these pooled predictions and transferred to the final model trained on all eligible records. These results describe the OOF procedure, not an independent test of the final refit. Recall and FPR can change on new collections.
52
+
53
+ ## Training and input
54
+
55
+ Initialized from [OpenJev 2B](https://huggingface.co/AlexWortega/openjev/tree/a20448012c213128955ca0c693e7c943865cab77/qwen3.5-2b-nli-v5), then trained for one epoch with seed 42, LoRA rank 8 and a trainable NLI classification head. The final weights use all 13,514 eligible records from [SkillScreen-Data at the pinned revision](https://huggingface.co/datasets/SkillScreen/SkillScreen-Data/tree/d59aabbd7e0e8d95b464a2d335f36e948577b832). Labels are inherited from upstream benchmarks. This deployment release is separate from the paper's train-only experiment checkpoints.
56
+
57
+ The fixed hypothesis is “This AI agent skill is malicious.” The score is the entailment softmax output, not a separately calibrated probability. The input is the original Markdown paired with the fixed hypothesis. The report records token truncation. Inputs above 8,192 tokens are truncated while retaining the hypothesis. A below-threshold score means no model alert, not a safety certification.
58
+
59
+ [Training configuration](training_config.json) · [Adapter](adapter/) · [Release checksums](checksums.json)
60
+
61
+ ## Citation and license
62
+
63
+ Please cite [SkillScreen](https://arxiv.org/abs/2610.17727) for the method and identify this deployment release when reporting results. Machine-readable citation details are in [CITATION.cff](CITATION.cff).
64
+
65
+ Jerrold Huang, Junwei Quan, Saiyue Lyu, Zihan Guo and Yingqiang Ge. *SkillScreen: High-Recall Screening of Malicious Agent Skills with Decision Models.* 2026.
66
+
67
+ Weights, code and configuration are [Apache-2.0](LICENSE) for rights held by the contributors. Documentation and predictions are [CC BY 4.0](LICENSE-CC-BY-4.0). Upstream model and dataset terms remain applicable. See [third-party notices](THIRD_PARTY_NOTICES.md).
THIRD_PARTY_NOTICES.md ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Third-party notices
2
+
3
+ The released model is derived from Alex Wortega's OpenJev 2B checkpoint
4
+ `AlexWortega/openjev/qwen3.5-2b-nli-v5` at revision
5
+ `a20448012c213128955ca0c693e7c943865cab77`, based on Qwen3.5-2B.
6
+ OpenJev's MIT notice and Qwen's Apache-2.0 notice are retained in
7
+ LICENSE-OPENJEV and LICENSE-QWEN.
8
+
9
+ Training uses SkillScreen/SkillScreen-Data at revision
10
+ `d59aabbd7e0e8d95b464a2d335f36e948577b832`. The dataset combines public benchmark
11
+ records and recovered skill files. Its upstream labels, source terms and
12
+ attribution are documented in the dataset card.
13
+
14
+ - MaliciousSkillBench, ProtectSkills. Metadata and annotations are CC BY 4.0.
15
+ https://huggingface.co/datasets/ProtectSkills/MaliciousSkillBench
16
+ - MalSkillBench benign additions, lxyeternal/MalSkillBench. Source notice states
17
+ academic research use only. https://github.com/lxyeternal/MalSkillBench
18
+ - Package recovery includes SkillTrustBench (CC BY-NC-SA 4.0), SkillFortifyBench
19
+ (MIT), and source-specific benchmark archives. See the dataset card for
20
+ revisions, coverage, attribution and per-record terms.
21
+
22
+ Training data and upstream packages are not relicensed by this model release.
23
+ The model's Apache-2.0 license applies to rights held by the SkillScreen
24
+ contributors and does not grant additional rights to upstream material.
25
+
26
+ Documentation and released OOF predictions are CC BY 4.0. Project-authored
27
+ code examples are Apache-2.0. Python dependencies retain their own licenses.
adapter/README.md ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # LoRA adapter and classification head
2
+
3
+ Complete merged weights are at the repository root. This folder retains the
4
+ original adapter for research and further fine-tuning. Install PEFT 0.21.2 to use it.
5
+
6
+ Load the pinned **2B subfolder**, not the OpenJev repository root:
7
+
8
+ ```python
9
+ import torch
10
+ from huggingface_hub import snapshot_download
11
+ from transformers import AutoModelForSequenceClassification
12
+ from peft import PeftModel
13
+
14
+ base_dir = snapshot_download(
15
+ "AlexWortega/openjev", revision="a20448012c213128955ca0c693e7c943865cab77",
16
+ allow_patterns="qwen3.5-2b-nli-v5/*")
17
+ base = AutoModelForSequenceClassification.from_pretrained(
18
+ base_dir + "/qwen3.5-2b-nli-v5", dtype=torch.bfloat16,
19
+ attn_implementation="sdpa", trust_remote_code=False)
20
+ model = PeftModel.from_pretrained(base, "SkillScreen/SkillScreen-2B-Markdown",
21
+ subfolder="adapter", revision="v0.1.0")
22
+ ```
23
+
24
+ Input formatting and frozen thresholds are in the root inference entry point.
25
+ Adapter weights are byte-identical to the final training export. Its config uses
26
+ the pinned Hub ID instead of the training machine's local base path.
adapter/adapter_config.json ADDED
@@ -0,0 +1,52 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "AlexWortega/openjev",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "kasa_config": null,
16
+ "layer_replication": null,
17
+ "layers_pattern": null,
18
+ "layers_to_transform": null,
19
+ "loftq_config": {},
20
+ "lora_alpha": 16,
21
+ "lora_bias": false,
22
+ "lora_dropout": 0.0,
23
+ "lora_ga_config": null,
24
+ "megatron_config": null,
25
+ "megatron_core": "megatron.core",
26
+ "modules_to_save": [
27
+ "score",
28
+ "classifier",
29
+ "score"
30
+ ],
31
+ "monteclora_config": null,
32
+ "peft_type": "LORA",
33
+ "peft_version": "0.21.2",
34
+ "qalora_group_size": 16,
35
+ "r": 8,
36
+ "rank_pattern": {},
37
+ "revision": "a20448012c213128955ca0c693e7c943865cab77",
38
+ "target_modules": [
39
+ "q_proj",
40
+ "out_proj",
41
+ "in_proj_qkv",
42
+ "v_proj"
43
+ ],
44
+ "target_parameters": null,
45
+ "task_type": "SEQ_CLS",
46
+ "trainable_token_indices": null,
47
+ "use_bdlora": null,
48
+ "use_dora": false,
49
+ "use_qalora": false,
50
+ "use_rslora": false,
51
+ "velora_config": null
52
+ }
adapter/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:da4424a0b6091820866e767769d82bca1c717bd7f488b484f766353d6ca6f48b
3
+ size 8775920
chat_template.jinja ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- if tools and tools is iterable and tools is not mapping %}
46
+ {{- '<|im_start|>system\n' }}
47
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
48
+ {%- for tool in tools %}
49
+ {{- "\n" }}
50
+ {{- tool | tojson }}
51
+ {%- endfor %}
52
+ {{- "\n</tools>" }}
53
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
54
+ {%- if messages[0].role == 'system' %}
55
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
56
+ {%- if content %}
57
+ {{- '\n\n' + content }}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {{- '<|im_end|>\n' }}
61
+ {%- else %}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
65
+ {%- endif %}
66
+ {%- endif %}
67
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
68
+ {%- for message in messages[::-1] %}
69
+ {%- set index = (messages|length - 1) - loop.index0 %}
70
+ {%- if ns.multi_step_tool and message.role == "user" %}
71
+ {%- set content = render_content(message.content, false)|trim %}
72
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
73
+ {%- set ns.multi_step_tool = false %}
74
+ {%- set ns.last_query_index = index %}
75
+ {%- endif %}
76
+ {%- endif %}
77
+ {%- endfor %}
78
+ {%- if ns.multi_step_tool %}
79
+ {{- raise_exception('No user query found in messages.') }}
80
+ {%- endif %}
81
+ {%- for message in messages %}
82
+ {%- set content = render_content(message.content, true)|trim %}
83
+ {%- if message.role == "system" %}
84
+ {%- if not loop.first %}
85
+ {{- raise_exception('System message must be at the beginning.') }}
86
+ {%- endif %}
87
+ {%- elif message.role == "user" %}
88
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
89
+ {%- elif message.role == "assistant" %}
90
+ {%- set reasoning_content = '' %}
91
+ {%- if message.reasoning_content is string %}
92
+ {%- set reasoning_content = message.reasoning_content %}
93
+ {%- else %}
94
+ {%- if '</think>' in content %}
95
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
96
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
97
+ {%- endif %}
98
+ {%- endif %}
99
+ {%- set reasoning_content = reasoning_content|trim %}
100
+ {%- if loop.index0 > ns.last_query_index %}
101
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
102
+ {%- else %}
103
+ {{- '<|im_start|>' + message.role + '\n' + content }}
104
+ {%- endif %}
105
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
106
+ {%- for tool_call in message.tool_calls %}
107
+ {%- if tool_call.function is defined %}
108
+ {%- set tool_call = tool_call.function %}
109
+ {%- endif %}
110
+ {%- if loop.first %}
111
+ {%- if content|trim %}
112
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
113
+ {%- else %}
114
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
115
+ {%- endif %}
116
+ {%- else %}
117
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
118
+ {%- endif %}
119
+ {%- if tool_call.arguments is defined %}
120
+ {%- for args_name, args_value in tool_call.arguments|items %}
121
+ {{- '<parameter=' + args_name + '>\n' }}
122
+ {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
123
+ {{- args_value }}
124
+ {{- '\n</parameter>\n' }}
125
+ {%- endfor %}
126
+ {%- endif %}
127
+ {{- '</function>\n</tool_call>' }}
128
+ {%- endfor %}
129
+ {%- endif %}
130
+ {{- '<|im_end|>\n' }}
131
+ {%- elif message.role == "tool" %}
132
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
133
+ {{- '<|im_start|>user' }}
134
+ {%- endif %}
135
+ {{- '\n<tool_response>\n' }}
136
+ {{- content }}
137
+ {{- '\n</tool_response>' }}
138
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
139
+ {{- '<|im_end|>\n' }}
140
+ {%- elif loop.last %}
141
+ {{- '<|im_end|>\n' }}
142
+ {%- endif %}
143
+ {%- else %}
144
+ {{- raise_exception('Unexpected message role.') }}
145
+ {%- endif %}
146
+ {%- endfor %}
147
+ {%- if add_generation_prompt %}
148
+ {{- '<|im_start|>assistant\n' }}
149
+ {%- if enable_thinking is defined and enable_thinking is true %}
150
+ {{- '<think>\n' }}
151
+ {%- else %}
152
+ {{- '<think>\n\n</think>\n\n' }}
153
+ {%- endif %}
154
+ {%- endif %}
checksums.json ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "CITATION.cff": "94c19278673b73108c6e5a50a2ec061907a183e0bc53ab4a8bcb30838312e6d0",
3
+ "LICENSE": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
4
+ "LICENSE-CC-BY-4.0": "d557539df68e771cc1eedcc91d13f70fca930e508d11eedcafa4b15db49e3744",
5
+ "LICENSE-OPENJEV": "e6ff2f45f5a9a0756196d24ec68714b0375dc5fec0bc466dcc7dce02f442706d",
6
+ "LICENSE-QWEN": "bbedc3fda3305820b977265f01b8619d87570a6739de3a5582c3464840f1e57a",
7
+ "NOTICE": "38da69918d844538eae2cd8e99f11bb33719ca62dcb4b8c5309d95a6b6142f62",
8
+ "README.md": "e2db801d1d277191ea60bb259a1f19d3d6505d6d3e0df1a229172faebef93361",
9
+ "THIRD_PARTY_NOTICES.md": "0d1a7dc90ca95032098e7fa2662e126f7fcf794390043ed13e656c963adadb44",
10
+ "adapter/README.md": "566a19b71ec5645c1701cabd4efb738091c9d4410ff9bce5d53db4bd1f893672",
11
+ "adapter/adapter_config.json": "08d3ecae6fb926088e4db85c5232be1dc84e554b9ce65ce4ab9970f528521630",
12
+ "adapter/adapter_model.safetensors": "da4424a0b6091820866e767769d82bca1c717bd7f488b484f766353d6ca6f48b",
13
+ "chat_template.jinja": "273d8e0e683b885071fb17e08d71e5f2a5ddfb5309756181681de4f5a1822d80",
14
+ "config.json": "2ff7404369a6c8efbef3da39842aeceaf91d5eeb053df51b38b0c68be07320e3",
15
+ "deployment_settings.json": "64280dd83da9b510ee490af73b8c565651db2c501716740abaacdb9cede345c8",
16
+ "evaluation-oof.csv": "935e97b51798f6b553da87c4cc0738510a7575c14b0fe4bc6db2726145baa888",
17
+ "evaluation.json": "0ec048e245ea30370b2fc615d1869d2cc25c4f9053f184a592f13460e5fa20bb",
18
+ "examples/local-inventory/SKILL.md": "89085ea9e08c843bd9670320ef01b7ff71db4596b4296a2ba2842b22f8d6a2e4",
19
+ "examples/local-inventory/references/output.md": "09914cee180a4e792357afb4dc14c622b14243e702ea18ff61b4e4868f5c650e",
20
+ "examples/local-inventory/requirements.txt": "53f5633bc406ce79d5224ef9656f8bef17b108298c4fdcb06c2b0e0e6d9d3e11",
21
+ "examples/local-inventory/scripts/inventory.py": "915c6a74c66a3214de7a53eda74c3cf5c0b3c3d3a6bd82f9393dff919ed00328",
22
+ "model.safetensors": "4b02bba1c37cabf85d09edd8757bd30abcf7ce57b6ba2d13c5831b4a7c7fafeb",
23
+ "requirements/inference.txt": "c8c8b41a2b26eaa19481575a17ec6d436e273ae6ef8f922bfdfcca5fbfb97094",
24
+ "scan.py": "7ec9f49ffbf57b87af5897a388e1f895161f17137e10640f826f527f09f1b1d2",
25
+ "scripts/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
26
+ "scripts/check_dependencies.py": "9e34b071a3c1b8065dabbb33f9808ec7f777f4c01434a06f180a2f50dd3d6b13",
27
+ "scripts/model_score.py": "0ee39d340954e04da9e92f8b94bc111107f1c968aad3022cb3bebaed7436139d",
28
+ "scripts/package_inputs.py": "352fbd1bfe0da57ae17fddb7fd323c123996ee72df39af517009c8a37117c000",
29
+ "scripts/skill_package.py": "de2572e85de55110883aec3fe84b59ba2525b6946ad2f7b996e306edcca525b2",
30
+ "tokenizer.json": "f399b3cd12fa270d51457bb749fb30863521e8359b8a27059c71b6c2f7d6dd6c",
31
+ "tokenizer_config.json": "1b4aa963b7ff602051283ab9d794c44e4a3b65a806668a66e2312f1b641c0caa",
32
+ "training_config.json": "fd6014851d7979ddb5fa57151fa748f36fa79c5c01d0b15f0033829f9722d543"
33
+ }
config.json ADDED
@@ -0,0 +1,119 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3_5ForSequenceClassification"
4
+ ],
5
+ "dtype": "bfloat16",
6
+ "id2label": {
7
+ "0": "contradiction",
8
+ "1": "entailment",
9
+ "2": "neutral"
10
+ },
11
+ "image_token_id": 248056,
12
+ "label2id": {
13
+ "contradiction": 0,
14
+ "entailment": 1,
15
+ "neutral": 2
16
+ },
17
+ "model_type": "qwen3_5",
18
+ "nli_template": "Premise: {premise}\nHypothesis: {hypothesis}",
19
+ "pad_token_id": 248044,
20
+ "problem_type": "single_label_classification",
21
+ "text_config": {
22
+ "attention_bias": false,
23
+ "attention_dropout": 0.0,
24
+ "attn_output_gate": true,
25
+ "bos_token_id": null,
26
+ "dtype": "bfloat16",
27
+ "eos_token_id": 248044,
28
+ "full_attention_interval": 4,
29
+ "head_dim": 256,
30
+ "hidden_act": "silu",
31
+ "hidden_size": 2048,
32
+ "initializer_range": 0.02,
33
+ "intermediate_size": 6144,
34
+ "layer_types": [
35
+ "linear_attention",
36
+ "linear_attention",
37
+ "linear_attention",
38
+ "full_attention",
39
+ "linear_attention",
40
+ "linear_attention",
41
+ "linear_attention",
42
+ "full_attention",
43
+ "linear_attention",
44
+ "linear_attention",
45
+ "linear_attention",
46
+ "full_attention",
47
+ "linear_attention",
48
+ "linear_attention",
49
+ "linear_attention",
50
+ "full_attention",
51
+ "linear_attention",
52
+ "linear_attention",
53
+ "linear_attention",
54
+ "full_attention",
55
+ "linear_attention",
56
+ "linear_attention",
57
+ "linear_attention",
58
+ "full_attention"
59
+ ],
60
+ "linear_conv_kernel_dim": 4,
61
+ "linear_key_head_dim": 128,
62
+ "linear_num_key_heads": 16,
63
+ "linear_num_value_heads": 16,
64
+ "linear_value_head_dim": 128,
65
+ "mamba_ssm_dtype": "float32",
66
+ "max_position_embeddings": 262144,
67
+ "mlp_only_layers": [],
68
+ "model_type": "qwen3_5_text",
69
+ "mtp_num_hidden_layers": 1,
70
+ "mtp_use_dedicated_embeddings": false,
71
+ "num_attention_heads": 8,
72
+ "num_hidden_layers": 24,
73
+ "num_key_value_heads": 2,
74
+ "pad_token_id": 248044,
75
+ "partial_rotary_factor": 0.25,
76
+ "rms_norm_eps": 1e-06,
77
+ "rope_parameters": {
78
+ "mrope_interleaved": true,
79
+ "mrope_section": [
80
+ 11,
81
+ 11,
82
+ 10
83
+ ],
84
+ "partial_rotary_factor": 0.25,
85
+ "rope_theta": 10000000,
86
+ "rope_type": "default"
87
+ },
88
+ "tie_word_embeddings": true,
89
+ "use_cache": false,
90
+ "vocab_size": 248320
91
+ },
92
+ "tie_word_embeddings": true,
93
+ "transformers_version": "5.17.0",
94
+ "use_cache": false,
95
+ "video_token_id": 248057,
96
+ "vision_config": {
97
+ "deepstack_visual_indexes": [],
98
+ "depth": 24,
99
+ "dtype": "bfloat16",
100
+ "hidden_act": "gelu_pytorch_tanh",
101
+ "hidden_size": 1024,
102
+ "in_channels": 3,
103
+ "initializer_range": 0.02,
104
+ "intermediate_size": 4096,
105
+ "model_type": "qwen3_5_vision",
106
+ "num_heads": 16,
107
+ "num_position_embeddings": 2304,
108
+ "out_hidden_size": 2048,
109
+ "patch_size": 16,
110
+ "rope_parameters": {
111
+ "rope_theta": 10000.0,
112
+ "rope_type": "axial"
113
+ },
114
+ "spatial_merge_size": 2,
115
+ "temporal_patch_size": 2
116
+ },
117
+ "vision_end_token_id": 248054,
118
+ "vision_start_token_id": 248053
119
+ }
deployment_settings.json ADDED
@@ -0,0 +1,133 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "training_scope": "all_data",
3
+ "seed": 42,
4
+ "input_mode": "markdown8k",
5
+ "max_length": 8192,
6
+ "hypothesis": "This AI agent skill is malicious.",
7
+ "dataset": {
8
+ "repo_id": "SkillScreen/SkillScreen-Data",
9
+ "revision": "d59aabbd7e0e8d95b464a2d335f36e948577b832"
10
+ },
11
+ "native_score": "entailment softmax",
12
+ "thresholds": {
13
+ "mode": "markdown8k",
14
+ "seed": 42,
15
+ "records": 13514,
16
+ "malicious": 7500,
17
+ "benign": 6014,
18
+ "auprc_average_precision": 0.9966339274117415,
19
+ "auroc": 0.9945986919410504,
20
+ "folds_sha256": "c11c6f68dd23fddb70d7090512d93f5be54bb216596175d44fcaa7e7df80e202",
21
+ "oof_sha256": "935e97b51798f6b553da87c4cc0738510a7575c14b0fe4bc6db2726145baa888",
22
+ "folds": [
23
+ {
24
+ "fold": 1,
25
+ "predictions_sha256": "a43bd55e2f3256b7e5d959822ce449ecee79dfc2affbb1a3c6a6fab04bbc1ec0",
26
+ "adapter_sha256": "79c2cfa9e230b5eb88dc26ae96a5f9b26ea86d6f7f5d187fbb9070b9518d7ec4"
27
+ },
28
+ {
29
+ "fold": 2,
30
+ "predictions_sha256": "608d14ae22ea7579928f0722dc4ba9a4a7dfad923a57e8542d7ad7ae132ed52e",
31
+ "adapter_sha256": "7683ceca69861929e1263dd46bf3befeb2534592b90686ad60b4abf21b2841dc"
32
+ },
33
+ {
34
+ "fold": 3,
35
+ "predictions_sha256": "a759c2e629bf2e21382806eb332e0760de792683b337f2529f019e0b730286ba",
36
+ "adapter_sha256": "3110418b5329db127917bad3135db46d97610cf126cc9ced6b08f048b7ba5100"
37
+ }
38
+ ],
39
+ "created_at": "2026-11-10T16:05:55.944354+00:00",
40
+ "default": {
41
+ "raw_threshold": 0.0,
42
+ "native_threshold": 0.5,
43
+ "tp": 7272,
44
+ "fp": 53,
45
+ "fn": 228,
46
+ "tn": 5961,
47
+ "recall": 0.9696,
48
+ "fpr": 0.008812770202859994,
49
+ "precision": 0.9927645051194539,
50
+ "f1": 0.9810455311973019,
51
+ "flagged_fraction": 0.5420304869024715
52
+ },
53
+ "targets": {
54
+ "0.95": {
55
+ "raw_threshold": 5.374836521935736,
56
+ "native_threshold": 0.9953896776695467,
57
+ "tp": 7125,
58
+ "fp": 9,
59
+ "fn": 375,
60
+ "tn": 6005,
61
+ "recall": 0.95,
62
+ "fpr": 0.0014965081476554707,
63
+ "precision": 0.9987384356602187,
64
+ "f1": 0.973759737597376,
65
+ "flagged_fraction": 0.5278969957081545
66
+ },
67
+ "0.98": {
68
+ "raw_threshold": -2.9843786119895848,
69
+ "native_threshold": 0.0481366056169061,
70
+ "tp": 7350,
71
+ "fp": 243,
72
+ "fn": 150,
73
+ "tn": 5771,
74
+ "recall": 0.98,
75
+ "fpr": 0.0404057199866977,
76
+ "precision": 0.9679968391939945,
77
+ "f1": 0.9739614390777181,
78
+ "flagged_fraction": 0.5618617729761729
79
+ },
80
+ "0.99": {
81
+ "raw_threshold": -4.441408183360328,
82
+ "native_threshold": 0.011642201918123018,
83
+ "tp": 7425,
84
+ "fp": 945,
85
+ "fn": 75,
86
+ "tn": 5069,
87
+ "recall": 0.99,
88
+ "fpr": 0.1571333555038244,
89
+ "precision": 0.8870967741935484,
90
+ "f1": 0.9357277882797732,
91
+ "flagged_fraction": 0.6193577031226876
92
+ }
93
+ },
94
+ "balanced": {
95
+ "target": 0.95,
96
+ "raw_threshold": 5.374836521935736,
97
+ "native_threshold": 0.9953896776695467,
98
+ "tp": 7125,
99
+ "fp": 9,
100
+ "fn": 375,
101
+ "tn": 6005,
102
+ "recall": 0.95,
103
+ "fpr": 0.0014965081476554707,
104
+ "precision": 0.9987384356602187,
105
+ "f1": 0.973759737597376,
106
+ "flagged_fraction": 0.5278969957081545
107
+ },
108
+ "high_recall": {
109
+ "raw_threshold": -4.441408183360328,
110
+ "native_threshold": 0.011642201918123018,
111
+ "tp": 7425,
112
+ "fp": 945,
113
+ "fn": 75,
114
+ "tn": 5069,
115
+ "recall": 0.99,
116
+ "fpr": 0.1571333555038244,
117
+ "precision": 0.8870967741935484,
118
+ "f1": 0.9357277882797732,
119
+ "flagged_fraction": 0.6193577031226876
120
+ },
121
+ "selection": "Balanced maximizes recall minus FPR among predeclared 95/98/99% OOF recall gates; ties prefer higher recall. High-recall targets 99% OOF recall.",
122
+ "scope": "Pooled three-fold out-of-fold scores. Every score is from a model that excluded that skill group. Thresholds transfer to a separate full-data refit; these are not held-out test results of that refit."
123
+ },
124
+ "metric_scope": "OOF metrics belong to the fold models, not an independent test of this final refit.",
125
+ "repo_id": "SkillScreen/SkillScreen-2B-Markdown",
126
+ "release": "v0.1.0",
127
+ "default_policy": "native",
128
+ "base_model": {
129
+ "repo_id": "AlexWortega/openjev",
130
+ "subfolder": "qwen3.5-2b-nli-v5",
131
+ "revision": "a20448012c213128955ca0c693e7c943865cab77"
132
+ }
133
+ }
evaluation-oof.csv ADDED
The diff for this file is too large to render. See raw diff
 
evaluation.json ADDED
@@ -0,0 +1,112 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "mode": "markdown8k",
3
+ "seed": 42,
4
+ "records": 13514,
5
+ "malicious": 7500,
6
+ "benign": 6014,
7
+ "auprc_average_precision": 0.9966339274117415,
8
+ "auroc": 0.9945986919410504,
9
+ "folds_sha256": "c11c6f68dd23fddb70d7090512d93f5be54bb216596175d44fcaa7e7df80e202",
10
+ "oof_sha256": "935e97b51798f6b553da87c4cc0738510a7575c14b0fe4bc6db2726145baa888",
11
+ "folds": [
12
+ {
13
+ "fold": 1,
14
+ "predictions_sha256": "a43bd55e2f3256b7e5d959822ce449ecee79dfc2affbb1a3c6a6fab04bbc1ec0",
15
+ "adapter_sha256": "79c2cfa9e230b5eb88dc26ae96a5f9b26ea86d6f7f5d187fbb9070b9518d7ec4"
16
+ },
17
+ {
18
+ "fold": 2,
19
+ "predictions_sha256": "608d14ae22ea7579928f0722dc4ba9a4a7dfad923a57e8542d7ad7ae132ed52e",
20
+ "adapter_sha256": "7683ceca69861929e1263dd46bf3befeb2534592b90686ad60b4abf21b2841dc"
21
+ },
22
+ {
23
+ "fold": 3,
24
+ "predictions_sha256": "a759c2e629bf2e21382806eb332e0760de792683b337f2529f019e0b730286ba",
25
+ "adapter_sha256": "3110418b5329db127917bad3135db46d97610cf126cc9ced6b08f048b7ba5100"
26
+ }
27
+ ],
28
+ "created_at": "2026-11-10T16:05:55.944354+00:00",
29
+ "default": {
30
+ "raw_threshold": 0.0,
31
+ "native_threshold": 0.5,
32
+ "tp": 7272,
33
+ "fp": 53,
34
+ "fn": 228,
35
+ "tn": 5961,
36
+ "recall": 0.9696,
37
+ "fpr": 0.008812770202859994,
38
+ "precision": 0.9927645051194539,
39
+ "f1": 0.9810455311973019,
40
+ "flagged_fraction": 0.5420304869024715
41
+ },
42
+ "targets": {
43
+ "0.95": {
44
+ "raw_threshold": 5.374836521935736,
45
+ "native_threshold": 0.9953896776695467,
46
+ "tp": 7125,
47
+ "fp": 9,
48
+ "fn": 375,
49
+ "tn": 6005,
50
+ "recall": 0.95,
51
+ "fpr": 0.0014965081476554707,
52
+ "precision": 0.9987384356602187,
53
+ "f1": 0.973759737597376,
54
+ "flagged_fraction": 0.5278969957081545
55
+ },
56
+ "0.98": {
57
+ "raw_threshold": -2.9843786119895848,
58
+ "native_threshold": 0.0481366056169061,
59
+ "tp": 7350,
60
+ "fp": 243,
61
+ "fn": 150,
62
+ "tn": 5771,
63
+ "recall": 0.98,
64
+ "fpr": 0.0404057199866977,
65
+ "precision": 0.9679968391939945,
66
+ "f1": 0.9739614390777181,
67
+ "flagged_fraction": 0.5618617729761729
68
+ },
69
+ "0.99": {
70
+ "raw_threshold": -4.441408183360328,
71
+ "native_threshold": 0.011642201918123018,
72
+ "tp": 7425,
73
+ "fp": 945,
74
+ "fn": 75,
75
+ "tn": 5069,
76
+ "recall": 0.99,
77
+ "fpr": 0.1571333555038244,
78
+ "precision": 0.8870967741935484,
79
+ "f1": 0.9357277882797732,
80
+ "flagged_fraction": 0.6193577031226876
81
+ }
82
+ },
83
+ "balanced": {
84
+ "target": 0.95,
85
+ "raw_threshold": 5.374836521935736,
86
+ "native_threshold": 0.9953896776695467,
87
+ "tp": 7125,
88
+ "fp": 9,
89
+ "fn": 375,
90
+ "tn": 6005,
91
+ "recall": 0.95,
92
+ "fpr": 0.0014965081476554707,
93
+ "precision": 0.9987384356602187,
94
+ "f1": 0.973759737597376,
95
+ "flagged_fraction": 0.5278969957081545
96
+ },
97
+ "high_recall": {
98
+ "raw_threshold": -4.441408183360328,
99
+ "native_threshold": 0.011642201918123018,
100
+ "tp": 7425,
101
+ "fp": 945,
102
+ "fn": 75,
103
+ "tn": 5069,
104
+ "recall": 0.99,
105
+ "fpr": 0.1571333555038244,
106
+ "precision": 0.8870967741935484,
107
+ "f1": 0.9357277882797732,
108
+ "flagged_fraction": 0.6193577031226876
109
+ },
110
+ "selection": "Balanced maximizes recall minus FPR among predeclared 95/98/99% OOF recall gates; ties prefer higher recall. High-recall targets 99% OOF recall.",
111
+ "scope": "Pooled three-fold out-of-fold scores. Every score is from a model that excluded that skill group. Thresholds transfer to a separate full-data refit; these are not held-out test results of that refit."
112
+ }
examples/local-inventory/SKILL.md ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ name: local-inventory
3
+ description: Count file extensions in a directory selected by the user.
4
+ ---
5
+
6
+ # Local inventory
7
+
8
+ Use this skill to summarize the kinds of files in one directory. Ask for the
9
+ directory to summarize if the user has not provided one.
10
+
11
+ The helper in scripts/inventory.py counts the immediate files by extension and
12
+ prints JSON. It does not read file contents, follow links, access the network, or
13
+ modify files. The output format is described in references/output.md.
14
+
15
+ This package is a small, project-authored example for SkillScreen. Its files can
16
+ be inspected as text; scanning the package does not run the helper.
examples/local-inventory/references/output.md ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ # Output
2
+
3
+ The helper prints an object whose keys are file extensions and whose values are
4
+ file counts. Files without an extension use the key "(none)". Counts concern
5
+ only immediate files in the chosen directory.
examples/local-inventory/requirements.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ # This example helper uses only the Python standard library.
examples/local-inventory/scripts/inventory.py ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Count immediate file extensions in a directory explicitly selected by the user."""
2
+ import argparse
3
+ from collections import Counter
4
+ import json
5
+ from pathlib import Path
6
+
7
+
8
+ def main():
9
+ parser = argparse.ArgumentParser(description=__doc__)
10
+ parser.add_argument('directory', type=Path)
11
+ args = parser.parse_args()
12
+ if not args.directory.is_dir():
13
+ parser.error('Choose an existing directory.')
14
+ counts = Counter(path.suffix.lower() or '(none)' for path in args.directory.iterdir()
15
+ if path.is_file() and not path.is_symlink())
16
+ print(json.dumps(dict(sorted(counts.items())), indent=2))
17
+
18
+
19
+ if __name__ == '__main__':
20
+ main()
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4b02bba1c37cabf85d09edd8757bd30abcf7ce57b6ba2d13c5831b4a7c7fafeb
3
+ size 4426571240
requirements/inference.txt ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ # Linux, Python 3.12, NVIDIA GPU with BF16 support. Verified on A100.
2
+ --extra-index-url https://download.pytorch.org/whl/cu126
3
+ torch==2.8.0+cu126
4
+ transformers==5.17.0
5
+ huggingface_hub==1.33.0
6
+ flash-linear-attention==0.3.2
scan.py ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Inspect or screen a local skill package. Python 3.12+. Sample files are read only."""
2
+ import argparse
3
+ import json
4
+ from datetime import datetime, timezone
5
+ from pathlib import Path
6
+ import sys
7
+
8
+ from scripts.skill_package import read_package, write_json
9
+
10
+
11
+ def scan(path, checks='model', mode=None, inspect_only=False, model_dir=None, operating_point='auto'):
12
+ row, coverage = read_package(path)
13
+ bundled = Path(model_dir) if model_dir else Path(__file__).resolve().parent
14
+ settings_file = bundled / 'deployment_settings.json'
15
+ if mode is None and settings_file.is_file():
16
+ saved_mode = json.loads(settings_file.read_text(encoding='utf-8'))['input_mode']
17
+ mode = {'package32k': 'package', 'markdown8k': 'markdown'}[saved_mode]
18
+ mode = mode or ('markdown' if coverage['input_kind'] == 'markdown_file' else 'package')
19
+ report = {'created_at': datetime.now(timezone.utc).isoformat(),
20
+ 'package': coverage, 'model': {'status': 'not_run'},
21
+ 'dependencies': {'status': 'not_run'}}
22
+ if inspect_only:
23
+ return report
24
+ if checks in ('dependencies', 'all'):
25
+ from scripts.check_dependencies import check_dependencies
26
+ report['dependencies'] = check_dependencies(row)
27
+ if checks in ('model', 'all'):
28
+ from scripts.model_score import score_model
29
+ try:
30
+ report['model'] = score_model(row, mode, model_dir=model_dir, operating_point=operating_point)
31
+ except (OSError, ValueError, RuntimeError, ImportError) as exc:
32
+ report['model'] = {'status': 'unavailable', 'reason': str(exc)}
33
+ return report
34
+
35
+
36
+ def print_summary(report):
37
+ coverage = report['package']
38
+ print(f"Read {coverage['text_files']} text files; skipped {len(coverage['skipped'])} entries.")
39
+ model = report['model']
40
+ if model['status'] == 'complete':
41
+ decision = 'review' if model['review'] else 'below the configured review threshold'
42
+ print(f"Model: {model['score']:.6f} / threshold {model['threshold']:.6f} ({decision}).")
43
+ print(f"Release: {model['repo_id']} @ {model['revision']}; operating point: {model['operating_point']}.")
44
+ print(f"Model input: {model['input_mode']}; {model['coverage']['used_tokens']} tokens; truncated={model['coverage']['truncated']}.")
45
+ else:
46
+ print('Model: ' + model['status'].replace('_', ' ') + '.')
47
+ if model.get('reason'):
48
+ print(model['reason'])
49
+ deps = report['dependencies']
50
+ print('Dependencies: ' + deps['status'].replace('_', ' ') + '.')
51
+ if deps['status'] != 'not_run':
52
+ print(f"Checked {len(deps['checked'])} exact pins; {len(deps['findings'])} advisory matches; {len(deps['unresolved'])} unresolved entries; {len(deps['errors'])} lookup errors.")
53
+ for finding in deps['findings']:
54
+ print(f" {finding['package']}=={finding['version']}: {finding['advisory_id']} at {finding['path']}:{finding['line']}")
55
+ print('Only declared exact PyPI pins are checked. A match is not a finding of malicious intent.')
56
+ if model['status'] == deps['status'] == 'not_run':
57
+ print('Inspection only. No security checks were run.')
58
+
59
+
60
+ def main(argv=None):
61
+ parser = argparse.ArgumentParser(description=__doc__)
62
+ parser.add_argument('path', type=Path, help='Skill directory containing SKILL.md, or a single SKILL.md')
63
+ parser.add_argument('--inspect', action='store_true', help='Offline inventory only; no models or network')
64
+ parser.add_argument('--checks', choices=('model', 'dependencies', 'all'), default='model')
65
+ parser.add_argument('--mode', choices=('package', 'markdown'), help='Defaults to package for a directory, markdown for a file')
66
+ parser.add_argument('--model-dir', type=Path, help='Use a downloaded deployment release instead of the Hub')
67
+ parser.add_argument('--operating-point', choices=('auto', 'native', 'high-recall', 'balanced'), default='auto',
68
+ help='Saved operating point; auto uses the release default, native uses 0.5')
69
+ parser.add_argument('--output', type=Path, help='Create a JSON report without raw file contents')
70
+ args = parser.parse_args(argv)
71
+ if args.output and (args.output.exists() or args.output.resolve().is_relative_to(args.path.resolve() if args.path.is_dir() else args.path.parent.resolve())):
72
+ parser.error('Choose a new output file outside the input package.')
73
+ report = scan(args.path, args.checks, args.mode, args.inspect, args.model_dir, args.operating_point)
74
+ if args.output:
75
+ write_json(args.output, report)
76
+ print_summary(report)
77
+ return 1 if any(report[key]['status'] == 'unavailable' or report[key].get('errors') for key in ('model', 'dependencies')) else 0
78
+
79
+
80
+ if __name__ == '__main__':
81
+ try:
82
+ sys.exit(main())
83
+ except (ValueError, OSError) as exc:
84
+ print(f'ERROR: {exc}', file=sys.stderr)
85
+ sys.exit(1)
scripts/__init__.py ADDED
File without changes
scripts/check_dependencies.py ADDED
@@ -0,0 +1,81 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Optional CPU check for exact PyPI pins in requirements*.txt, using OSV."""
2
+ from datetime import datetime, timezone
3
+ import json
4
+ from pathlib import PurePosixPath
5
+ import re
6
+ from urllib.request import Request, urlopen
7
+
8
+
9
+ OSV_URL = 'https://api.osv.dev/v1/query'
10
+ PIN = re.compile(r'([A-Za-z0-9][A-Za-z0-9._-]*)(?:\[[A-Za-z0-9_,.-]+\])?==([A-Za-z0-9][A-Za-z0-9.!+_-]*)')
11
+ OTHER_MANIFESTS = {'package.json', 'package-lock.json', 'yarn.lock', 'pnpm-lock.yaml',
12
+ 'composer.json', 'composer.lock', 'pyproject.toml', 'poetry.lock',
13
+ 'uv.lock', 'pipfile', 'pipfile.lock', 'dockerfile', 'go.mod', 'cargo.lock'}
14
+
15
+
16
+ def query_osv(name, version):
17
+ payload = json.dumps({'package': {'name': name, 'ecosystem': 'PyPI'}, 'version': version}).encode()
18
+ request = Request(OSV_URL, data=payload, headers={'Content-Type': 'application/json',
19
+ 'User-Agent': 'SkillScreen/0.1'})
20
+ with urlopen(request, timeout=15) as response:
21
+ raw = response.read(4_194_305)
22
+ if len(raw) > 4_194_304:
23
+ raise ValueError('OSV response exceeded 4 MiB.')
24
+ result = json.loads(raw)
25
+ if not isinstance(result, dict) or result.get('error') or not isinstance(result.get('vulns', []), list):
26
+ raise ValueError('Unexpected OSV response.')
27
+ if result.get('next_page_token'):
28
+ raise ValueError('OSV response is paginated; this check is incomplete.')
29
+ return result.get('vulns', [])
30
+
31
+
32
+ def check_dependencies(row, lookup=None):
33
+ lookup = query_osv if lookup is None else lookup
34
+ checked, unresolved, findings, errors = [], [], [], []
35
+ cache = {}
36
+ manifests = []
37
+ for item in row['files']:
38
+ name = PurePosixPath(item['path']).name.lower()
39
+ if not (name.startswith('requirements') and name.endswith('.txt')):
40
+ if name in OTHER_MANIFESTS:
41
+ unresolved.append({'path': item['path'], 'reason': 'unsupported_manifest'})
42
+ continue
43
+ manifests.append(item['path'])
44
+ for line_number, raw in enumerate(item['text'].splitlines(), 1):
45
+ line = re.split(r'\s+#', raw, maxsplit=1)[0].strip()
46
+ if not line or line.startswith('#'):
47
+ continue
48
+ location = {'path': item['path'], 'line': line_number}
49
+ match = PIN.fullmatch(line)
50
+ if not match:
51
+ unresolved.append({**location, 'reason': 'not_an_unconditional_exact_pin'})
52
+ continue
53
+ package, version = match.groups()
54
+ package = re.sub(r'[-_.]+', '-', package).lower()
55
+ key = package, version
56
+ if key not in cache:
57
+ try:
58
+ advisories = lookup(*key)
59
+ if not isinstance(advisories, list) or any(not isinstance(a, dict) or not a.get('id') for a in advisories):
60
+ raise ValueError('Invalid advisory response.')
61
+ cache[key] = [a for a in advisories if not a.get('withdrawn')]
62
+ except (OSError, ValueError, TimeoutError) as exc:
63
+ cache[key] = str(exc)
64
+ result = cache[key]
65
+ dependency = {**location, 'ecosystem': 'PyPI', 'package': package, 'version': version}
66
+ if isinstance(result, str):
67
+ errors.append({**dependency, 'reason': result})
68
+ continue
69
+ checked.append({**dependency, 'advisory_count': len(result)})
70
+ for advisory in result:
71
+ findings.append({**dependency, 'type': 'affected_dependency_version',
72
+ 'advisory_id': advisory['id'], 'aliases': advisory.get('aliases', []),
73
+ 'summary': advisory.get('summary', ''),
74
+ 'url': 'https://osv.dev/vulnerability/' + advisory['id'],
75
+ 'advisory_modified': advisory.get('modified')})
76
+ status = ('partial' if errors and checked else 'unavailable' if errors else
77
+ 'partial' if unresolved else 'complete' if manifests else 'not_applicable')
78
+ return {'status': status, 'checked_at': datetime.now(timezone.utc).isoformat(),
79
+ 'scope': 'Exact, unconditional PyPI pins in requirements*.txt; other files and transitive dependencies are not resolved.',
80
+ 'manifests': manifests, 'checked': checked, 'findings': findings,
81
+ 'unresolved': unresolved, 'errors': errors}
scripts/model_score.py ADDED
@@ -0,0 +1,107 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Local inference with versioned SkillScreen deployment models."""
2
+ import json
3
+ import math
4
+ from pathlib import Path
5
+
6
+ from .package_inputs import encode_package
7
+
8
+
9
+ ROOT = Path(__file__).resolve().parents[1]
10
+ HYPOTHESIS = 'This AI agent skill is malicious.'
11
+
12
+
13
+ def encode_input(tokenizer, row, mode, max_length):
14
+ if mode not in ('package', 'markdown'):
15
+ raise ValueError('Choose package or markdown input mode.')
16
+ if tokenizer.num_special_tokens_to_add(False) != 0:
17
+ raise ValueError('The released model requires a tokenizer without added special tokens.')
18
+ if mode == 'package':
19
+ return encode_package(tokenizer, row, HYPOTHESIS, max_length)
20
+ suffix = '\nHypothesis: ' + HYPOTHESIS
21
+ suffix_ids = tokenizer.encode(suffix, add_special_tokens=False)
22
+ if type(max_length) is not int or max_length <= len(suffix_ids) + 8:
23
+ raise ValueError('Input budget is too small for the premise and fixed hypothesis.')
24
+ ids = tokenizer.encode('Premise: ' + row['skill_text'] + suffix, add_special_tokens=False)
25
+ original = len(ids)
26
+ if original > max_length:
27
+ ids = ids[:max_length - len(suffix_ids)] + suffix_ids
28
+ return ids, original, {'original_tokens': original, 'used_tokens': len(ids),
29
+ 'truncated': original > max_length, 'files': ['SKILL.md']}
30
+
31
+
32
+ def load_release(mode, model_dir=None):
33
+ """Resolve one local release or the pinned Hub release for this input mode."""
34
+ if model_dir is None and (ROOT / 'deployment_settings.json').is_file():
35
+ model_dir = ROOT
36
+ if model_dir is not None:
37
+ source = str(Path(model_dir).resolve())
38
+ settings = json.loads((Path(source) / 'deployment_settings.json').read_text(encoding='utf-8'))
39
+ options = {'local_files_only': True, 'trust_remote_code': False}
40
+ revision = settings['release']
41
+ else:
42
+ from huggingface_hub import hf_hub_download
43
+ spec = json.loads((ROOT / 'configs/community-models.json').read_text(encoding='utf-8'))[mode]
44
+ source, revision = spec['repo_id'], spec['revision']
45
+ path = hf_hub_download(source, 'deployment_settings.json', revision=revision)
46
+ settings = json.loads(Path(path).read_text(encoding='utf-8'))
47
+ if settings['repo_id'] != source:
48
+ raise ValueError('Model identity differs from the pinned release.')
49
+ options = {'revision': revision, 'trust_remote_code': False}
50
+ if settings['input_mode'] != {'package': 'package32k', 'markdown': 'markdown8k'}[mode]:
51
+ raise ValueError('This model uses a different input mode.')
52
+ if settings['hypothesis'] != HYPOTHESIS:
53
+ raise ValueError('Unexpected hypothesis in deployment settings.')
54
+ return source, options, settings, revision
55
+
56
+
57
+ def select_threshold(settings, operating_point='auto'):
58
+ policy = settings['default_policy'] if operating_point == 'auto' else operating_point
59
+ key = {'native': 'default', 'high-recall': 'high_recall', 'balanced': 'balanced'}.get(policy)
60
+ if key is None:
61
+ raise ValueError('Unknown operating point.')
62
+ gate = settings['thresholds'][key]
63
+ raw, native = gate['raw_threshold'], gate['native_threshold']
64
+ if not all(math.isfinite(x) for x in (raw, native)) or not 0 < native < 1:
65
+ raise ValueError('Invalid saved threshold.')
66
+ if not math.isclose(raw, math.log(native) - math.log1p(-native), abs_tol=1e-9):
67
+ raise ValueError('Raw and native thresholds disagree.')
68
+ return policy, gate
69
+
70
+
71
+ def score_model(row, mode, model_dir=None, operating_point='auto'):
72
+ try:
73
+ import torch
74
+ from transformers import AutoModelForSequenceClassification, AutoTokenizer
75
+ except ImportError as exc:
76
+ raise ValueError('Model inference needs requirements/inference.txt. Package inspection and dependency checks need only Python.') from exc
77
+ if not torch.cuda.is_available() or not torch.cuda.is_bf16_supported():
78
+ raise ValueError('The released inference recipe requires a CUDA GPU with BF16 support. Use --inspect or --checks dependencies on this machine.')
79
+ source, options, settings, revision = load_release(mode, model_dir)
80
+ policy, gate = select_threshold(settings, operating_point)
81
+ torch.set_num_threads(8)
82
+ torch.manual_seed(42)
83
+ tokenizer = AutoTokenizer.from_pretrained(source, **options)
84
+ ids, _, coverage = encode_input(tokenizer, row, mode, settings['max_length'])
85
+ model = AutoModelForSequenceClassification.from_pretrained(
86
+ source, dtype=torch.bfloat16, attn_implementation='sdpa', **options)
87
+ if model.config.label2id != {'contradiction': 0, 'entailment': 1, 'neutral': 2}:
88
+ raise ValueError('Unexpected model label mapping.')
89
+ tokenizer.pad_token_id = model.config.pad_token_id
90
+ model.config.use_cache = model.config.text_config.use_cache = False
91
+ model = model.cuda().eval()
92
+ inputs = torch.tensor([ids], dtype=torch.long, device='cuda')
93
+ with torch.inference_mode(), torch.autocast('cuda', dtype=torch.bfloat16):
94
+ logits = model(input_ids=inputs, attention_mask=torch.ones_like(inputs)).logits[0].float().double()
95
+ if not torch.isfinite(logits).all():
96
+ raise ValueError('Model returned nonfinite logits.')
97
+ score = logits.softmax(-1)[1].item()
98
+ raw = (logits[1] - torch.logsumexp(logits[[0, 2]], dim=0)).item()
99
+ return {'status': 'complete', 'repo_id': settings['repo_id'], 'revision': revision,
100
+ 'release': settings['release'], 'training_scope': settings['training_scope'],
101
+ 'dataset': settings['dataset']['repo_id'], 'dataset_revision': settings['dataset']['revision'],
102
+ 'protocol': 'three_fold_oof_then_all_data_refit', 'operating_point': policy,
103
+ 'threshold_source': 'native 0.5 reference' if policy == 'native' else 'frozen three-fold OOF threshold',
104
+ 'input_mode': mode, 'score': score, 'score_kind': 'native entailment softmax',
105
+ 'raw_log_odds': raw, 'threshold': gate['native_threshold'],
106
+ 'raw_threshold': gate['raw_threshold'], 'review': raw >= gate['raw_threshold'],
107
+ 'coverage': coverage}
scripts/package_inputs.py ADDED
@@ -0,0 +1,134 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Package input recipe copied from the released SkillScreen Package model.
2
+
3
+ Only input formatting and token coverage are retained; see the HF v1.1 model.
4
+ """
5
+
6
+
7
+ from bisect import bisect_left, bisect_right
8
+ from pathlib import PurePosixPath
9
+ import re
10
+
11
+
12
+ CODE_SUFFIXES = frozenset(".py .sh .bash .zsh .fish .js .mjs .cjs .jsx .ts .tsx .rb .pl .php .ps1 .psm1 .bat .cmd .go .rs .java .c .h .cc .cpp .hpp .cs .swift .lua .r .jl .sql".split())
13
+
14
+
15
+ CONFIG_SUFFIXES = frozenset(".json .jsonc .yaml .yml .toml .ini .cfg .conf .env .xml .properties".split())
16
+
17
+
18
+ BULK_SUFFIXES = frozenset(".csv .tsv .jsonl .ndjson .geojson .log .svg .map".split())
19
+
20
+
21
+ BULK_DIRECTORIES = frozenset(("data", "assets", "output", "outputs", "reports"))
22
+
23
+
24
+ def safe_relative_path(value):
25
+ """Validate a serialized path. Never resolve, open or execute that path."""
26
+ if (not isinstance(value, str) or not value or "\\" in value or
27
+ value.startswith("/") or re.match(r"^[A-Za-z]:", value) or
28
+ any(ord(c) < 32 or ord(c) == 127 for c in value) or
29
+ any(part in ("", ".", "..") for part in value.split("/"))):
30
+ raise ValueError("Expected an inert, unambiguous relative package path")
31
+ value.encode("utf-8")
32
+ return value
33
+
34
+
35
+ def file_priority(item):
36
+ path = PurePosixPath(safe_relative_path(item["path"]))
37
+ suffix = path.suffix.lower()
38
+ parts = {p.lower() for p in path.parts[:-1]}
39
+ if parts & BULK_DIRECTORIES or suffix in BULK_SUFFIXES:
40
+ priority = 3
41
+ elif suffix in CODE_SUFFIXES or path.name.lower() in ("makefile", "dockerfile"):
42
+ priority = 0
43
+ elif suffix in CONFIG_SUFFIXES or path.name.startswith(".env"):
44
+ priority = 1
45
+ else:
46
+ priority = 2
47
+ return priority, item["path"]
48
+
49
+
50
+ def ordered_files(row):
51
+ """Order only explicit companion text files using label-independent rules."""
52
+ if not isinstance(row.get("skill_text"), str) or not isinstance(row.get("files"), list):
53
+ raise ValueError("Package records require original Markdown and companion files")
54
+ files = row["files"]
55
+ for item in files:
56
+ if not isinstance(item, dict) or set(item) != {"path", "text"} or not isinstance(item["text"], str):
57
+ raise ValueError("Companions must contain only path and inert text")
58
+ safe_relative_path(item["path"])
59
+ if item["path"].lower() == "skill.md":
60
+ raise ValueError("SKILL.md must come from the original skill_text field")
61
+ if len({item["path"] for item in files}) != len(files):
62
+ raise ValueError("Duplicate companion path")
63
+ return sorted(files, key=file_priority)
64
+
65
+
66
+ def _render(row):
67
+ chunks = ["[SKILL.md]\n", row["skill_text"]]
68
+ cursor = len(chunks[0])
69
+ spans = [{"path": "SKILL.md", "start": cursor, "end": cursor + len(row["skill_text"]), "header_start": 0}]
70
+ cursor += len(row["skill_text"])
71
+ for item in ordered_files(row):
72
+ header = "\n\n[FILE " + item["path"] + "]\n"
73
+ header_start = cursor
74
+ cursor += len(header)
75
+ spans.append({"path": item["path"], "start": cursor, "end": cursor + len(item["text"]),
76
+ "header_start": header_start})
77
+ chunks.extend((header, item["text"]))
78
+ cursor += len(item["text"])
79
+ return "".join(chunks), spans
80
+
81
+
82
+ def encode_package(tokenizer, row, hypothesis, max_length):
83
+ """Return IDs, original token count and exact retained file-token coverage.
84
+
85
+ A fast Hugging Face tokenizer is required for Unicode character offsets. The
86
+ premise and fixed hypothesis are encoded at an explicit segment boundary.
87
+ Truncation removes only a suffix of the premise. File counts concern content
88
+ tokens, excluding headers. Boundary-overlapping tokens count in both spans.
89
+ """
90
+ if not getattr(tokenizer, "is_fast", False):
91
+ raise ValueError("Package coverage requires a fast tokenizer with character offsets")
92
+ if not isinstance(hypothesis, str) or not hypothesis.strip():
93
+ raise ValueError("A fixed hypothesis is required")
94
+ ordered_files(row)
95
+ premise, spans = _render(row)
96
+ suffix_ids = list(tokenizer.encode("\nHypothesis: " + hypothesis, add_special_tokens=False))
97
+ if type(max_length) is not int or max_length <= len(suffix_ids) + 8:
98
+ raise ValueError("Input budget is too small for the premise and fixed hypothesis")
99
+ prefix = "Premise: "
100
+ encoding = tokenizer(prefix + premise, add_special_tokens=False, truncation=False,
101
+ return_offsets_mapping=True, return_attention_mask=False,
102
+ return_token_type_ids=False)
103
+ premise_ids = encoding["input_ids"]
104
+ offsets = encoding["offset_mapping"]
105
+ if len(premise_ids) != len(offsets) or any(start > end for start, end in offsets):
106
+ raise ValueError("Invalid tokenizer offsets")
107
+ used = min(len(premise_ids), max_length - len(suffix_ids))
108
+ ids = list(premise_ids[:used]) + suffix_ids
109
+ starts = [start for start, _ in offsets]
110
+ ends = [end for _, end in offsets]
111
+ if any(a > b for a, b in zip(starts, starts[1:])) or any(a > b for a, b in zip(ends, ends[1:])):
112
+ raise ValueError("Tokenizer offsets must be monotone for retained file coverage")
113
+ files = []
114
+ for span in spans:
115
+ start, end = len(prefix) + span["start"], len(prefix) + span["end"]
116
+ first = bisect_right(ends, start)
117
+ last = bisect_left(starts, end)
118
+ original_tokens = max(0, last - first) if start != end else 0
119
+ used_tokens = max(0, min(last, used) - first) if original_tokens else 0
120
+ if original_tokens:
121
+ status = "full" if used_tokens == original_tokens else ("partial" if used_tokens else "omitted")
122
+ else:
123
+ header_end = bisect_left(starts, start)
124
+ header_first = bisect_right(ends, len(prefix) + span["header_start"])
125
+ status = "full" if used >= header_end else ("partial" if used > header_first else "omitted")
126
+ files.append({"path": span["path"], "original_tokens": original_tokens,
127
+ "used_tokens": used_tokens, "status": status})
128
+ original = len(premise_ids) + len(suffix_ids)
129
+ return ids, original, {"original_tokens": original, "used_tokens": len(ids),
130
+ "premise_original_tokens": len(premise_ids), "premise_used_tokens": used,
131
+ "hypothesis_tokens": len(suffix_ids), "truncated": used < len(premise_ids),
132
+ "files_full": sum(f["status"] == "full" for f in files),
133
+ "files_partial": sum(f["status"] == "partial" for f in files),
134
+ "files_omitted": sum(f["status"] == "omitted" for f in files), "files": files}
scripts/skill_package.py ADDED
@@ -0,0 +1,100 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Read a skill as inert text; never import, install, or execute its files."""
2
+ import hashlib
3
+ import json
4
+ import os
5
+ from pathlib import Path
6
+ import stat
7
+
8
+
9
+ IGNORE_DIRS = {'.git', '.venv', 'venv', 'node_modules', '__pycache__', '.ssh', '.aws'}
10
+ MAX_FILE_BYTES = 1_048_576
11
+ MAX_PACKAGE_BYTES = 16_777_216
12
+ MAX_ENTRIES = 1000
13
+
14
+
15
+ def digest(value):
16
+ return hashlib.sha256(value).hexdigest()
17
+
18
+
19
+ def read_package(path):
20
+ path = Path(path).absolute()
21
+ if path.is_symlink() or path.is_junction():
22
+ raise ValueError('Choose a real package directory or SKILL.md, not a link.')
23
+ single = path.is_file()
24
+ if single and path.name != 'SKILL.md':
25
+ raise ValueError('A file input must be named SKILL.md.')
26
+ root = path.parent if single else path
27
+ if not root.is_dir():
28
+ raise ValueError('Package directory does not exist.')
29
+ root = root.resolve()
30
+ files, skipped, total, entries = [], [], 0, 0
31
+ if single:
32
+ candidates = [(root, [], ['SKILL.md'])]
33
+ else:
34
+ candidates = os.walk(root, followlinks=False, onerror=lambda exc: (_ for _ in ()).throw(exc))
35
+ for directory, dirs, names in candidates:
36
+ directory = Path(directory)
37
+ entries += len(dirs) + len(names)
38
+ if entries > MAX_ENTRIES:
39
+ raise ValueError(f'Package exceeds {MAX_ENTRIES} entries; select a smaller skill directory.')
40
+ for name in sorted(dirs[:]):
41
+ item = directory / name
42
+ reason = ('excluded_directory' if name.lower() in IGNORE_DIRS else
43
+ 'link' if item.is_symlink() or item.is_junction() else None)
44
+ if reason:
45
+ dirs.remove(name)
46
+ skipped.append({'path': item.relative_to(root).as_posix() + '/', 'reason': reason})
47
+ dirs.sort()
48
+ for name in sorted(names):
49
+ item = directory / name
50
+ relative = item.relative_to(root).as_posix()
51
+ reason = None
52
+ if item.is_symlink() or item.is_junction():
53
+ reason = 'link'
54
+ elif name.lower() == '.env' or name.lower().startswith('.env.') or item.suffix.lower() in {'.pem', '.key'}:
55
+ reason = 'private_configuration'
56
+ elif not stat.S_ISREG(item.stat().st_mode):
57
+ reason = 'not_regular_file'
58
+ elif any(ord(char) < 32 or ord(char) == 127 for char in relative) or '\\' in relative:
59
+ reason = 'unsupported_path'
60
+ elif not item.resolve().is_relative_to(root):
61
+ raise ValueError('A file resolved outside the package.')
62
+ if reason:
63
+ skipped.append({'path': relative, 'reason': reason})
64
+ continue
65
+ with item.open('rb') as stream:
66
+ raw = stream.read(MAX_FILE_BYTES + 1)
67
+ if len(raw) > MAX_FILE_BYTES:
68
+ skipped.append({'path': relative, 'reason': 'file_size_limit'})
69
+ continue
70
+ total += len(raw)
71
+ if total > MAX_PACKAGE_BYTES:
72
+ raise ValueError('Package exceeds the 16 MiB text-read budget.')
73
+ try:
74
+ text = raw.decode('utf-8')
75
+ if '\x00' in text:
76
+ raise UnicodeError('NUL byte')
77
+ except UnicodeError:
78
+ skipped.append({'path': relative, 'reason': 'not_utf8_text'})
79
+ continue
80
+ files.append({'path': relative, 'sha256': digest(raw), 'bytes': len(raw), 'text': text})
81
+ files.sort(key=lambda item: item['path'])
82
+ markdown = next((item for item in files if item['path'] == 'SKILL.md'), None)
83
+ if markdown is None or not markdown['text'].strip():
84
+ raise ValueError('A nonempty UTF-8 SKILL.md is required at the package root.')
85
+ manifest = [{key: item[key] for key in ('path', 'sha256', 'bytes')} for item in files]
86
+ fingerprint = digest(json.dumps(manifest, sort_keys=True, separators=(',', ':')).encode())
87
+ row = {'skill_text': markdown['text'],
88
+ 'files': [{'path': item['path'], 'text': item['text']} for item in files if item is not markdown]}
89
+ coverage = {'input_kind': 'markdown_file' if single else 'package_directory',
90
+ 'text_manifest_sha256': fingerprint, 'text_files': len(files),
91
+ 'text_bytes': sum(item['bytes'] for item in files), 'files': manifest, 'skipped': skipped}
92
+ return row, coverage
93
+
94
+
95
+ def write_json(path, value):
96
+ """Create outputs explicitly; do not replace a user's existing file."""
97
+ path = Path(path)
98
+ path.parent.mkdir(parents=True, exist_ok=True)
99
+ with path.open('x', encoding='utf-8', newline='\n') as stream:
100
+ stream.write(json.dumps(value, indent=2, ensure_ascii=False, allow_nan=False) + '\n')
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f399b3cd12fa270d51457bb749fb30863521e8359b8a27059c71b6c2f7d6dd6c
3
+ size 19989424
tokenizer_config.json ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|im_end|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": true,
13
+ "local_files_only": true,
14
+ "max_length": 1024,
15
+ "model_max_length": 262144,
16
+ "model_specific_special_tokens": {
17
+ "audio_bos_token": "<|audio_start|>",
18
+ "audio_eos_token": "<|audio_end|>",
19
+ "audio_token": "<|audio_pad|>",
20
+ "image_token": "<|image_pad|>",
21
+ "video_token": "<|video_pad|>",
22
+ "vision_bos_token": "<|vision_start|>",
23
+ "vision_eos_token": "<|vision_end|>"
24
+ },
25
+ "pad_token": "<|endoftext|>",
26
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
27
+ "split_special_tokens": false,
28
+ "stride": 0,
29
+ "tokenizer_class": "Qwen2Tokenizer",
30
+ "truncation_side": "right",
31
+ "truncation_strategy": "longest_first",
32
+ "unk_token": null,
33
+ "video_token": "<|video_pad|>",
34
+ "vision_bos_token": "<|vision_start|>",
35
+ "vision_eos_token": "<|vision_end|>"
36
+ }
training_config.json ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "train_records": 13514,
3
+ "max_length": 8192,
4
+ "epochs": 1,
5
+ "learning_rate": 0.0002,
6
+ "effective_batch": 16,
7
+ "base_model": {
8
+ "repo_id": "AlexWortega/openjev",
9
+ "subfolder": "qwen3.5-2b-nli-v5",
10
+ "revision": "a20448012c213128955ca0c693e7c943865cab77"
11
+ },
12
+ "trainable_parameters": 2193408,
13
+ "checkpoint_policy": "Fixed end of epoch",
14
+ "runtime": {
15
+ "torch": "2.8.0+cu126",
16
+ "transformers": "5.17.0",
17
+ "peft": "0.21.2",
18
+ "flash-linear-attention": "0.3.2",
19
+ "triton": "3.4.0"
20
+ },
21
+ "adapter_sha256": "da4424a0b6091820866e767769d82bca1c717bd7f488b484f766353d6ca6f48b",
22
+ "seed": 42,
23
+ "training_scope": "all_data",
24
+ "dataset": {
25
+ "repo_id": "SkillScreen/SkillScreen-Data",
26
+ "revision": "d59aabbd7e0e8d95b464a2d335f36e948577b832"
27
+ },
28
+ "lora": {
29
+ "r": 8,
30
+ "alpha": 16,
31
+ "dropout": 0,
32
+ "target_modules": [
33
+ "q_proj",
34
+ "v_proj",
35
+ "in_proj_qkv",
36
+ "out_proj"
37
+ ],
38
+ "classification_head_trainable": true
39
+ },
40
+ "weight_decay": 0.01,
41
+ "warmup_fraction": 0.05,
42
+ "gradient_clip_norm": 1.0,
43
+ "precision": "bfloat16",
44
+ "folds": 3,
45
+ "fold_grouping": "Normalized Markdown content groups; stratified by source and label.",
46
+ "selection": "Fixed final checkpoint after one epoch; no early stopping."
47
+ }