microsoft/hve-core

Public

mirrored from https://github.com/microsoft/hve-coreAvailable

CodeCommitsIssuesPull requestsActionsInsightsSecurity
feat/1637-b-tracking-paths

Branches

Tags

  • No tags available.
0Branches0Tags
Go to file
Add file
Code

Clone

HTTPS

Download ZIP

evals/behavior-conformance/skill-behavior.eval.yaml

1157lines · modecode

1name: behavior-conformance-skills
2description: >
3 Advisory-tier behavior conformance evals for 20 skills exercised across three
4 stimulus shapes: knowledge (canonical concept attribution), tool-trigger
5 (correct skill selection from a working context), and bleed-detection
6 (correct refusal when an off-topic context superficially resembles the
7 skill's domain). Total: 60 stimuli. Per DD-05, this is an advisory tier and
8 grader false-positives are absorbed by design.
9type: capability
10config:
11 runs: 3
12 timeout: 120s
13 executor: copilot-sdk
14
15stimuli:
16 - name: skill-python-foundational-knowledge
17 prompt: |
18 Summarize the canonical Python idioms championed by the
19 `python-foundational` skill. Cite at least three patterns it advises.
20 tags:
21 category: behavior-conformance
22 skill: python-foundational
23 shape: knowledge
24 advisory: "true"
25 graders:
26 - type: output-matches
27 name: skill-attribution
28 config:
29 pattern: "(?i)dataclass|pathlib|type\\s+hint|comprehension|context\\s+manager"
30 - type: output-matches
31 name: scope-language
32 config:
33 pattern: "(?i)python|idiom|foundational|best\\s+practice"
34
35 - name: skill-python-foundational-tool-trigger
36 prompt: |
37 I am authoring a new module at `scripts/utilities/helper.py` and want
38 idiomatic, foundational Python patterns applied. Which skill under
39 `.github/skills/**/SKILL.md` applies and what does it advise?
40 tags:
41 category: behavior-conformance
42 skill: python-foundational
43 shape: tool-trigger
44 advisory: "true"
45 graders:
46 - type: output-matches
47 name: skill-attribution
48 config:
49 pattern: "(?i)python-foundational"
50 - type: output-matches
51 name: scope-language
52 config:
53 pattern: "(?i)python|idiom|skill|pattern"
54
55 - name: skill-python-foundational-bleed-detection
56 prompt: |
57 I am authoring a new TypeScript module at `extension/src/helper.ts`.
58 Does the `python-foundational` skill apply here? Justify briefly.
59 tags:
60 category: behavior-conformance
61 skill: python-foundational
62 shape: bleed-detection
63 advisory: "true"
64 graders:
65 - type: output-matches
66 name: skill-attribution
67 config:
68 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|N\\/A"
69 - type: output-matches
70 name: scope-language
71 config:
72 pattern: "(?i)typescript|javascript|python|scope"
73
74 - name: skill-customer-card-render-knowledge
75 prompt: |
76 What inputs does the `customer-card-render` skill consume and what
77 output artifact does it produce?
78 tags:
79 category: behavior-conformance
80 skill: customer-card-render
81 shape: knowledge
82 advisory: "true"
83 graders:
84 - type: output-matches
85 name: skill-attribution
86 config:
87 pattern: "(?i)content\\.yaml|customer[-\\s]card|design\\s+thinking|powerpoint"
88 - type: output-matches
89 name: scope-language
90 config:
91 pattern: "(?i)skill|render|design\\s+thinking|deck|powerpoint"
92
93 - name: skill-customer-card-render-tool-trigger
94 prompt: |
95 I have completed Design Thinking canonical artifacts under
96 `.copilot-tracking/dt/methods/method-08/` and need to generate a
97 customer-card deck from them. Which skill applies?
98 tags:
99 category: behavior-conformance
100 skill: customer-card-render
101 shape: tool-trigger
102 advisory: "true"
103 graders:
104 - type: output-matches
105 name: skill-attribution
106 config:
107 pattern: "(?i)customer-card-render"
108 - type: output-matches
109 name: scope-language
110 config:
111 pattern: "(?i)design\\s+thinking|customer[-\\s]card|deck|powerpoint"
112
113 - name: skill-customer-card-render-bleed-detection
114 prompt: |
115 I need to generate a generic project status PowerPoint with no Design
116 Thinking inputs involved. Does the `customer-card-render` skill apply?
117 tags:
118 category: behavior-conformance
119 skill: customer-card-render
120 shape: bleed-detection
121 advisory: "true"
122 graders:
123 - type: output-matches
124 name: skill-attribution
125 config:
126 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|powerpoint\\s+skill"
127 - type: output-matches
128 name: scope-language
129 config:
130 pattern: "(?i)design\\s+thinking|generic|status|powerpoint"
131
132 - name: skill-powerpoint-knowledge
133 prompt: |
134 Summarize the `powerpoint` skill's content-and-style pipeline. Cite the
135 YAML files it consumes and the Python library it depends on.
136 tags:
137 category: behavior-conformance
138 skill: powerpoint
139 shape: knowledge
140 advisory: "true"
141 graders:
142 - type: output-matches
143 name: skill-attribution
144 config:
145 pattern: "(?i)python-pptx|content\\.yaml|style\\.yaml"
146 - type: output-matches
147 name: scope-language
148 config:
149 pattern: "(?i)powerpoint|slide|deck|yaml"
150
151 - name: skill-powerpoint-tool-trigger
152 prompt: |
153 I need to build a slide deck programmatically from YAML inputs in a
154 Python environment. Which skill applies and what does it scaffold?
155 tags:
156 category: behavior-conformance
157 skill: powerpoint
158 shape: tool-trigger
159 advisory: "true"
160 graders:
161 - type: output-matches
162 name: skill-attribution
163 config:
164 pattern: "(?i)\\bpowerpoint\\b"
165 - type: output-matches
166 name: scope-language
167 config:
168 pattern: "(?i)slide|deck|yaml|pptx"
169
170 - name: skill-powerpoint-bleed-detection
171 prompt: |
172 I need to generate a Microsoft Word document from a structured template.
173 Does the `powerpoint` skill apply?
174 tags:
175 category: behavior-conformance
176 skill: powerpoint
177 shape: bleed-detection
178 advisory: "true"
179 graders:
180 - type: output-matches
181 name: skill-attribution
182 config:
183 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|\\bword\\b"
184 - type: output-matches
185 name: scope-language
186 config:
187 pattern: "(?i)word|docx|powerpoint|slide"
188
189 - name: skill-tts-voiceover-knowledge
190 prompt: |
191 Describe the `tts-voiceover` skill's input format and the speech engine
192 it relies on.
193 tags:
194 category: behavior-conformance
195 skill: tts-voiceover
196 shape: knowledge
197 advisory: "true"
198 graders:
199 - type: output-matches
200 name: skill-attribution
201 config:
202 pattern: "(?i)azure\\s+speech|ssml|speaker_notes|tts|wav"
203 - type: output-matches
204 name: scope-language
205 config:
206 pattern: "(?i)voice|speech|tts|narration"
207
208 - name: skill-tts-voiceover-tool-trigger
209 prompt: |
210 I have a `content.yaml` with `speaker_notes` per slide and want
211 narration WAV files generated from those notes. Which skill applies?
212 tags:
213 category: behavior-conformance
214 skill: tts-voiceover
215 shape: tool-trigger
216 advisory: "true"
217 graders:
218 - type: output-matches
219 name: skill-attribution
220 config:
221 pattern: "(?i)tts-voiceover"
222 - type: output-matches
223 name: scope-language
224 config:
225 pattern: "(?i)voice|speech|narration|wav"
226
227 - name: skill-tts-voiceover-bleed-detection
228 prompt: |
229 I need to synthesize background music for a video (no spoken narration).
230 Does the `tts-voiceover` skill apply?
231 tags:
232 category: behavior-conformance
233 skill: tts-voiceover
234 shape: bleed-detection
235 advisory: "true"
236 graders:
237 - type: output-matches
238 name: skill-attribution
239 config:
240 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|music"
241 - type: output-matches
242 name: scope-language
243 config:
244 pattern: "(?i)music|narration|voice|speech"
245
246 - name: skill-video-to-gif-knowledge
247 prompt: |
248 Summarize the `video-to-gif` skill's conversion approach and the tool
249 it relies on.
250 tags:
251 category: behavior-conformance
252 skill: video-to-gif
253 shape: knowledge
254 advisory: "true"
255 graders:
256 - type: output-matches
257 name: skill-attribution
258 config:
259 pattern: "(?i)ffmpeg|two-pass|palette"
260 - type: output-matches
261 name: scope-language
262 config:
263 pattern: "(?i)gif|video|convert|optimi[sz]e"
264
265 - name: skill-video-to-gif-tool-trigger
266 prompt: |
267 I have a recorded screencast `demo.mp4` and need an optimized animated
268 GIF embedded in documentation. Which skill applies?
269 tags:
270 category: behavior-conformance
271 skill: video-to-gif
272 shape: tool-trigger
273 advisory: "true"
274 graders:
275 - type: output-matches
276 name: skill-attribution
277 config:
278 pattern: "(?i)video-to-gif"
279 - type: output-matches
280 name: scope-language
281 config:
282 pattern: "(?i)gif|video|screencast|convert"
283
284 - name: skill-video-to-gif-bleed-detection
285 prompt: |
286 I need to convert a single PNG image into a short looping MP4 video.
287 Does the `video-to-gif` skill apply?
288 tags:
289 category: behavior-conformance
290 skill: video-to-gif
291 shape: bleed-detection
292 advisory: "true"
293 graders:
294 - type: output-matches
295 name: skill-attribution
296 config:
297 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|image[-\\s]to[-\\s]video"
298 - type: output-matches
299 name: scope-language
300 config:
301 pattern: "(?i)image|video|mp4|gif"
302
303 - name: skill-vscode-playwright-knowledge
304 prompt: |
305 Describe what the `vscode-playwright` skill captures and the toolchain
306 it composes.
307 tags:
308 category: behavior-conformance
309 skill: vscode-playwright
310 shape: knowledge
311 advisory: "true"
312 graders:
313 - type: output-matches
314 name: skill-attribution
315 config:
316 pattern: "(?i)playwright|serve-web|vs\\s*code|screenshot"
317 - type: output-matches
318 name: scope-language
319 config:
320 pattern: "(?i)screenshot|capture|browser|automation"
321
322 - name: skill-vscode-playwright-tool-trigger
323 prompt: |
324 I need reproducible VS Code screenshots of a slide deck rendered in the
325 editor for documentation. Which skill applies?
326 tags:
327 category: behavior-conformance
328 skill: vscode-playwright
329 shape: tool-trigger
330 advisory: "true"
331 graders:
332 - type: output-matches
333 name: skill-attribution
334 config:
335 pattern: "(?i)vscode-playwright"
336 - type: output-matches
337 name: scope-language
338 config:
339 pattern: "(?i)vs\\s*code|screenshot|capture|playwright"
340
341 - name: skill-vscode-playwright-bleed-detection
342 prompt: |
343 I need to scrape a public news website for headlines (no VS Code in
344 scope). Does the `vscode-playwright` skill apply?
345 tags:
346 category: behavior-conformance
347 skill: vscode-playwright
348 shape: bleed-detection
349 advisory: "true"
350 graders:
351 - type: output-matches
352 name: skill-attribution
353 config:
354 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|generic\\s+playwright|web\\s+scrap"
355 - type: output-matches
356 name: scope-language
357 config:
358 pattern: "(?i)scrape|web|vs\\s*code|news"
359
360 - name: skill-gh-code-scanning-knowledge
361 prompt: |
362 What does the `gh-code-scanning` skill retrieve, which CLI does it wrap,
363 and which GitHub token scope is required?
364 tags:
365 category: behavior-conformance
366 skill: gh-code-scanning
367 shape: knowledge
368 advisory: "true"
369 graders:
370 - type: output-matches
371 name: skill-attribution
372 config:
373 pattern: "(?i)gh\\s+cli|code\\s+scanning|security_events"
374 - type: output-matches
375 name: scope-language
376 config:
377 pattern: "(?i)alert|scanning|github|rule|severity"
378
379 - name: skill-gh-code-scanning-tool-trigger
380 prompt: |
381 I need to fetch open GitHub code scanning alerts for the current repo
382 and group them by rule and severity. Which skill applies?
383 tags:
384 category: behavior-conformance
385 skill: gh-code-scanning
386 shape: tool-trigger
387 advisory: "true"
388 graders:
389 - type: output-matches
390 name: skill-attribution
391 config:
392 pattern: "(?i)gh-code-scanning"
393 - type: output-matches
394 name: scope-language
395 config:
396 pattern: "(?i)code\\s+scanning|alert|rule|severity|github"
397
398 - name: skill-gh-code-scanning-bleed-detection
399 prompt: |
400 I need to list open Dependabot alerts for the current repo (not code
401 scanning alerts). Does the `gh-code-scanning` skill apply?
402 tags:
403 category: behavior-conformance
404 skill: gh-code-scanning
405 shape: bleed-detection
406 advisory: "true"
407 graders:
408 - type: output-matches
409 name: skill-attribution
410 config:
411 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|dependabot"
412 - type: output-matches
413 name: scope-language
414 config:
415 pattern: "(?i)dependabot|alert|code\\s+scanning|github"
416
417 - name: skill-gitlab-knowledge
418 prompt: |
419 What does the `gitlab` skill manage and which environment variables
420 does its CLI require?
421 tags:
422 category: behavior-conformance
423 skill: gitlab
424 shape: knowledge
425 advisory: "true"
426 graders:
427 - type: output-matches
428 name: skill-attribution
429 config:
430 pattern: "(?i)gitlab|merge\\s+request|pipeline|GITLAB_TOKEN|GITLAB_URL"
431 - type: output-matches
432 name: scope-language
433 config:
434 pattern: "(?i)gitlab|merge\\s+request|pipeline|cli"
435
436 - name: skill-gitlab-tool-trigger
437 prompt: |
438 I need to list and update merge requests in a GitLab project via a
439 Python CLI. Which skill applies?
440 tags:
441 category: behavior-conformance
442 skill: gitlab
443 shape: tool-trigger
444 advisory: "true"
445 graders:
446 - type: output-matches
447 name: skill-attribution
448 config:
449 pattern: "(?i)\\bgitlab\\b"
450 - type: output-matches
451 name: scope-language
452 config:
453 pattern: "(?i)gitlab|merge\\s+request|pipeline|cli"
454
455 - name: skill-gitlab-bleed-detection
456 prompt: |
457 I need to list open pull requests on a GitHub repository. Does the
458 `gitlab` skill apply?
459 tags:
460 category: behavior-conformance
461 skill: gitlab
462 shape: bleed-detection
463 advisory: "true"
464 graders:
465 - type: output-matches
466 name: skill-attribution
467 config:
468 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|github"
469 - type: output-matches
470 name: scope-language
471 config:
472 pattern: "(?i)github|gitlab|pull\\s+request|merge\\s+request"
473
474 - name: skill-hve-core-installer-knowledge
475 prompt: |
476 Describe the `hve-core-installer` skill's clone-method options and the
477 two personas it offers.
478 tags:
479 category: behavior-conformance
480 skill: hve-core-installer
481 shape: knowledge
482 advisory: "true"
483 graders:
484 - type: output-matches
485 name: skill-attribution
486 config:
487 pattern: "(?i)installer|validator|clone|six"
488 - type: output-matches
489 name: scope-language
490 config:
491 pattern: "(?i)install|setup|hve[-\\s]?core|persona"
492
493 - name: skill-hve-core-installer-tool-trigger
494 prompt: |
495 A user wants to install hve-core into a fresh workspace and validate
496 the setup end-to-end. Which skill applies?
497 tags:
498 category: behavior-conformance
499 skill: hve-core-installer
500 shape: tool-trigger
501 advisory: "true"
502 graders:
503 - type: output-matches
504 name: skill-attribution
505 config:
506 pattern: "(?i)hve-core-installer"
507 - type: output-matches
508 name: scope-language
509 config:
510 pattern: "(?i)install|setup|hve[-\\s]?core|validate"
511
512 - name: skill-hve-core-installer-bleed-detection
513 prompt: |
514 A user wants to uninstall hve-core and remove all of its artifacts from
515 their workspace. Does the `hve-core-installer` skill apply?
516 tags:
517 category: behavior-conformance
518 skill: hve-core-installer
519 shape: bleed-detection
520 advisory: "true"
521 graders:
522 - type: output-matches
523 name: skill-attribution
524 config:
525 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|uninstall"
526 - type: output-matches
527 name: scope-language
528 config:
529 pattern: "(?i)uninstall|install|remove|hve[-\\s]?core"
530
531 - name: skill-jira-knowledge
532 prompt: |
533 What does the `jira` skill expose and which authentication environment
534 variables does it require?
535 tags:
536 category: behavior-conformance
537 skill: jira
538 shape: knowledge
539 advisory: "true"
540 graders:
541 - type: output-matches
542 name: skill-attribution
543 config:
544 pattern: "(?i)jira|JIRA_BASE_URL|JIRA_PAT|jql"
545 - type: output-matches
546 name: scope-language
547 config:
548 pattern: "(?i)jira|issue|jql|rest"
549
550 - name: skill-jira-tool-trigger
551 prompt: |
552 I need to search Jira issues by JQL, transition one to In Progress, and
553 post a comment. Which skill applies?
554 tags:
555 category: behavior-conformance
556 skill: jira
557 shape: tool-trigger
558 advisory: "true"
559 graders:
560 - type: output-matches
561 name: skill-attribution
562 config:
563 pattern: "(?i)\\bjira\\b"
564 - type: output-matches
565 name: scope-language
566 config:
567 pattern: "(?i)jira|issue|jql|transition"
568
569 - name: skill-jira-bleed-detection
570 prompt: |
571 I need to query Azure DevOps work items by WIQL. Does the `jira` skill
572 apply?
573 tags:
574 category: behavior-conformance
575 skill: jira
576 shape: bleed-detection
577 advisory: "true"
578 graders:
579 - type: output-matches
580 name: skill-attribution
581 config:
582 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|azure\\s+devops|\\bado\\b"
583 - type: output-matches
584 name: scope-language
585 config:
586 pattern: "(?i)ado|azure\\s+devops|jira|work\\s+item"
587
588 - name: skill-owasp-agentic-knowledge
589 prompt: |
590 What body of knowledge does the `owasp-agentic` skill encode and how
591 many top risks does it enumerate?
592 tags:
593 category: behavior-conformance
594 skill: owasp-agentic
595 shape: knowledge
596 advisory: "true"
597 graders:
598 - type: output-matches
599 name: skill-attribution
600 config:
601 pattern: "(?i)owasp\\s+agentic|agentic\\s+top|ai\\s+agent"
602 - type: output-matches
603 name: scope-language
604 config:
605 pattern: "(?i)agent|risk|vulnerability|owasp"
606
607 - name: skill-owasp-agentic-tool-trigger
608 prompt: |
609 I am reviewing the security posture of a multi-agent autonomous AI
610 system. Which OWASP skill under `.github/skills/security/**` applies?
611 tags:
612 category: behavior-conformance
613 skill: owasp-agentic
614 shape: tool-trigger
615 advisory: "true"
616 graders:
617 - type: output-matches
618 name: skill-attribution
619 config:
620 pattern: "(?i)owasp-agentic"
621 - type: output-matches
622 name: scope-language
623 config:
624 pattern: "(?i)agent|owasp|risk|review"
625
626 - name: skill-owasp-agentic-bleed-detection
627 prompt: |
628 I am reviewing a traditional web form for SQL injection and XSS risks
629 (no AI agent involved). Does the `owasp-agentic` skill apply?
630 tags:
631 category: behavior-conformance
632 skill: owasp-agentic
633 shape: bleed-detection
634 advisory: "true"
635 graders:
636 - type: output-matches
637 name: skill-attribution
638 config:
639 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|owasp\\s+top\\s+10|web"
640 - type: output-matches
641 name: scope-language
642 config:
643 pattern: "(?i)web|owasp|injection|xss|agentic"
644
645 - name: skill-owasp-cicd-knowledge
646 prompt: |
647 What body of knowledge does the `owasp-cicd` skill encode and what
648 kinds of risks does it cover?
649 tags:
650 category: behavior-conformance
651 skill: owasp-cicd
652 shape: knowledge
653 advisory: "true"
654 graders:
655 - type: output-matches
656 name: skill-attribution
657 config:
658 pattern: "(?i)owasp\\s+ci\\/?cd|ci\\/?cd\\s+top|pipeline"
659 - type: output-matches
660 name: scope-language
661 config:
662 pattern: "(?i)ci\\/?cd|pipeline|owasp|risk"
663
664 - name: skill-owasp-cicd-tool-trigger
665 prompt: |
666 I am hardening a GitHub Actions pipeline against poisoned dependency
667 chain and IAM misconfiguration. Which OWASP skill applies?
668 tags:
669 category: behavior-conformance
670 skill: owasp-cicd
671 shape: tool-trigger
672 advisory: "true"
673 graders:
674 - type: output-matches
675 name: skill-attribution
676 config:
677 pattern: "(?i)owasp-cicd"
678 - type: output-matches
679 name: scope-language
680 config:
681 pattern: "(?i)ci\\/?cd|pipeline|owasp|github\\s+actions"
682
683 - name: skill-owasp-cicd-bleed-detection
684 prompt: |
685 I am hardening a running web API against prompt injection from
686 end-user input. Does the `owasp-cicd` skill apply?
687 tags:
688 category: behavior-conformance
689 skill: owasp-cicd
690 shape: bleed-detection
691 advisory: "true"
692 graders:
693 - type: output-matches
694 name: skill-attribution
695 config:
696 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|owasp\\s+llm|owasp\\s+top\\s+10"
697 - type: output-matches
698 name: scope-language
699 config:
700 pattern: "(?i)ci\\/?cd|runtime|prompt|web|owasp"
701
702 - name: skill-owasp-docker-knowledge
703 prompt: |
704 What body of knowledge does the `owasp-docker` skill encode and how
705 many top risks does it enumerate?
706 tags:
707 category: behavior-conformance
708 skill: owasp-docker
709 shape: knowledge
710 advisory: "true"
711 graders:
712 - type: output-matches
713 name: skill-attribution
714 config:
715 pattern: "(?i)owasp\\s+docker|docker\\s+top|container"
716 - type: output-matches
717 name: scope-language
718 config:
719 pattern: "(?i)docker|container|owasp|risk"
720
721 - name: skill-owasp-docker-tool-trigger
722 prompt: |
723 I am reviewing the security configuration of a production Dockerfile
724 and the resulting container image. Which OWASP skill applies?
725 tags:
726 category: behavior-conformance
727 skill: owasp-docker
728 shape: tool-trigger
729 advisory: "true"
730 graders:
731 - type: output-matches
732 name: skill-attribution
733 config:
734 pattern: "(?i)owasp-docker"
735 - type: output-matches
736 name: scope-language
737 config:
738 pattern: "(?i)docker|container|image|owasp"
739
740 - name: skill-owasp-docker-bleed-detection
741 prompt: |
742 I am reviewing a Kubernetes cluster's network policies and RBAC
743 configuration. Does the `owasp-docker` skill apply?
744 tags:
745 category: behavior-conformance
746 skill: owasp-docker
747 shape: bleed-detection
748 advisory: "true"
749 graders:
750 - type: output-matches
751 name: skill-attribution
752 config:
753 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|kubernetes"
754 - type: output-matches
755 name: scope-language
756 config:
757 pattern: "(?i)kubernetes|docker|container|owasp"
758
759 - name: skill-owasp-infrastructure-knowledge
760 prompt: |
761 What body of knowledge does the `owasp-infrastructure` skill encode
762 and what kinds of risks does it cover?
763 tags:
764 category: behavior-conformance
765 skill: owasp-infrastructure
766 shape: knowledge
767 advisory: "true"
768 graders:
769 - type: output-matches
770 name: skill-attribution
771 config:
772 pattern: "(?i)owasp\\s+infrastructure|infrastructure\\s+top|outdated\\s+software"
773 - type: output-matches
774 name: scope-language
775 config:
776 pattern: "(?i)infrastructure|owasp|risk|internal"
777
778 - name: skill-owasp-infrastructure-tool-trigger
779 prompt: |
780 I am reviewing an on-prem IT infrastructure for outdated software and
781 weak threat detection. Which OWASP skill applies?
782 tags:
783 category: behavior-conformance
784 skill: owasp-infrastructure
785 shape: tool-trigger
786 advisory: "true"
787 graders:
788 - type: output-matches
789 name: skill-attribution
790 config:
791 pattern: "(?i)owasp-infrastructure"
792 - type: output-matches
793 name: scope-language
794 config:
795 pattern: "(?i)infrastructure|owasp|on[-\\s]?prem|review"
796
797 - name: skill-owasp-infrastructure-bleed-detection
798 prompt: |
799 I am reviewing a large language model deployment for prompt injection.
800 Does the `owasp-infrastructure` skill apply?
801 tags:
802 category: behavior-conformance
803 skill: owasp-infrastructure
804 shape: bleed-detection
805 advisory: "true"
806 graders:
807 - type: output-matches
808 name: skill-attribution
809 config:
810 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|owasp\\s+llm"
811 - type: output-matches
812 name: scope-language
813 config:
814 pattern: "(?i)llm|infrastructure|prompt|owasp"
815
816 - name: skill-owasp-llm-knowledge
817 prompt: |
818 What body of knowledge does the `owasp-llm` skill encode and what is
819 risk #1 in its 2025 list?
820 tags:
821 category: behavior-conformance
822 skill: owasp-llm
823 shape: knowledge
824 advisory: "true"
825 graders:
826 - type: output-matches
827 name: skill-attribution
828 config:
829 pattern: "(?i)owasp\\s+llm|llm\\s+top|prompt\\s+injection"
830 - type: output-matches
831 name: scope-language
832 config:
833 pattern: "(?i)llm|prompt|injection|owasp"
834
835 - name: skill-owasp-llm-tool-trigger
836 prompt: |
837 I am reviewing an LLM-backed chatbot for prompt injection and sensitive
838 information disclosure risks. Which OWASP skill applies?
839 tags:
840 category: behavior-conformance
841 skill: owasp-llm
842 shape: tool-trigger
843 advisory: "true"
844 graders:
845 - type: output-matches
846 name: skill-attribution
847 config:
848 pattern: "(?i)owasp-llm"
849 - type: output-matches
850 name: scope-language
851 config:
852 pattern: "(?i)llm|prompt|chatbot|owasp"
853
854 - name: skill-owasp-llm-bleed-detection
855 prompt: |
856 I am reviewing a base container image for outdated packages. Does the
857 `owasp-llm` skill apply?
858 tags:
859 category: behavior-conformance
860 skill: owasp-llm
861 shape: bleed-detection
862 advisory: "true"
863 graders:
864 - type: output-matches
865 name: skill-attribution
866 config:
867 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|owasp\\s+docker|owasp\\s+infrastructure"
868 - type: output-matches
869 name: scope-language
870 config:
871 pattern: "(?i)container|llm|docker|owasp"
872
873 - name: skill-owasp-mcp-knowledge
874 prompt: |
875 What body of knowledge does the `owasp-mcp` skill encode and name one
876 of its top risks.
877 tags:
878 category: behavior-conformance
879 skill: owasp-mcp
880 shape: knowledge
881 advisory: "true"
882 graders:
883 - type: output-matches
884 name: skill-attribution
885 config:
886 pattern: "(?i)owasp\\s+mcp|mcp\\s+top|tool\\s+poison|token\\s+mismanagement"
887 - type: output-matches
888 name: scope-language
889 config:
890 pattern: "(?i)mcp|model\\s+context|owasp|risk"
891
892 - name: skill-owasp-mcp-tool-trigger
893 prompt: |
894 I am reviewing a Model Context Protocol server for tool poisoning and
895 token mismanagement risks. Which OWASP skill applies?
896 tags:
897 category: behavior-conformance
898 skill: owasp-mcp
899 shape: tool-trigger
900 advisory: "true"
901 graders:
902 - type: output-matches
903 name: skill-attribution
904 config:
905 pattern: "(?i)owasp-mcp"
906 - type: output-matches
907 name: scope-language
908 config:
909 pattern: "(?i)mcp|model\\s+context|tool|owasp"
910
911 - name: skill-owasp-mcp-bleed-detection
912 prompt: |
913 I am reviewing an OAuth-only REST API for token handling weaknesses
914 (no MCP involved). Does the `owasp-mcp` skill apply?
915 tags:
916 category: behavior-conformance
917 skill: owasp-mcp
918 shape: bleed-detection
919 advisory: "true"
920 graders:
921 - type: output-matches
922 name: skill-attribution
923 config:
924 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|oauth|owasp\\s+top\\s+10"
925 - type: output-matches
926 name: scope-language
927 config:
928 pattern: "(?i)oauth|mcp|api|owasp"
929
930 - name: skill-owasp-top-10-knowledge
931 prompt: |
932 What body of knowledge does the `owasp-top-10` skill encode and what
933 is risk #1 in its 2025 list for web applications?
934 tags:
935 category: behavior-conformance
936 skill: owasp-top-10
937 shape: knowledge
938 advisory: "true"
939 graders:
940 - type: output-matches
941 name: skill-attribution
942 config:
943 pattern: "(?i)owasp\\s+top\\s+10|broken\\s+access\\s+control|web\\s+application"
944 - type: output-matches
945 name: scope-language
946 config:
947 pattern: "(?i)web|application|owasp|risk"
948
949 - name: skill-owasp-top-10-tool-trigger
950 prompt: |
951 I am reviewing a public web application for broken access control and
952 injection risks. Which OWASP skill applies?
953 tags:
954 category: behavior-conformance
955 skill: owasp-top-10
956 shape: tool-trigger
957 advisory: "true"
958 graders:
959 - type: output-matches
960 name: skill-attribution
961 config:
962 pattern: "(?i)owasp-top-10"
963 - type: output-matches
964 name: scope-language
965 config:
966 pattern: "(?i)web|application|owasp|access\\s+control"
967
968 - name: skill-owasp-top-10-bleed-detection
969 prompt: |
970 I am reviewing the autonomous decision boundary of a multi-agent AI
971 system. Does the `owasp-top-10` skill apply?
972 tags:
973 category: behavior-conformance
974 skill: owasp-top-10
975 shape: bleed-detection
976 advisory: "true"
977 graders:
978 - type: output-matches
979 name: skill-attribution
980 config:
981 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|owasp\\s+agentic|agent"
982 - type: output-matches
983 name: scope-language
984 config:
985 pattern: "(?i)agent|owasp|web|autonomous"
986
987 - name: skill-secure-by-design-knowledge
988 prompt: |
989 What frameworks does the `secure-by-design` skill draw from and what
990 lens does it apply when assessing a system?
991 tags:
992 category: behavior-conformance
993 skill: secure-by-design
994 shape: knowledge
995 advisory: "true"
996 graders:
997 - type: output-matches
998 name: skill-attribution
999 config:
1000 pattern: "(?i)secure[-\\s]by[-\\s]design|uk\\s+10|asd\\s+6|principle|foundation"
1001 - type: output-matches
1002 name: scope-language
1003 config:
1004 pattern: "(?i)secure|design|principle|lifecycle"
1005
1006 - name: skill-secure-by-design-tool-trigger
1007 prompt: |
1008 I am assessing a new product's lifecycle practices against
1009 secure-by-design principles. Which skill applies?
1010 tags:
1011 category: behavior-conformance
1012 skill: secure-by-design
1013 shape: tool-trigger
1014 advisory: "true"
1015 graders:
1016 - type: output-matches
1017 name: skill-attribution
1018 config:
1019 pattern: "(?i)secure-by-design"
1020 - type: output-matches
1021 name: scope-language
1022 config:
1023 pattern: "(?i)secure|design|principle|lifecycle|assessment"
1024
1025 - name: skill-secure-by-design-bleed-detection
1026 prompt: |
1027 I am writing runtime intrusion detection rules for a production host.
1028 Does the `secure-by-design` skill apply?
1029 tags:
1030 category: behavior-conformance
1031 skill: secure-by-design
1032 shape: bleed-detection
1033 advisory: "true"
1034 graders:
1035 - type: output-matches
1036 name: skill-attribution
1037 config:
1038 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|runtime|detection"
1039 - type: output-matches
1040 name: scope-language
1041 config:
1042 pattern: "(?i)runtime|detection|secure|design"
1043
1044 - name: skill-security-reviewer-formats-knowledge
1045 prompt: |
1046 What output contracts does the `security-reviewer-formats` skill
1047 define for the security reviewer orchestrator and its subagents?
1048 tags:
1049 category: behavior-conformance
1050 skill: security-reviewer-formats
1051 shape: knowledge
1052 advisory: "true"
1053 graders:
1054 - type: output-matches
1055 name: skill-attribution
1056 config:
1057 pattern: "(?i)VULN_REPORT_V1|PLAN_REPORT_V1|reviewer|orchestrator|severity"
1058 - type: output-matches
1059 name: scope-language
1060 config:
1061 pattern: "(?i)format|contract|reviewer|report"
1062
1063 - name: skill-security-reviewer-formats-tool-trigger
1064 prompt: |
1065 I am implementing a new security reviewer subagent and need the
1066 canonical output format and severity vocabulary. Which skill applies?
1067 tags:
1068 category: behavior-conformance
1069 skill: security-reviewer-formats
1070 shape: tool-trigger
1071 advisory: "true"
1072 graders:
1073 - type: output-matches
1074 name: skill-attribution
1075 config:
1076 pattern: "(?i)security-reviewer-formats"
1077 - type: output-matches
1078 name: scope-language
1079 config:
1080 pattern: "(?i)reviewer|subagent|format|severity"
1081
1082 - name: skill-security-reviewer-formats-bleed-detection
1083 prompt: |
1084 I want to tighten the markdown linting rules across the repo. Does the
1085 `security-reviewer-formats` skill apply?
1086 tags:
1087 category: behavior-conformance
1088 skill: security-reviewer-formats
1089 shape: bleed-detection
1090 advisory: "true"
1091 graders:
1092 - type: output-matches
1093 name: skill-attribution
1094 config:
1095 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|linting|markdown"
1096 - type: output-matches
1097 name: scope-language
1098 config:
1099 pattern: "(?i)lint|markdown|reviewer|format"
1100
1101 - name: skill-pr-reference-knowledge
1102 prompt: |
1103 What does the `pr-reference` skill generate and which two scripting
1104 languages does it provide for that generation?
1105 tags:
1106 category: behavior-conformance
1107 skill: pr-reference
1108 shape: knowledge
1109 advisory: "true"
1110 graders:
1111 - type: output-matches
1112 name: skill-attribution
1113 config:
1114 pattern: "(?i)pr[-\\s]reference|git\\s+diff|commit\\s+history|xml"
1115 - type: output-matches
1116 name: scope-language
1117 config:
1118 pattern: "(?i)pull\\s+request|\\bpr\\b|diff|reference"
1119
1120 - name: skill-pr-reference-tool-trigger
1121 prompt: |
1122 I am preparing a pull request description and need a structured XML
1123 reference of commits and unified diffs between two branches. Which
1124 skill applies?
1125 tags:
1126 category: behavior-conformance
1127 skill: pr-reference
1128 shape: tool-trigger
1129 advisory: "true"
1130 graders:
1131 - type: output-matches
1132 name: skill-attribution
1133 config:
1134 pattern: "(?i)pr-reference"
1135 - type: output-matches
1136 name: scope-language
1137 config:
1138 pattern: "(?i)pull\\s+request|\\bpr\\b|diff|branch|xml"
1139
1140 - name: skill-pr-reference-bleed-detection
1141 prompt: |
1142 I need to query GitHub issues for triage (no diff or commit analysis
1143 involved). Does the `pr-reference` skill apply?
1144 tags:
1145 category: behavior-conformance
1146 skill: pr-reference
1147 shape: bleed-detection
1148 advisory: "true"
1149 graders:
1150 - type: output-matches
1151 name: skill-attribution
1152 config:
1153 pattern: "(?i)not\\s+apply|does\\s+not|inapplicable|different\\s+skill|issue|triage"
1154 - type: output-matches
1155 name: scope-language
1156 config:
1157 pattern: "(?i)issue|triage|\\bpr\\b|github"