gabbe-kit 1.0.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (502) hide show
  1. package/LICENSE +25 -0
  2. package/LICENSE-CODE +202 -0
  3. package/README.md +817 -0
  4. package/agents/AGENTS.md +542 -0
  5. package/agents/CONSTITUTION.md +188 -0
  6. package/agents/guides/00-index.md +142 -0
  7. package/agents/guides/ai/a2ui-protocols.md +71 -0
  8. package/agents/guides/ai/actor-agent-frameworks.md +82 -0
  9. package/agents/guides/ai/agent-communication.md +146 -0
  10. package/agents/guides/ai/agent-identity-trust.md +69 -0
  11. package/agents/guides/ai/agent-only-cognition.md +73 -0
  12. package/agents/guides/ai/agent-only-swarms.md +70 -0
  13. package/agents/guides/ai/agent-operating-ergonomics.md +52 -0
  14. package/agents/guides/ai/agent-ui.md +106 -0
  15. package/agents/guides/ai/ai-agentic.md +309 -0
  16. package/agents/guides/ai/beyond-llms.md +60 -0
  17. package/agents/guides/ai/dynamic-capability-loading.md +121 -0
  18. package/agents/guides/ai/gabbe-innovations.md +111 -0
  19. package/agents/guides/ai/model-customization-decisions.md +78 -0
  20. package/agents/guides/ai/multi-agent-systems.md +85 -0
  21. package/agents/guides/ai/self-evolving-skills.md +137 -0
  22. package/agents/guides/ai/self-healing-summary.md +21 -0
  23. package/agents/guides/ai/visual-mcp-integration.md +106 -0
  24. package/agents/guides/architecture/api-standards.md +61 -0
  25. package/agents/guides/architecture/blockchain-dlt.md +36 -0
  26. package/agents/guides/architecture/critical-systems-arch.md +66 -0
  27. package/agents/guides/architecture/event-driven-architecture.md +39 -0
  28. package/agents/guides/architecture/microservices.md +292 -0
  29. package/agents/guides/architecture/monolith.md +303 -0
  30. package/agents/guides/architecture/orchestration.md +75 -0
  31. package/agents/guides/architecture/system-qualities.md +51 -0
  32. package/agents/guides/architecture/system-scalability.md +47 -0
  33. package/agents/guides/architecture/systems-architecture.md +506 -0
  34. package/agents/guides/data/semantic-web.md +23 -0
  35. package/agents/guides/data/sql-nosql.md +305 -0
  36. package/agents/guides/industry/engineering-standards.md +36 -0
  37. package/agents/guides/industry/global-standards.md +24 -0
  38. package/agents/guides/industry/healthcare-fhir.md +33 -0
  39. package/agents/guides/industry/industrial-iot.md +33 -0
  40. package/agents/guides/industry/telecom-networks.md +34 -0
  41. package/agents/guides/languages/c-sharp.md +66 -0
  42. package/agents/guides/languages/go-lang.md +251 -0
  43. package/agents/guides/languages/java.md +73 -0
  44. package/agents/guides/languages/js-ts-nodejs.md +361 -0
  45. package/agents/guides/languages/nodejs-advanced.md +56 -0
  46. package/agents/guides/languages/php-laravel.md +376 -0
  47. package/agents/guides/languages/python-fastapi-ai.md +257 -0
  48. package/agents/guides/languages/rust.md +73 -0
  49. package/agents/guides/ops/compliance-audit.md +177 -0
  50. package/agents/guides/ops/cost-optimization.md +57 -0
  51. package/agents/guides/ops/dev-environments.md +67 -0
  52. package/agents/guides/ops/dev-workflow.md +44 -0
  53. package/agents/guides/ops/gabbe-cli-workflows.md +252 -0
  54. package/agents/guides/ops/production-health.md +62 -0
  55. package/agents/guides/ops/troubleshooting-guide.md +69 -0
  56. package/agents/guides/patterns/agentic-design-patterns.md +308 -0
  57. package/agents/guides/patterns/agentic-patterns.md +43 -0
  58. package/agents/guides/patterns/ai-native-scenarios.md +71 -0
  59. package/agents/guides/patterns/autonomous-swarm-patterns.md +42 -0
  60. package/agents/guides/patterns/design-patterns.md +68 -0
  61. package/agents/guides/patterns/enterprise-migration-scenario.md +58 -0
  62. package/agents/guides/patterns/enterprise-patterns.md +56 -0
  63. package/agents/guides/patterns/software-design-patterns.md +106 -0
  64. package/agents/guides/patterns/time-complexity-analysis.md +116 -0
  65. package/agents/guides/planning/product-requirements.md +407 -0
  66. package/agents/guides/planning/strategic-analysis.md +98 -0
  67. package/agents/guides/planning/visual-product-specs.md +162 -0
  68. package/agents/guides/principles/clean-code-standards.md +51 -0
  69. package/agents/guides/principles/computing-foundations.md +87 -0
  70. package/agents/guides/principles/diagramming-standards.md +568 -0
  71. package/agents/guides/principles/evaluation-strategy.md +103 -0
  72. package/agents/guides/principles/future-tech.md +35 -0
  73. package/agents/guides/principles/human-agent-collaboration.md +54 -0
  74. package/agents/guides/principles/legacy-tech.md +61 -0
  75. package/agents/guides/principles/mathematical-foundations.md +103 -0
  76. package/agents/guides/principles/modeling-methods.md +97 -0
  77. package/agents/guides/principles/no-code-integration.md +84 -0
  78. package/agents/guides/principles/testing-strategy.md +97 -0
  79. package/agents/guides/principles/visual-design-system.md +44 -0
  80. package/agents/guides/processes/RARV_CYCLE.md +262 -0
  81. package/agents/guides/processes/extension-protocol.md +183 -0
  82. package/agents/guides/processes/full-system-lifecycle.md +47 -0
  83. package/agents/guides/processes/knowledge-integration.md +82 -0
  84. package/agents/guides/processes/loki-sdlc-phases.md +446 -0
  85. package/agents/guides/product/green-software.md +38 -0
  86. package/agents/guides/product/sustainability-checks.md +29 -0
  87. package/agents/guides/security/ai-risk-standards-map.md +98 -0
  88. package/agents/guides/security/api-security.md +51 -0
  89. package/agents/guides/security/cryptography-standards.md +38 -0
  90. package/agents/guides/security/privacy-data-protection.md +41 -0
  91. package/agents/guides/security/secure-architecture.md +43 -0
  92. package/agents/guides/security/threat-modeling.md +46 -0
  93. package/agents/memory/AUDIT_LOG.md +48 -0
  94. package/agents/memory/CONTINUITY.md +49 -0
  95. package/agents/memory/PROJECT_STATE.md +134 -0
  96. package/agents/memory/RESUME_POINTER.md +15 -0
  97. package/agents/memory/episodic/DECISION_LOG_TEMPLATE.md +104 -0
  98. package/agents/memory/semantic/PROJECT_KNOWLEDGE_TEMPLATE.md +194 -0
  99. package/agents/memory/semantic/VECTOR_DB_CONFIG_TEMPLATE.json +46 -0
  100. package/agents/personas/00-index.md +148 -0
  101. package/agents/personas/biz-compliance.md +112 -0
  102. package/agents/personas/biz-legal.md +131 -0
  103. package/agents/personas/eng-api.md +110 -0
  104. package/agents/personas/eng-backend.md +113 -0
  105. package/agents/personas/eng-data.md +50 -0
  106. package/agents/personas/eng-database.md +95 -0
  107. package/agents/personas/eng-frontend.md +75 -0
  108. package/agents/personas/eng-infra.md +105 -0
  109. package/agents/personas/eng-messaging.md +50 -0
  110. package/agents/personas/eng-ml.md +50 -0
  111. package/agents/personas/eng-mobile.md +75 -0
  112. package/agents/personas/eng-perf.md +90 -0
  113. package/agents/personas/eng-qa.md +125 -0
  114. package/agents/personas/eng-tooling.md +49 -0
  115. package/agents/personas/ops-compliance.md +109 -0
  116. package/agents/personas/ops-cost.md +92 -0
  117. package/agents/personas/ops-devops.md +85 -0
  118. package/agents/personas/ops-incident.md +123 -0
  119. package/agents/personas/ops-monitor.md +89 -0
  120. package/agents/personas/ops-release.md +91 -0
  121. package/agents/personas/ops-security.md +104 -0
  122. package/agents/personas/ops-sre.md +85 -0
  123. package/agents/personas/orch-coordinator.md +140 -0
  124. package/agents/personas/orch-judge.md +135 -0
  125. package/agents/personas/orch-planner.md +114 -0
  126. package/agents/personas/orch-researcher.md +165 -0
  127. package/agents/personas/prod-architect.md +110 -0
  128. package/agents/personas/prod-design.md +112 -0
  129. package/agents/personas/prod-ethicist.md +50 -0
  130. package/agents/personas/prod-integration.md +120 -0
  131. package/agents/personas/prod-pm.md +115 -0
  132. package/agents/personas/prod-product-ops.md +128 -0
  133. package/agents/personas/prod-research.md +49 -0
  134. package/agents/personas/prod-safety-engineer.md +28 -0
  135. package/agents/personas/prod-tech-lead.md +119 -0
  136. package/agents/personas/ui-design.md +81 -0
  137. package/agents/scripts/check_skills_docs.py +120 -0
  138. package/agents/scripts/compile_skills.py +284 -0
  139. package/agents/scripts/comprehensive_checker.py +187 -0
  140. package/agents/scripts/dump-context.sh +37 -0
  141. package/agents/scripts/eval_skills.py +268 -0
  142. package/agents/scripts/preflight_summary.sh +80 -0
  143. package/agents/scripts/report_duplicates.py +125 -0
  144. package/agents/scripts/setup-context.ps1 +250 -0
  145. package/agents/scripts/setup-context.sh +398 -0
  146. package/agents/scripts/state_export.sh +84 -0
  147. package/agents/scripts/state_import.sh +99 -0
  148. package/agents/scripts/validate_integrity.py +98 -0
  149. package/agents/scripts/validate_links.py +96 -0
  150. package/agents/scripts/validate_methodology_graph.py +194 -0
  151. package/agents/scripts/validate_skills.py +81 -0
  152. package/agents/scripts/verify_e2e_init_flows.py +246 -0
  153. package/agents/scripts/verify_setup_simulation.py +104 -0
  154. package/agents/scripts/verify_triggers_and_mcps.py +159 -0
  155. package/agents/scripts/verify_use_cases.py +498 -0
  156. package/agents/skills/00-index.md +322 -0
  157. package/agents/skills/ai/a2ui-protocols.skill.md +31 -0
  158. package/agents/skills/ai/actor-agent-frameworks.skill.md +31 -0
  159. package/agents/skills/ai/agent-communication.skill.md +36 -0
  160. package/agents/skills/ai/agent-ui.skill.md +31 -0
  161. package/agents/skills/ai/autonomous-swarm-patterns.skill.md +31 -0
  162. package/agents/skills/ai/beyond-llms.skill.md +31 -0
  163. package/agents/skills/ai/context-engineering.skill.md +79 -0
  164. package/agents/skills/ai/llm-as-judge.skill.md +97 -0
  165. package/agents/skills/ai/multi-agent-systems.skill.md +31 -0
  166. package/agents/skills/ai/rag-evaluation.skill.md +93 -0
  167. package/agents/skills/ai/reasoning-patterns.skill.md +125 -0
  168. package/agents/skills/architecture/adaptive-architecture.skill.md +51 -0
  169. package/agents/skills/architecture/api-design.skill.md +92 -0
  170. package/agents/skills/architecture/api-standards.skill.md +31 -0
  171. package/agents/skills/architecture/arch-debt.skill.md +180 -0
  172. package/agents/skills/architecture/arch-design.skill.md +304 -0
  173. package/agents/skills/architecture/arch-patterns.skill.md +52 -0
  174. package/agents/skills/architecture/arch-review.skill.md +326 -0
  175. package/agents/skills/architecture/architecture-governance.skill.md +62 -0
  176. package/agents/skills/architecture/attribute-driven-design.skill.md +108 -0
  177. package/agents/skills/architecture/blockchain-dlt.skill.md +28 -0
  178. package/agents/skills/architecture/compatibility-design.skill.md +51 -0
  179. package/agents/skills/architecture/critical-systems-arch.skill.md +31 -0
  180. package/agents/skills/architecture/design-patterns.skill.md +59 -0
  181. package/agents/skills/architecture/diagramming.skill.md +380 -0
  182. package/agents/skills/architecture/domain-model.skill.md +294 -0
  183. package/agents/skills/architecture/enterprise-migration-scenario.skill.md +31 -0
  184. package/agents/skills/architecture/enterprise-patterns.skill.md +31 -0
  185. package/agents/skills/architecture/error-handling-strategy.skill.md +58 -0
  186. package/agents/skills/architecture/event-driven-architecture.skill.md +31 -0
  187. package/agents/skills/architecture/event-governance.skill.md +55 -0
  188. package/agents/skills/architecture/fitness-functions.skill.md +111 -0
  189. package/agents/skills/architecture/graphql-schema.skill.md +64 -0
  190. package/agents/skills/architecture/legacy-modernization.skill.md +51 -0
  191. package/agents/skills/architecture/microservices.skill.md +31 -0
  192. package/agents/skills/architecture/middleware-design.skill.md +59 -0
  193. package/agents/skills/architecture/monolith.skill.md +31 -0
  194. package/agents/skills/architecture/realtime-comm.skill.md +52 -0
  195. package/agents/skills/architecture/state-management.skill.md +72 -0
  196. package/agents/skills/architecture/system-qualities.skill.md +31 -0
  197. package/agents/skills/architecture/system-scalability.skill.md +31 -0
  198. package/agents/skills/architecture/systems-architecture.skill.md +31 -0
  199. package/agents/skills/architecture/visual-whiteboarding.skill.md +33 -0
  200. package/agents/skills/brain/README.md +70 -0
  201. package/agents/skills/brain/active-inference.skill.md +74 -0
  202. package/agents/skills/brain/brain-mode.skill.md +125 -0
  203. package/agents/skills/brain/cognitive-architectures.skill.md +53 -0
  204. package/agents/skills/brain/cognitive-testing.skill.md +62 -0
  205. package/agents/skills/brain/consciousness-loop.skill.md +68 -0
  206. package/agents/skills/brain/cost-benefit-router.skill.md +89 -0
  207. package/agents/skills/brain/episodic-consolidation.skill.md +43 -0
  208. package/agents/skills/brain/epistemology-knowledge.skill.md +54 -0
  209. package/agents/skills/brain/global-workspace.skill.md +77 -0
  210. package/agents/skills/brain/learning-adaptation.skill.md +68 -0
  211. package/agents/skills/brain/loki-mode.skill.md +267 -0
  212. package/agents/skills/brain/neuroscience-foundations.skill.md +52 -0
  213. package/agents/skills/brain/scripts/active_inference_loop.py +92 -0
  214. package/agents/skills/brain/scripts/cognition_test.py +80 -0
  215. package/agents/skills/brain/scripts/global_workspace.py +104 -0
  216. package/agents/skills/brain/self-improvement.skill.md +80 -0
  217. package/agents/skills/brain/sensory-motor.skill.md +61 -0
  218. package/agents/skills/brain/sequential-thinking.skill.md +48 -0
  219. package/agents/skills/brain/working-memory.skill.md +47 -0
  220. package/agents/skills/coding/agent-trajectory-eval.skill.md +92 -0
  221. package/agents/skills/coding/artisan-commands.skill.md +52 -0
  222. package/agents/skills/coding/browser-tdd.skill.md +131 -0
  223. package/agents/skills/coding/chaos-fault-injection.skill.md +116 -0
  224. package/agents/skills/coding/ci-autofix.skill.md +53 -0
  225. package/agents/skills/coding/clean-coder.skill.md +58 -0
  226. package/agents/skills/coding/code-review.skill.md +100 -0
  227. package/agents/skills/coding/debug.skill.md +100 -0
  228. package/agents/skills/coding/documentation.skill.md +149 -0
  229. package/agents/skills/coding/eval-driven-development.skill.md +107 -0
  230. package/agents/skills/coding/evals/code-review.eval.yaml +29 -0
  231. package/agents/skills/coding/excalidraw.skill.md +90 -0
  232. package/agents/skills/coding/file-processing.skill.md +55 -0
  233. package/agents/skills/coding/git-workflow.skill.md +153 -0
  234. package/agents/skills/coding/metamorphic-testing.skill.md +113 -0
  235. package/agents/skills/coding/mobile-dev.skill.md +41 -0
  236. package/agents/skills/coding/nodejs-expert.skill.md +23 -0
  237. package/agents/skills/coding/output-validation.skill.md +57 -0
  238. package/agents/skills/coding/pbt-strategy.skill.md +117 -0
  239. package/agents/skills/coding/performant-ai.skill.md +62 -0
  240. package/agents/skills/coding/performant-go.skill.md +61 -0
  241. package/agents/skills/coding/performant-laravel.skill.md +70 -0
  242. package/agents/skills/coding/performant-nodejs.skill.md +67 -0
  243. package/agents/skills/coding/performant-php.skill.md +53 -0
  244. package/agents/skills/coding/performant-python.skill.md +62 -0
  245. package/agents/skills/coding/refactor.skill.md +124 -0
  246. package/agents/skills/coding/secure-coding.skill.md +57 -0
  247. package/agents/skills/coding/sketch-to-diagram.skill.md +106 -0
  248. package/agents/skills/coding/spec-driven-development.skill.md +110 -0
  249. package/agents/skills/coding/tdd-cycle.skill.md +92 -0
  250. package/agents/skills/coding/testing-strategy.skill.md +74 -0
  251. package/agents/skills/coding/time-complexity.skill.md +112 -0
  252. package/agents/skills/coding/tldraw-canvas.skill.md +109 -0
  253. package/agents/skills/coding/tool-construction.skill.md +47 -0
  254. package/agents/skills/coding/ui-gen.skill.md +64 -0
  255. package/agents/skills/coding/vibe-coding.skill.md +63 -0
  256. package/agents/skills/coding/visual-design.skill.md +65 -0
  257. package/agents/skills/coordination/agent-interop.skill.md +72 -0
  258. package/agents/skills/coordination/agent-protocol.skill.md +62 -0
  259. package/agents/skills/coordination/agent-workflow-patterns.skill.md +123 -0
  260. package/agents/skills/coordination/agentic-linter.skill.md +134 -0
  261. package/agents/skills/coordination/agentic-patterns.skill.md +79 -0
  262. package/agents/skills/coordination/meta-optimize.skill.md +62 -0
  263. package/agents/skills/coordination/meta-prompting.skill.md +46 -0
  264. package/agents/skills/coordination/multi-agent-orch.skill.md +78 -0
  265. package/agents/skills/coordination/persona-selector.skill.md +78 -0
  266. package/agents/skills/coordination/self-optimize.skill.md +81 -0
  267. package/agents/skills/coordination/swarm-consensus.skill.md +52 -0
  268. package/agents/skills/core/agent-analytics.skill.md +74 -0
  269. package/agents/skills/core/audit-trail.skill.md +122 -0
  270. package/agents/skills/core/clarify.skill.md +89 -0
  271. package/agents/skills/core/emerging-tech.skill.md +52 -0
  272. package/agents/skills/core/empirical-methods.skill.md +101 -0
  273. package/agents/skills/core/evals/clarify.eval.yaml +29 -0
  274. package/agents/skills/core/evals/integrity-check.eval.yaml +23 -0
  275. package/agents/skills/core/evals/research.eval.yaml +23 -0
  276. package/agents/skills/core/final-review.skill.md +66 -0
  277. package/agents/skills/core/integrity-check.skill.md +149 -0
  278. package/agents/skills/core/knowledge-connect.skill.md +77 -0
  279. package/agents/skills/core/knowledge-gap.skill.md +123 -0
  280. package/agents/skills/core/preflight.skill.md +106 -0
  281. package/agents/skills/core/research.skill.md +107 -0
  282. package/agents/skills/core/retrospective.skill.md +59 -0
  283. package/agents/skills/core/sdlc-checkpoint.skill.md +151 -0
  284. package/agents/skills/core/self-heal.skill.md +163 -0
  285. package/agents/skills/core/session-resume.skill.md +125 -0
  286. package/agents/skills/core/skills-registry.skill.md +65 -0
  287. package/agents/skills/core/state-portability.skill.md +89 -0
  288. package/agents/skills/core/state-preserve.skill.md +76 -0
  289. package/agents/skills/core/system-lifecycle.skill.md +51 -0
  290. package/agents/skills/core/update-scan.skill.md +92 -0
  291. package/agents/skills/data/data-engineering.skill.md +56 -0
  292. package/agents/skills/data/data-governance.skill.md +49 -0
  293. package/agents/skills/data/db-migration.skill.md +136 -0
  294. package/agents/skills/data/semantic-web.skill.md +28 -0
  295. package/agents/skills/data/sql-optimization.skill.md +78 -0
  296. package/agents/skills/industry/engineering-standards.skill.md +31 -0
  297. package/agents/skills/industry/global-standards.skill.md +30 -0
  298. package/agents/skills/industry/healthcare-fhir.skill.md +28 -0
  299. package/agents/skills/industry/industrial-iot.skill.md +30 -0
  300. package/agents/skills/industry/telecom-networks.skill.md +30 -0
  301. package/agents/skills/ops/caching-strategy.skill.md +77 -0
  302. package/agents/skills/ops/capacity-planning.skill.md +67 -0
  303. package/agents/skills/ops/cloud-deploy.skill.md +55 -0
  304. package/agents/skills/ops/configuration-management.skill.md +107 -0
  305. package/agents/skills/ops/cost-optimization.skill.md +39 -0
  306. package/agents/skills/ops/decommission-sunset.skill.md +64 -0
  307. package/agents/skills/ops/dependency-lifecycle.skill.md +63 -0
  308. package/agents/skills/ops/deployment.skill.md +217 -0
  309. package/agents/skills/ops/dev-environments.skill.md +31 -0
  310. package/agents/skills/ops/dev-workflow.skill.md +28 -0
  311. package/agents/skills/ops/docker-dev.skill.md +52 -0
  312. package/agents/skills/ops/enterprise-integration.skill.md +53 -0
  313. package/agents/skills/ops/feature-flag-management.skill.md +59 -0
  314. package/agents/skills/ops/incident-response.skill.md +40 -0
  315. package/agents/skills/ops/infra-devops.skill.md +53 -0
  316. package/agents/skills/ops/k8s-dev.skill.md +51 -0
  317. package/agents/skills/ops/memory-optimization.skill.md +45 -0
  318. package/agents/skills/ops/observability-stack-setup.skill.md +62 -0
  319. package/agents/skills/ops/performance-audit.skill.md +178 -0
  320. package/agents/skills/ops/performance-optimization.skill.md +51 -0
  321. package/agents/skills/ops/production-health.skill.md +85 -0
  322. package/agents/skills/ops/production-verifier.skill.md +50 -0
  323. package/agents/skills/ops/queue-management.skill.md +52 -0
  324. package/agents/skills/ops/release-management.skill.md +39 -0
  325. package/agents/skills/ops/release-validation.skill.md +51 -0
  326. package/agents/skills/ops/reliability-sre.skill.md +51 -0
  327. package/agents/skills/ops/runbook-authoring.skill.md +64 -0
  328. package/agents/skills/ops/system-benchmark.skill.md +69 -0
  329. package/agents/skills/ops/tech-debt.skill.md +143 -0
  330. package/agents/skills/ops/troubleshooting-guide.skill.md +31 -0
  331. package/agents/skills/product/accessibility.skill.md +143 -0
  332. package/agents/skills/product/adr-writer.skill.md +110 -0
  333. package/agents/skills/product/business-case.skill.md +70 -0
  334. package/agents/skills/product/change-management.skill.md +90 -0
  335. package/agents/skills/product/decompose.skill.md +75 -0
  336. package/agents/skills/product/design-thinking.skill.md +62 -0
  337. package/agents/skills/product/estimation-sizing.skill.md +87 -0
  338. package/agents/skills/product/evals/spec-writer.eval.yaml +22 -0
  339. package/agents/skills/product/financial-governance.skill.md +88 -0
  340. package/agents/skills/product/green-software.skill.md +46 -0
  341. package/agents/skills/product/ideation-facilitation.skill.md +87 -0
  342. package/agents/skills/product/market-analysis.skill.md +56 -0
  343. package/agents/skills/product/opportunity-assessment.skill.md +89 -0
  344. package/agents/skills/product/product-analytics.skill.md +92 -0
  345. package/agents/skills/product/professional-practice.skill.md +105 -0
  346. package/agents/skills/product/req-elicitation.skill.md +292 -0
  347. package/agents/skills/product/req-review.skill.md +280 -0
  348. package/agents/skills/product/spec-analyze.skill.md +98 -0
  349. package/agents/skills/product/spec-writer.skill.md +114 -0
  350. package/agents/skills/product/stakeholder-management.skill.md +54 -0
  351. package/agents/skills/product/sustainability-checks.skill.md +29 -0
  352. package/agents/skills/product/systems-thinking.skill.md +56 -0
  353. package/agents/skills/product/user-research-synthesis.skill.md +85 -0
  354. package/agents/skills/product/user-story-mapping.skill.md +54 -0
  355. package/agents/skills/product/visual-specs.skill.md +112 -0
  356. package/agents/skills/security/access-control.skill.md +45 -0
  357. package/agents/skills/security/agent-sandboxing.skill.md +69 -0
  358. package/agents/skills/security/ai-ethics-compliance.skill.md +46 -0
  359. package/agents/skills/security/ai-red-teaming.skill.md +74 -0
  360. package/agents/skills/security/ai-safety-guardrails.skill.md +60 -0
  361. package/agents/skills/security/api-security.skill.md +31 -0
  362. package/agents/skills/security/backup-recovery.skill.md +49 -0
  363. package/agents/skills/security/compliance-review.skill.md +152 -0
  364. package/agents/skills/security/cryptography-standards.skill.md +31 -0
  365. package/agents/skills/security/dependency-security.skill.md +47 -0
  366. package/agents/skills/security/hazard-analysis.skill.md +53 -0
  367. package/agents/skills/security/legal-review.skill.md +41 -0
  368. package/agents/skills/security/log-analysis.skill.md +68 -0
  369. package/agents/skills/security/network-security.skill.md +64 -0
  370. package/agents/skills/security/privacy-audit.skill.md +137 -0
  371. package/agents/skills/security/privacy-data-protection.skill.md +31 -0
  372. package/agents/skills/security/prompt-injection-defense.skill.md +67 -0
  373. package/agents/skills/security/reliability-engineering.skill.md +47 -0
  374. package/agents/skills/security/safety-scan.skill.md +49 -0
  375. package/agents/skills/security/secrets-management.skill.md +65 -0
  376. package/agents/skills/security/secure-architecture.skill.md +31 -0
  377. package/agents/skills/security/security-audit.skill.md +164 -0
  378. package/agents/skills/security/threat-model.skill.md +120 -0
  379. package/agents/skills/security/traceability-audit.skill.md +48 -0
  380. package/agents/templates/00-index.md +162 -0
  381. package/agents/templates/architecture/ADAPTIVE_SYSTEM_TEMPLATE.md +35 -0
  382. package/agents/templates/architecture/ADD_WORKBOOK_TEMPLATE.md +81 -0
  383. package/agents/templates/architecture/ADR_TEMPLATE.md +136 -0
  384. package/agents/templates/architecture/ARCHITECTURE_DECISION_MATRIX.md +169 -0
  385. package/agents/templates/architecture/ARCHITECTURE_REVIEW_TEMPLATE.md +241 -0
  386. package/agents/templates/architecture/ARCHITECTURE_VIEWS_TEMPLATE.md +353 -0
  387. package/agents/templates/architecture/ARCH_DECISION_FRAMEWORK.md +32 -0
  388. package/agents/templates/architecture/C4_ARCHITECTURE_TEMPLATE.md +185 -0
  389. package/agents/templates/architecture/CAPABILITY_MAP_TEMPLATE.md +153 -0
  390. package/agents/templates/architecture/CONTEXT_MAP_TEMPLATE.md +196 -0
  391. package/agents/templates/architecture/DESIGN_PATTERN_USAGE.md +24 -0
  392. package/agents/templates/architecture/DOMAIN_MODEL_TEMPLATE.md +254 -0
  393. package/agents/templates/architecture/FITNESS_FUNCTION_TEMPLATE.md +59 -0
  394. package/agents/templates/architecture/INTEGRATION_SPEC_TEMPLATE.md +38 -0
  395. package/agents/templates/architecture/LEGACY_AUDIT_TEMPLATE.md +34 -0
  396. package/agents/templates/architecture/QUALITY_ATTRIBUTES_TEMPLATE.md +210 -0
  397. package/agents/templates/architecture/SCALABILITY_ANALYSIS_TEMPLATE.md +46 -0
  398. package/agents/templates/architecture/SMART_CONTRACT_TEMPLATE.md +44 -0
  399. package/agents/templates/architecture/SYSTEM_CONTEXT_TEMPLATE.md +197 -0
  400. package/agents/templates/brain/ACTIVE_INFERENCE_LOOP_TEMPLATE.md +42 -0
  401. package/agents/templates/brain/EPISODIC_MEMORY_LOG_TEMPLATE.md +28 -0
  402. package/agents/templates/brain/GLOBAL_WORKSPACE_CONFIG_TEMPLATE.md +44 -0
  403. package/agents/templates/brain/KNOWLEDGE_MAP_TEMPLATE.md +53 -0
  404. package/agents/templates/brain/OODA_LOOP_TRACE_TEMPLATE.md +44 -0
  405. package/agents/templates/brain/PREDICTION_ERROR_LOG_TEMPLATE.md +33 -0
  406. package/agents/templates/brain/PROJECT_CONTEXT_TEMPLATE.md +36 -0
  407. package/agents/templates/brain/SELF_IMPROVEMENT_LOG_TEMPLATE.md +39 -0
  408. package/agents/templates/brain/WORKING_MEMORY_TEMPLATE.md +34 -0
  409. package/agents/templates/coding/CLEAN_CODE_CHECKLIST.md +30 -0
  410. package/agents/templates/coding/DESIGN_TOKENS_TEMPLATE.json +54 -0
  411. package/agents/templates/coding/DEVCONTAINER_TEMPLATE.json +42 -0
  412. package/agents/templates/coding/DEV_SPACE_TEMPLATE.yaml +48 -0
  413. package/agents/templates/coding/E2E_TEST_SUITE_TEMPLATE.md +36 -0
  414. package/agents/templates/coding/EVAL_PLAN_TEMPLATE.md +85 -0
  415. package/agents/templates/coding/EVAL_RUBRIC_TEMPLATE.md +57 -0
  416. package/agents/templates/coding/GOLDEN_DATASET_TEMPLATE.md +73 -0
  417. package/agents/templates/coding/PROPERTY_TEST_CHECKLIST.md +77 -0
  418. package/agents/templates/coding/SKETCH_TO_DIAGRAM_TEMPLATE.md +56 -0
  419. package/agents/templates/coding/TEST_CASE_TEMPLATE.md +28 -0
  420. package/agents/templates/coding/TEST_PLAN_TEMPLATE.md +36 -0
  421. package/agents/templates/coding/TIME_COMPLEXITY_REPORT_TEMPLATE.md +61 -0
  422. package/agents/templates/coordination/AGENTS_TEMPLATE.md +540 -0
  423. package/agents/templates/coordination/AGENT_HANDSHAKE_TEMPLATE.json +37 -0
  424. package/agents/templates/coordination/AGENT_PROFILE_TEMPLATE.md +28 -0
  425. package/agents/templates/coordination/SWARM_ARCHITECTURE_TEMPLATE.md +53 -0
  426. package/agents/templates/coordination/SWARM_CONFIG_TEMPLATE.json +45 -0
  427. package/agents/templates/core/AUDIT_LOG_TEMPLATE.md +75 -0
  428. package/agents/templates/core/BUG_REPORT_TEMPLATE.md +136 -0
  429. package/agents/templates/core/DEPENDENCY_GRAPH_TEMPLATE.md +63 -0
  430. package/agents/templates/core/ESTIMATION_TEMPLATE.md +70 -0
  431. package/agents/templates/core/MCP_CONFIG_TEMPLATE.json +807 -0
  432. package/agents/templates/core/PLAN_TEMPLATE.md +162 -0
  433. package/agents/templates/core/PROJECT_RETROSPECTIVE_TEMPLATE.md +67 -0
  434. package/agents/templates/core/SDLC_TRACKER.md +217 -0
  435. package/agents/templates/core/SESSION_SNAPSHOT_TEMPLATE.md +168 -0
  436. package/agents/templates/core/SYSTEM_ANALYSIS_TEMPLATE.md +52 -0
  437. package/agents/templates/core/TASKS_TEMPLATE.md +104 -0
  438. package/agents/templates/core/TRACEABILITY_MATRIX_TEMPLATE.md +18 -0
  439. package/agents/templates/core/WHITEBOARD_DESIGN_TEMPLATE.md +46 -0
  440. package/agents/templates/data/DATABASE_SCHEMA_TEMPLATE.md +38 -0
  441. package/agents/templates/data/DATA_PIPELINE_TEMPLATE.md +44 -0
  442. package/agents/templates/data/ONTOLOGY_TEMPLATE.md +52 -0
  443. package/agents/templates/industry/ENGINEERING_STANDARDS_REVIEW_TEMPLATE.md +43 -0
  444. package/agents/templates/industry/FHIR_INTEGRATION_TEMPLATE.md +53 -0
  445. package/agents/templates/industry/GLOBAL_STANDARDS_AUDIT_TEMPLATE.md +45 -0
  446. package/agents/templates/industry/IOT_TELEMETRY_TEMPLATE.md +59 -0
  447. package/agents/templates/industry/TELECOM_API_TEMPLATE.md +50 -0
  448. package/agents/templates/ops/BENCHMARK_REPORT_TEMPLATE.md +48 -0
  449. package/agents/templates/ops/CAPACITY_PLAN_TEMPLATE.md +41 -0
  450. package/agents/templates/ops/COST_OPTIMIZATION_REPORT_TEMPLATE.md +40 -0
  451. package/agents/templates/ops/DECOMMISSION_PLAN_TEMPLATE.md +102 -0
  452. package/agents/templates/ops/DEPLOY_CONFIG_TEMPLATE.md +88 -0
  453. package/agents/templates/ops/INCIDENT_POSTMORTEM_TEMPLATE.md +39 -0
  454. package/agents/templates/ops/INFRA_PLAN_TEMPLATE.md +37 -0
  455. package/agents/templates/ops/RELEASE_READINESS_REPORT.md +35 -0
  456. package/agents/templates/ops/RUNBOOK_TEMPLATE.md +103 -0
  457. package/agents/templates/ops/TECH_DEBT_TEMPLATE.md +102 -0
  458. package/agents/templates/product/BUSINESS_CASE_TEMPLATE.md +42 -0
  459. package/agents/templates/product/CHANGE_MANAGEMENT_PLAN_TEMPLATE.md +68 -0
  460. package/agents/templates/product/EMPATHY_MAP_TEMPLATE.md +47 -0
  461. package/agents/templates/product/EXPERIMENT_PLAN_TEMPLATE.md +84 -0
  462. package/agents/templates/product/GREEN_SOFTWARE_REPORT_TEMPLATE.md +42 -0
  463. package/agents/templates/product/IDEATION_LOG_TEMPLATE.md +42 -0
  464. package/agents/templates/product/NFR_TEMPLATE.md +31 -0
  465. package/agents/templates/product/PRD_TEMPLATE.md +222 -0
  466. package/agents/templates/product/PROBLEM_STATEMENT_TEMPLATE.md +32 -0
  467. package/agents/templates/product/REQUIREMENTS_REVIEW_TEMPLATE.md +173 -0
  468. package/agents/templates/product/SPEC_TEMPLATE.md +259 -0
  469. package/agents/templates/product/STAKEHOLDER_REGISTER_TEMPLATE.md +127 -0
  470. package/agents/templates/product/USER_STORY_MAP_TEMPLATE.md +35 -0
  471. package/agents/templates/product/VISUAL_SPEC_PACKAGE_TEMPLATE.md +207 -0
  472. package/agents/templates/security/DATASHEET_TEMPLATE.md +95 -0
  473. package/agents/templates/security/ETHICAL_IMPACT_ASSESSMENT.md +32 -0
  474. package/agents/templates/security/HAZARD_LOG.md +17 -0
  475. package/agents/templates/security/MODEL_CARD_TEMPLATE.md +120 -0
  476. package/agents/templates/security/SAFETY_CASE.md +39 -0
  477. package/agents/templates/security/SAFETY_POLICY.md +25 -0
  478. package/agents/templates/security/SECURITY_CHECKLIST.md +157 -0
  479. package/agents/templates/security/SYSTEM_CARD_TEMPLATE.md +139 -0
  480. package/agents/templates/security/THREAT_MODEL_TEMPLATE.md +171 -0
  481. package/bin/install.js +572 -0
  482. package/docs/CLI_REFERENCE.md +527 -0
  483. package/docs/INSTALL.md +68 -0
  484. package/docs/MCP_CONFIGURATIONS.md +1765 -0
  485. package/docs/PLATFORM_CONTROLS.md +324 -0
  486. package/docs/POST_INSTALL.md +75 -0
  487. package/docs/QUICK_COMMANDS.md +131 -0
  488. package/docs/QUICK_GUIDE.md +914 -0
  489. package/docs/README_FULL.md +1451 -0
  490. package/docs/SCHEMA.md +164 -0
  491. package/docs/V1_RELEASE_CHECKLIST.md +144 -0
  492. package/docs/VERIFICATION_GUIDE.md +283 -0
  493. package/docs/adr/0001-record-architecture-decisions.md +26 -0
  494. package/docs/adr/0002-brain-active-inference-runtime.md +27 -0
  495. package/docs/adr/0003-additive-only-emitter-schema-evolution.md +33 -0
  496. package/docs/assets/GABBE.png +0 -0
  497. package/docs/gabbe.config.example.json +9 -0
  498. package/docs/reports/duplication-report.md +30 -0
  499. package/install.ps1 +90 -0
  500. package/install.sh +89 -0
  501. package/package.json +49 -0
  502. package/scripts/init.py +1036 -0
@@ -0,0 +1,85 @@
1
+ ---
2
+ name: user-research-synthesis
3
+ description: Synthesize raw user-research data into personas, a journey map, and prioritized insight statements.
4
+ triggers: [synthesize user research, affinity mapping interviews, jobs to be done analysis, build user personas, empathy and journey mapping, write insight statements, prioritize research findings]
5
+ tags: [product, research, ux]
6
+ core: false
7
+ context_cost: medium
8
+ ---
9
+ # User Research Synthesis
10
+
11
+ ## Goal
12
+ Convert raw, messy research data — interview notes, survey responses, support transcripts, observation
13
+ logs — into structured, decision-ready artifacts: synthesized personas, a journey map, and a prioritized
14
+ list of insight statements. Synthesis is the act of finding the signal in noise: turning individual
15
+ anecdotes into patterns, and patterns into insights that change what the team builds. The discipline is
16
+ staying grounded in evidence so insights are *discovered*, not invented.
17
+
18
+ ## Steps
19
+ 1. **Tag and affinity-map the raw data.**
20
+ - Break notes into atomic observations (one quote or fact per unit), then cluster by similarity into
21
+ affinity groups. Name each group by the pattern it represents, not the topic it covers.
22
+ - Track how many distinct participants support each cluster — a pattern seen once is an anecdote, a
23
+ pattern seen across many participants is a finding.
24
+ 2. **Extract Jobs-To-Be-Done (JTBD).**
25
+ - For recurring goals, write job statements in the canonical form: "When [situation], I want to [motivation],
26
+ so I can [expected outcome]." JTBD captures the progress a user is trying to make, independent of any
27
+ specific solution, which keeps the team from over-fitting to current features.
28
+ 3. **Synthesize personas.**
29
+ - Build personas as composites grounded in the clusters — goals, jobs, pains, and context — not
30
+ demographic stereotypes. Each persona attribute should trace to observed evidence.
31
+ - Keep the set small (2-4); too many personas dilute focus.
32
+ 4. **Map the journey.**
33
+ - For the primary persona, lay out the end-to-end journey in stages. For each stage capture actions,
34
+ thoughts, emotional highs/lows, and pain points. Mark the moments of greatest friction — these are
35
+ where the journey map earns its keep.
36
+ 5. **Write insight statements.**
37
+ - Convert findings into insight statements that pair an observation with its implication:
38
+ "We observed [pattern]; this matters because [consequence]; therefore [opportunity]." An insight is
39
+ not a raw fact — it carries a "so what."
40
+ 6. **Prioritize the insights.**
41
+ - Rank insights by evidence strength (how many participants) and potential impact, separating
42
+ high-confidence, high-impact insights from interesting-but-thin ones.
43
+
44
+ ## Constraints
45
+ - Every persona attribute, journey pain point, and insight MUST trace back to observed data; the agent
46
+ MUST NOT invent quotes, metrics, or behaviors to fill gaps, and labels any inferred claim as inference.
47
+ - Distinguish what users *say* from what they *do*; weight observed behavior over stated preference when
48
+ they conflict.
49
+ - Note the evidence base for each insight (number of participants, method) so consumers can judge how far
50
+ to trust it — a finding from 2 interviews is not the same as one from 30.
51
+ - Personas are synthesized composites, never real individuals; do not reproduce a single participant as a
52
+ persona.
53
+
54
+ ## Output Format
55
+ Produce a synthesis package containing:
56
+ - 2-4 personas, each with goals, primary JTBD statements, key pains, and context, every attribute grounded
57
+ in evidence.
58
+ - A journey map for the primary persona: stages with actions, emotions, and flagged friction points.
59
+ - A prioritized list of insight statements (observation → implication → opportunity), each tagged with its
60
+ evidence strength (participant count / method) and a confidence level.
61
+
62
+ ## Security & Guardrails
63
+
64
+ ### 1. Skill Security
65
+ - **Risk**: Confirmation-biased synthesis — the agent surfaces only patterns matching a stakeholder's prior
66
+ hypothesis and buries disconfirming data; mitigation: the agent MUST report contradicting observations
67
+ alongside supporting ones and never drop a cluster solely because it conflicts with a desired conclusion.
68
+ - **Risk**: Evidence inflation — a single anecdote is presented as a widespread pattern; mitigation: every
69
+ finding carries its participant count and the agent refuses to label an n=1 observation as a "pattern."
70
+
71
+ ### 2. System Integration Security
72
+ - **Risk**: PII exposure — raw research contains real names, contact details, recordings, and sensitive
73
+ disclosures; mitigation: the agent strips and pseudonymizes all participant identifiers before any
74
+ persona or insight is emitted, and keeps raw transcripts in a restricted working artifact only.
75
+ - **Risk**: Re-identification — a "synthetic" persona reproduces one participant closely enough to identify
76
+ them; mitigation: the agent composites across multiple participants and removes uniquely identifying
77
+ detail so no persona maps to a single real person.
78
+
79
+ ### 3. LLM & Agent Guardrails
80
+ - **Risk**: Hallucinated evidence — the model fabricates a verbatim quote or statistic to support a tidy
81
+ insight; mitigation: quotes are used only verbatim from source and any number must trace to data, with
82
+ unsupported claims marked as inference rather than fact.
83
+ - **Risk**: Sampling and demographic bias — synthesis over-represents the loudest or most available segment
84
+ and erases under-served users; mitigation: the agent reports who is and is not represented in the sample
85
+ and flags segments with no coverage as gaps rather than silently generalizing.
@@ -0,0 +1,54 @@
1
+ ---
2
+ name: user-story-mapping
3
+ description: Visualizing user journeys and slicing deliverables into releases.
4
+ triggers: [user story mapping]
5
+ tags: [product]
6
+ role: prod-pm
7
+ context_cost: low
8
+ ---
9
+ # user-story-mapping Skill
10
+
11
+ ## Goal
12
+ This skill guides the creation of a 2D map of user needs (The Backbone) vs details (The Body) to scope releases.
13
+
14
+ ## Steps
15
+ ## 1. The Matrix (Backbone vs Body)
16
+ - **X-Axis (Backbone)**: The narrative flow. "User Signs Up" -> "Searches Product" -> "Adds to Cart" -> "Checks Out".
17
+ - **Y-Axis (Priority)**: Depth of implementation.
18
+ - *Slice 1 (MVP)*: "Sign up via Email".
19
+ - *Slice 2 (v1.1)*: "Sign up via Google".
20
+ - *Slice 3 (v2)*: "Sign up via SSO".
21
+
22
+ ## 2. Process
23
+ 1. **Define the Persona**: Who are we mapping for?
24
+ 2. **Map the Steps (Activities)**: High-level goals (e.g., "Manage Profile").
25
+ 3. **Break into Tasks**: Concrete actions (e.g., "Upload Avatar").
26
+ 4. **Slice Lines**: Draw horizontal lines to define Release 1, Release 2, Release 3.
27
+
28
+ ## 3. Output Format
29
+ Use `templates/product/USER_STORY_MAP_TEMPLATE.md` to document the outcome.
30
+
31
+ ## 4. The "Walking Skeleton" Strategy
32
+ - **Goal**: Implement a tiny slice of the *entire* backbone first.
33
+ - **Why**: Proves end-to-end connectivity (Frontend -> Backend -> DB) in Week 1.
34
+ - **Example**: "User clicks Buy -> DB records order -> Payment Stubbed -> Email Sent".
35
+
36
+ ## 5. Prioritization Rules (MoSCoW)
37
+ - **Must Have**: Without this, it doesn't work.
38
+ - **Should Have**: Important constraints (Performance, key features).
39
+ - **Could Have**: Delightful add-ons.
40
+ - **Won't Have**: Out of scope for now.
41
+
42
+ ## Security & Guardrails
43
+
44
+ ### 1. Skill Security (User Story Mapping)
45
+ - **MVP Security Mandate**: The agent must enforce the rule that "Security is not an add-on." Basic authentication, authorization, and data validation tasks *must* be included in Slice 1 (MVP). The agent must actively fight prompts that attempt to push foundational security controls into "v1.1" or "v2".
46
+ - **Backbone Vulnerability Analysis**: Once the "Walking Skeleton" (The Backbone) is defined, the agent must trigger a specialized prompt to identify the single highest-risk security node in that end-to-end chain (e.g., the payment stub API interface) and mark it for immediate validation.
47
+
48
+ ### 2. System Integration Security
49
+ - **Slice Dependency Integrity**: The agent must prevent the user from defining an integration sequence that causes a security paradox. (For example, attempting to place "Implement User Data Export" in Release 1, but "Implement User Authentication" in Release 2).
50
+ - **Graceful Degradation Mapping**: When prioritizing via MoSCoW, the agent must systematically map failure states. If a "Must Have" component fails, what is the defined fallback? The story map must include "Error Handling" stories as native "Body" elements, not as afterthoughts.
51
+
52
+ ### 3. LLM & Agent Guardrails
53
+ - **Persona Privilege Escalation**: The LLM must verify that the defined Persona operating the User Journey map does not require unauthorized privilege escalation to complete the targeted "Activities" (e.g., mapping a standard "Client Persona" performing an "Admin Database Dump" activity).
54
+ - **Scope Creep Hallucination**: The agent must not spontaneously invent new "Could Have" features to make the story map look more comprehensive. The map must strictly represent the validated constraints defined in the original `PRD.md` or `BUSINESS_CASE.md`.
@@ -0,0 +1,112 @@
1
+ ---
2
+ name: visual-specs
3
+ description: Process visual inputs (scans, sketches, diagrams) into structured product specifications before implementation.
4
+ triggers: [visual spec, design phase, visual requirements, scan to spec, diagram spec, visual design, sketch spec]
5
+ tags: [product, design, visual, architecture, requirements]
6
+ context_cost: high
7
+ ---
8
+ # Visual Specs Skill
9
+
10
+ ## Goal
11
+ Process visual inputs — whiteboard photos, napkin sketches, architecture diagrams, UI wireframes, state machines, ER models — into structured, version-controlled design artifacts that feed into the PRD and Spec templates. This skill produces a **Visual Spec Package** that serves as the design readiness gate before implementation.
12
+
13
+ ## Prerequisites
14
+ - **Image Recognition MCP**: `image-recognition` must be configured (requires vision API key)
15
+ - **At least one output MCP** (recommended: all):
16
+ - `excalidraw` — for architecture diagrams
17
+ - `tldraw` — for UI wireframes
18
+ - `drawio` — for enterprise diagrams (optional)
19
+ - Mermaid — no MCP required (text-based)
20
+ - **Approved PRD**: This skill runs after PRD approval (S01 complete)
21
+
22
+ ## Steps
23
+
24
+ ### Phase 1: Collect Visual Inputs
25
+ 1. **Inventory all visual inputs** provided by the user:
26
+ - Whiteboard photos, napkin drawings, paper wireframes
27
+ - Architecture sketches, system diagrams, flowcharts
28
+ - State machine drawings, ER model sketches
29
+ - Figma exports, existing screenshots, mockup images
30
+ 2. **Classify each input** by diagram type:
31
+ - Architecture | Flowchart | State | ER | Sequence | Wireframe | Component | Deployment | Mind map
32
+ 3. **Document in Visual Spec Package**: Section 6 (Visual Input Log)
33
+
34
+ ### Phase 2: Recognize and Structure
35
+ 4. **Process each visual input** through `image-recognition` MCP:
36
+ - Identify shapes, text labels, connections, groupings
37
+ - Classify diagram type if not already known
38
+ 5. **Structure recognized elements** into typed nodes and edges
39
+ 6. **Present structured list to user** for verification (MANDATORY)
40
+ 7. **Apply corrections** based on user feedback
41
+
42
+ ### Phase 3: Generate Formal Diagrams
43
+ 8. **Route each diagram to appropriate output tool**:
44
+
45
+ | Diagram Type | Default Tool | Alternative |
46
+ |---|---|---|
47
+ | Architecture (C4) | Mermaid `flowchart` | Excalidraw (>15 nodes) |
48
+ | UI wireframes | tldraw | — |
49
+ | Flowcharts | Mermaid `flowchart` | Excalidraw (complex) |
50
+ | State machines | Mermaid `stateDiagram-v2` | — |
51
+ | ER diagrams | Mermaid `erDiagram` | Draw.io (detailed) |
52
+ | Sequence diagrams | Mermaid `sequenceDiagram` | — |
53
+ | Component diagrams | Mermaid `graph TD` | Excalidraw |
54
+ | Deployment diagrams | Draw.io | Excalidraw |
55
+ | Mind maps | tldraw | Excalidraw |
56
+
57
+ 9. **Generate formal diagrams** using the selected tool
58
+ 10. **Verify output** — call `getFullDiagramState` (Excalidraw) or `tldraw_read` (tldraw) to confirm
59
+
60
+ ### Phase 4: Assemble Visual Spec Package
61
+ 11. **Fill `VISUAL_SPEC_PACKAGE_TEMPLATE.md`**:
62
+ - Section 1: System Architecture diagram(s)
63
+ - Section 2: Workflow & State diagrams
64
+ - Section 3: UI/UX wireframes
65
+ - Section 4: Data Model / ER diagrams
66
+ - Section 5: Use Case / Sequence diagrams
67
+ - Section 6: Visual Input Log (original images + recognition results)
68
+ - Section 7: Design Readiness Checklist
69
+ 12. **Update PRD** — link generated diagrams in Sections 6, 7, 8
70
+ 13. **Update SPEC** — embed architecture and data model diagrams
71
+
72
+ ### Phase 5: Design Readiness Gate
73
+ 14. **Run Design Readiness Checklist** (all items must pass):
74
+ - Architecture diagram present and verified
75
+ - Primary user workflows diagrammed
76
+ - Data model documented (if DB involved)
77
+ - UI wireframes created (if user-facing)
78
+ - State diagrams for entities with >3 states
79
+ - All visual inputs processed and documented
80
+ - All diagrams committed to version control
81
+ 15. **Present for human approval** before proceeding to S02/S03
82
+
83
+ ## Deliverables
84
+ - `VISUAL_SPEC_PACKAGE_TEMPLATE.md` — filled with all visual design artifacts
85
+ - Generated diagram files (`.excalidraw`, `.tldr`, `.drawio`, Mermaid in `.md`)
86
+ - Updated PRD and Spec templates with diagram references
87
+ - Design Readiness Checklist passed
88
+
89
+ ## Integration with Existing Skills
90
+
91
+ | Skill | Interaction |
92
+ |---|---|
93
+ | `spec-writer.skill` | Visual specs feed into SPEC_TEMPLATE.md |
94
+ | `req-review.skill` | Reviews visual specs for completeness |
95
+ | `excalidraw.skill` | Generates architecture diagrams |
96
+ | `tldraw-canvas.skill` | Generates UI wireframes |
97
+ | `sketch-to-diagram.skill` | Processes individual scans |
98
+ | `design-thinking.skill` | Empathy maps → visual user flows |
99
+ | `user-story-mapping.skill` | Use case diagrams link to user stories |
100
+
101
+ ## Security & Guardrails
102
+
103
+ ### 1. Visual Input Security
104
+ - Scanned images may contain sensitive information (credentials, internal IPs, customer data)
105
+ - The agent MUST warn before sending whiteboard photos to external vision APIs
106
+ - For sensitive diagrams: use verbal description or local Tesseract OCR
107
+
108
+ ### 2. Agent Guardrails
109
+ - **Human verification is MANDATORY** after recognition, before formal diagram generation
110
+ - **Design Readiness Gate** must pass before proceeding to implementation
111
+ - **Text-first rule**: Default to Mermaid (text-based, git-diffable) unless visual complexity demands otherwise
112
+ - **State sync**: All visual diagrams must have corresponding text descriptions in PRD/SPEC
@@ -0,0 +1,45 @@
1
+ ---
2
+ name: access-control
3
+ description: Design RBAC/ABAC models, Principle of Least Privilege, Identity Management.
4
+ triggers: [access control]
5
+ tags: [security]
6
+ role: ops-security, eng-backend
7
+ context_cost: low
8
+ ---
9
+ # access-control Skill
10
+
11
+ ## Goal
12
+ Identity is the new perimeter. Authorization must be strict, consistent, and auditable.
13
+
14
+ ## Steps
15
+ ## 1. Models
16
+ - **RBAC (Role-Based)**: `Admin`, `Editor`, `Viewer`. Simple, coarse-grained.
17
+ - **ABAC (Attribute-Based)**: `CanEdit if user.dept == doc.dept AND time < 5pm`. Flexible, complex.
18
+ - **ReBAC (Relationship-Based)**: `CanView if user is friend_of document.owner`. (Graph-based, e.g., Zanzibar).
19
+
20
+ ## 2. Best Practices
21
+ - **Least Privilege**: Start with `Deny All`. Explicitly `Allow` specific actions.
22
+ - **Decoupling**: Decouple logic (`if user.isAdmin`) from policy (`can(user, 'delete:report')`). Use libraries like CASL or Oso.
23
+ - **No Hardcoded IDs**: Never checks `if (user.id === '123')`.
24
+
25
+ ## 3. Infrastructure IAM
26
+ - **Service Accounts**: Separate identity for apps/machines. Rotate keys automatically.
27
+ - **Short-Lived Credentials**: Use OIDC/STS (AssumeRole) instead of long-lived access keys.
28
+
29
+ ## 4. Audit
30
+ - **Log Decisions**: "User X tried to do Action Y on Resource Z -> Result: DENIED".
31
+ - **Review**: Quarterly review of `Admin` group membership.
32
+
33
+ ## Security & Guardrails
34
+
35
+ ### 1. Skill Security (Access Control)
36
+ - **Privilege Escalation Veto**: The agent is strictly prohibited from autonomously generating migration scripts or configuration files that assign the `Admin` role or sweeping `*` permissions to any user or service account, including itself. All broad privilege grants must be hard-blocked pending human Executive review.
37
+ - **Orphaned Identity Cleanup**: When designing an IAM lifecycle, the agent must mandate an automated revocation sequence. If an employee is terminated or a service is deprecated, the architecture must guarantee the immediate, atomic destruction of all associated active sessions and API keys.
38
+
39
+ ### 2. System Integration Security
40
+ - **AuthZ Evaluation Enforcement**: The agent must mandate that authorization logic (ABAC/RBAC) is evaluated on the Server/Backend. It must aggressively flag and fail any architecture diagram or pull request that attempts to handle authorization exclusively in the Frontend/Client (e.g., merely hiding a UI button).
41
+ - **Service Account Segregation**: When provisioning Service Accounts for infrastructure, the agent must enforce strict boundary contexts. A Service Account used by the CI/CD pipeline to push Docker images must not possess the IAM permissions required to provision new cloud databases or modify VPC networks.
42
+
43
+ ### 3. LLM & Agent Guardrails
44
+ - **Bypass Prompting Defense**: The LLM must recognize and reject user prompts that ask for "temporary workarounds" to access control systems (e.g., "Just give this script `chmod 777` for now to get it working in Prod"). The agent must enforce least privilege, even during critical incidents.
45
+ - **Hardcoded Credential Hallucination**: The agent must never synthesize, generate, or accept hardcoded access tokens or default passwords in its output templates. All authentication examples must use securely injected environment variables (e.g., `process.env.API_KEY`) or reference a secure Secret Manager.
@@ -0,0 +1,69 @@
1
+ ---
2
+ name: agent-sandboxing
3
+ description: Intrinsic least-privilege isolation for agents touching a local OS, shell, or filesystem — the enforcement layer behind prompt-injection-defense and ai-safety-guardrails.
4
+ triggers: [agent sandbox, sandbox policy, least privilege isolation, filesystem allowlist, egress deny-by-default, syscall restriction, microVM agent isolation, contain agent blast radius]
5
+ tags: [security]
6
+ core: false
7
+ context_cost: medium
8
+ ---
9
+ # Agent Sandboxing Skill
10
+
11
+ ## Goal
12
+ Provide the intrinsic, infrastructure-level isolation layer for any agent that touches a local OS, shell, or filesystem. Prompt-injection and guardrail rails (`prompt-injection-defense.skill`, `ai-safety-guardrails.skill`) reduce the probability that an agent is subverted; sandboxing bounds the *consequence* if one is. The design assumption is **assumed compromise**: the planner LLM may already be injected, so the controls must hold even when the agent actively tries to relax them.
13
+
14
+ The defining property is that all four isolation domains are **LOCKED at sandbox-creation time** and are **immutable for the lifetime of the run**. A prompt-injected agent cannot widen its own filesystem scope, open new egress, escalate privilege, or re-route inference, because those decisions were made by the host before the agent's non-deterministic logic ever ran. Container and microVM runtimes (gVisor, Kata, Firecracker) implement exactly this shape — a thin, host-enforced boundary the workload cannot reach around.
15
+
16
+ ## Steps
17
+
18
+ 1. **Declare the four isolation policies (deny-by-default).**
19
+ Author a single declarative policy covering all four domains. Everything not explicitly allowed is denied.
20
+ - **Filesystem**: read/write permitted ONLY inside explicit path allowlists (e.g. the task workspace and a scratch dir). Host binaries, system config directories, credential stores, SSH keys, and the agent's own runtime config are unreachable — not merely read-only, but absent from the mount namespace.
21
+ - **Network**: egress is deny-by-default. A forward proxy is the only route out, and it enforces allow rules at **HTTP method + path granularity** (not just host/port), so `GET /v1/models` can be allowed while `POST /exfil` on the same host is refused. No raw sockets, no DNS to arbitrary resolvers.
22
+ - **Process**: no privilege escalation (no setuid, `no_new_privs` set), a restrictive syscall filter blocking dangerous calls (ptrace, mount, kernel-module load, raw `clone` for daemons), no spawning of background daemons or detached processes, and capped CPU/memory/PID/file-descriptor limits.
23
+ - **Inference routing**: all model calls are routed through controlled backends so prompt context and tool output stay within the trust boundary. The agent cannot redirect inference to an attacker-chosen endpoint, which would itself be an exfiltration channel.
24
+
25
+ 2. **Lock the policy at sandbox creation.**
26
+ The host (runtime supervisor / `policy` engine) materializes the sandbox from the declared policy and seals it BEFORE handing control to the agent. The four domains become immutable for the run: there is no in-band API, tool, or syscall by which the agent can edit its own allowlists, add egress rules, drop the syscall filter, or change the inference backend. Policy changes require tearing down and recreating the sandbox from outside — never a mid-run mutation. This is what makes the boundary robust to prompt injection: the attacker can only ask the agent to do something, and the agent has no privilege to grant it.
27
+
28
+ 3. **Verify escape attempts fail (negative testing).**
29
+ Treat the sandbox as untrusted-until-proven and run the escape-verification checklist (below) as part of provisioning and CI. Each probe MUST fail closed. A probe that succeeds is a sandbox defect, not an agent bug. On any verified escape or a runtime policy-violation signal, hand off to `self-heal.skill` to quarantine the sandbox, snapshot evidence for audit, and rebuild from the sealed policy rather than patching the live instance.
30
+
31
+ ## Constraints
32
+ - Policies are immutable mid-run. If a legitimate task genuinely needs broader scope, the correct path is to tear down and recreate a new sandbox with a new sealed policy under human review — NEVER to relax a live one.
33
+ - Deny-by-default everywhere. An empty or missing allowlist means "nothing," not "everything."
34
+ - Network allow rules are matched at method + path granularity; host/port allowlisting alone is insufficient because it leaves wide exfiltration paths open on otherwise-legitimate hosts.
35
+ - Read-only is not enough for the filesystem: sensitive host paths must be absent from the namespace, so they cannot be enumerated or leaked even when only readable.
36
+ - Sandboxing bounds blast radius; it does not replace the upstream rails. Pair it with `prompt-injection-defense.skill` (so fewer compromises happen) and HITL gates on trifecta-complete actions (so the remaining ones are caught).
37
+ - The inference route is part of the trust boundary: an unrestricted model endpoint is an egress channel and must be locked like any other.
38
+
39
+ ## Output Format
40
+ 1. **Declarative sandbox policy** — a single document with four sections:
41
+ - `filesystem`: read paths, write paths, explicitly-denied/absent paths.
42
+ - `network`: proxy endpoint + ordered allow rules as `METHOD path-prefix → allow|deny`, default `deny`.
43
+ - `process`: privilege flags (`no_new_privs`, no setuid), syscall allow/deny set, resource caps (CPU, memory, PIDs, FDs), daemon-spawn = denied.
44
+ - `inference`: approved backend(s) and the enforcement point that pins them.
45
+ Plus `runtime`: the chosen isolation runtime (gVisor / Kata / Firecracker / equivalent) and the `lock-at-creation: true` flag.
46
+ 2. **Escape-verification checklist** — each item expected to FAIL CLOSED:
47
+ - [ ] Write outside the filesystem allowlist is refused.
48
+ - [ ] Read of a host credential/config path returns "not found" (path absent from namespace).
49
+ - [ ] Egress to a non-allowlisted host is blocked at the proxy.
50
+ - [ ] An allowlisted host but non-allowlisted method+path is blocked.
51
+ - [ ] Privilege escalation (setuid / capability gain) is refused.
52
+ - [ ] A blocked syscall (ptrace / mount / module load) is refused.
53
+ - [ ] Spawning a detached daemon/background process is refused.
54
+ - [ ] An attempt to re-point inference at an external endpoint is refused.
55
+ - [ ] An in-band attempt to edit any of the four policies mid-run has no effect (immutability holds).
56
+
57
+ ## Security & Guardrails
58
+
59
+ ### 1. Skill Security
60
+ - **Risk**: a self-modifying or overly-permissive policy lets the agent widen its own scope — mitigation: policies are declarative, externally owned, and sealed by the host at creation; there is no in-band mutation path, and policy changes require an out-of-band teardown-and-recreate under review.
61
+ - **Risk**: a default-allow gap (empty allowlist read as "everything") silently grants full access — mitigation: deny-by-default is the only default; provisioning fails closed if any of the four domains is undeclared, and the escape checklist must pass before the sandbox is handed to the agent.
62
+
63
+ ### 2. System Integration Security
64
+ - **Risk**: a sandbox escape pivots into the host OS, credential stores, or the broader network — mitigation: enforce isolation with a host-level runtime (gVisor / Kata / Firecracker), keep sensitive host paths absent from the namespace, route all egress through a method+path-granular proxy, and block ptrace/mount/module-load syscalls so kernel-boundary escapes have no primitive.
65
+ - **Risk**: the model endpoint becomes an unmonitored exfiltration channel — mitigation: pin inference to controlled backends inside the trust boundary, treat the inference route as egress, and refuse any redirect to an external endpoint.
66
+
67
+ ### 3. LLM & Agent Guardrails
68
+ - **Risk**: a prompt-injected agent is socially engineered into trying to relax its own sandbox — mitigation: the four domains are immutable mid-run, so the agent has nothing to grant; the worst case is a refused request, not a widened boundary, and the refusal is logged as a violation signal.
69
+ - **Risk**: a compromised agent spawns daemons or background processes to outlive the run or stage exfiltration — mitigation: daemon/detached-process spawning is denied by the syscall filter and PID caps, and any violation triggers `self-heal.skill` to quarantine, snapshot, and rebuild from the sealed policy rather than patch the live instance.
@@ -0,0 +1,46 @@
1
+ ---
2
+ name: ai-ethics-compliance
3
+ description: Assessing Bias, Fairness, Transparency, and Legal (EU AI Act, GDPR).
4
+ triggers: [ai ethics compliance]
5
+ tags: [security]
6
+ role: prod-ethicist
7
+ context_cost: low
8
+ ---
9
+ # ai-ethics-compliance Skill
10
+
11
+ ## Goal
12
+ This skill ensures the system respects human rights and legal frameworks.
13
+
14
+ ## Steps
15
+ ## 1. Regulatory Checklist (EU AI Act / GDPR)
16
+ - **Transparency**: Does the user know they are interacting with AI? (Art 52).
17
+ - **Explanation**: Can we explain *why* the AI made a decision? (Right to Explanation).
18
+ - **Data Rights**: Can the user request their data be deleted? (Right to be Forgotten).
19
+ - **Risk Category**: Is this "High Risk" (CV recruitment, credit scoring)? If yes, Conformity Assessment required.
20
+
21
+ ## 2. Bias Testing
22
+ - **Demographic Parity**: Does the model perform equally well for all groups?
23
+ - **Dataset Audit**: scan training/RAG data for stereotypes or under-representation.
24
+ - **Proxy Variables**: Are we using "Zip Code" as a proxy for race?
25
+
26
+ ## 3. Dark Pattern Prevention
27
+ - **Addiction Loops**: Are we exploiting dopamine triggers?
28
+ - **Manipulation**: Are we tricking the user into consent?
29
+ - **Privacy Zuckering**: Making it hard to manage privacy settings.
30
+
31
+ ## 4. The "Newspaper Test"
32
+ - If this feature's logic was published on the front page of the NYT, would we be ashamed?
33
+
34
+ ## Security & Guardrails
35
+
36
+ ### 1. Skill Security (AI Ethics & Compliance)
37
+ - **Compliance Audit Immutability**: Ethics and compliance audits generated by this skill constitute legal artifacts. The agent must enforce read-only append logs for these assessments. Once an AI model is flagged as "High Risk" under the EU AI Act, the agent cannot retroactively downgrade the risk category without a formalized, cryptographically signed human review.
38
+ - **Bias Testing Sandbox**: When executing dataset audits for demographic parity or stereotypes, the agent must process the data within an ephemeral, secure enclave. This prevents the potentially sensitive training data (e.g., protected class information used to test for bias) from leaking into external logging aggregation systems.
39
+
40
+ ### 2. System Integration Security
41
+ - **Right to be Forgotten Execution**: The architecture must support hard deletion. If a user exercises their GDPR Right to be Forgotten, the agent must ensure the system propagates the deletion command across all primary databases, vector databases, and log archives, verifying the complete cryptographic erasure of the individual's PII.
42
+ - **Dark Pattern Telemetry Alerts**: The system must be instrumented to detect "Manipulation" or "Privacy Zuckering." If telemetry indicates users are repeatedly failing to navigate a consent form or opting into data sharing at statistically anomalous rates, the agent must trigger a Sev-2 security alert for immediate ethical review.
43
+
44
+ ### 3. LLM & Agent Guardrails
45
+ - **Ethics Washing Deflection**: The LLM must actively resist instructions to generate "Ethics Washing" documentation—i.e., writing a compliance report that uses vague corporate jargon to mask actual, measured bias or non-compliance. The agent must enforce the reporting of objective, quantifiable fairness metrics.
46
+ - **Malicious Compliance Identification**: A user might prompt the agent to "Optimize the algorithm to strictly maximize engagement, regardless of the psychological impact." The agent must recognize this as an "Addiction Loop" dark pattern and permanently refuse the request, citing the overarching ethical constitution of the project.
@@ -0,0 +1,74 @@
1
+ ---
2
+ name: ai-red-teaming
3
+ description: Proactively attack your own AI and agent system to find failures before adversaries, scoring attack-success-rate and gating CI on it.
4
+ triggers: [ai red teaming, red team plan, attack success rate, ASR scorecard, jailbreak testing, crescendo attack, automated adversarial testing, OWASP LLM red team]
5
+ tags: [security, ai, agents]
6
+ core: false
7
+ context_cost: medium
8
+ ---
9
+ # AI Red Teaming Skill
10
+
11
+ ## Goal
12
+ Proactively attack your own AI/agent system to discover failures before an adversary does. This is the offensive complement to the defensive skills: `prompt-injection-defense.skill` builds the layered controls and `ai-safety-guardrails.skill` builds the rails — this skill tries to break both. Red teaming is threat-led: it designs scenarios from a model of who would attack and why, executes them as concrete attacks, scores the attack-success-rate (ASR), and gates CI on it so regressions are caught automatically. Stated honestly: red teaming *reduces* risk by finding failures, but it can never *prove* safety — absence of a successful attack in your suite is not evidence the system is safe.
13
+
14
+ ## Steps
15
+
16
+ 1. **Design threat-led scenarios mapped to OWASP LLM Top 10**
17
+ - Start from threat actors and goals (insider, untrusted external content author, compromised tool/MCP server, end user seeking disallowed output), not from a generic checklist. Reuse the surface enumeration from `threat-model.skill` rather than re-deriving it.
18
+ - Map each scenario to the relevant OWASP LLM Top 10 category (e.g. LLM01 Prompt Injection, LLM02 Sensitive Information Disclosure, LLM06 Excessive Agency, LLM07 System Prompt Leakage) so coverage is auditable and gaps are visible.
19
+
20
+ 2. **Build the attack technique library**
21
+ - **Direct injection**: malicious instructions in the user's own prompt.
22
+ - **Indirect injection**: instructions hidden in ingested content — RAG documents, web fetches, file contents, tool output, tool metadata, image alt-text.
23
+ - **Jailbreaks**: persona/role-play, obfuscation, encoding, and refusal-bypass prompts targeting the safety policy.
24
+ - **Multi-turn / crescendo**: benign-looking openers that escalate across turns to a disallowed goal that a single-turn filter would miss.
25
+ - **Data exfiltration**: weaponizing a benign tool (link unfurl, image fetch, URL parameter, DNS) as a covert channel for private data the agent holds.
26
+ - **Tool abuse / excessive agency**: coercing the agent into unauthorized, irreversible, or out-of-scope tool calls.
27
+ - **Training-data extraction**: probing for memorized PII, secrets, or proprietary text in model outputs.
28
+
29
+ 3. **Automate with red-team tooling**
30
+ - Use an orchestration framework to generate and run attacks at scale (named generically: PyRIT for orchestration and scoring, Garak for probe-based vulnerability scanning).
31
+ - Codify the highest-value, reproducible attacks as a regression suite that runs in CI (named generically: Promptfoo for assertion-based red-team tests in the pipeline).
32
+ - Combine seed prompts with automated mutation/transformation so the suite explores variants rather than a fixed list an implementer can overfit to.
33
+
34
+ 4. **Score attack-success-rate (ASR) and gate CI**
35
+ - Define success per scenario with an explicit, automatable judge (a successful exfiltration, a policy-violating completion, a leaked system prompt). Where an LLM judge scores attacks, apply `llm-as-judge.skill` and spot-check the judge against human labels.
36
+ - Compute ASR = successful attacks / total attempts, broken out per OWASP category and per technique.
37
+ - Gate the pipeline: fail CI when ASR on a category exceeds its agreed threshold, or when ASR regresses versus the last green run. Pin thresholds so "we made it slightly worse" cannot ship silently.
38
+
39
+ 5. **Run human red-teaming for the long tail**
40
+ - Automated suites miss creative, context-specific, and socially-engineered attacks. Schedule human red-team exercises for novel scenarios, then promote anything they find into the automated regression suite so it can never recur unnoticed.
41
+
42
+ 6. **Make it continuous**
43
+ - Re-run the suite on every model, prompt, tool, or retrieval-source change — any of these can reopen a closed vulnerability. Treat red teaming as a standing process tied to release, not a one-time audit.
44
+
45
+ ## Constraints
46
+ - NEVER present a passing red-team run as proof of safety. Report it as "no successful attack in this suite as of this run" — a falsifiable claim, not a guarantee.
47
+ - NEVER let the system under test see the attack corpus at train/eval time; a model that memorizes the suite produces a flattering ASR that does not transfer to real adversaries.
48
+ - Red teaming is the offensive complement to, not a replacement for, the structural defenses in `prompt-injection-defense.skill`. Findings drive fixes there; they do not substitute for them.
49
+ - Run attacks only against systems and data you are authorized to test, in an isolated environment; never exfiltrate real user data or trigger real external side effects during a test.
50
+ - An ASR gate is only as honest as its judge — a weak judge inflates safety. Validate the judge before trusting the gate.
51
+
52
+ ## Output Format
53
+ A red-team plan plus an ASR scorecard:
54
+ - **Threat-led scenarios**: actors, goals, and the OWASP LLM category each maps to.
55
+ - **Technique coverage**: which attack families are exercised and which are out of scope (stated explicitly).
56
+ - **Tooling**: orchestration/scanner/CI-regression tools and where each runs.
57
+ - **ASR scorecard**: a table of OWASP category × technique × attempts × successes × ASR, with the prior run for comparison.
58
+ - **CI gates**: the per-category ASR thresholds and the regression rule.
59
+ - **Findings & fixes**: each successful attack, its severity, the owning defensive skill, and remediation status.
60
+ - **Residual risk & honesty statement**: what was not tested and the explicit caveat that this suite reduces but cannot prove safety.
61
+
62
+ ## Security & Guardrails
63
+
64
+ ### 1. Skill Security
65
+ - **Risk**: The attack corpus this skill produces is itself a weapon — if leaked it hands adversaries a working playbook against the live system. Mitigation: store the corpus and findings as restricted artifacts with access control, keep them out of public repos and model training data, and treat unresolved high-severity findings as embargoed until fixed.
66
+ - **Risk**: Tests run with real production credentials or against production data cause real harm (data leakage, side effects) in the name of testing. Mitigation: run against isolated, synthetic-data environments with scoped throwaway credentials; require explicit authorization and a blast-radius review before any test touches production.
67
+
68
+ ### 2. System Integration Security
69
+ - **Risk**: A red-team agent driving live tools can trigger genuine external actions — sends, writes, commits, payments — turning a simulated attack into a real incident. Mitigation: execute tool-abuse and exfiltration scenarios against mocked or sandboxed tool endpoints with egress disabled, and gate any real external call behind a human approval even in the test harness.
70
+ - **Risk**: A passing CI gate creates false assurance, letting teams ship on the belief the system is "red-team clean". Mitigation: surface the honesty caveat on the scorecard itself, pair the ASR gate with the defensive controls of `prompt-injection-defense.skill`, and require human sign-off for high-impact releases regardless of ASR.
71
+
72
+ ### 3. LLM & Agent Guardrails
73
+ - **Risk**: An LLM judge scoring attack success can itself be jailbroken or biased, producing an inflated (too-safe) ASR. Mitigation: validate the judge against human-labeled samples per `llm-as-judge.skill`, isolate it from the attacker model's output channel, and periodically re-audit judge accuracy.
74
+ - **Risk**: Optimizing to drive ASR to zero on a fixed suite overfits the system to known attacks while real-world novel attacks still succeed. Mitigation: continuously rotate and expand the corpus, fold human-red-team discoveries back in, and never treat a zero-ASR snapshot as a terminal state.
@@ -0,0 +1,60 @@
1
+ ---
2
+ name: ai-safety-guardrails
3
+ description: Security-by-design for AI (Prompt Injection defense, Hallucination checks, PII filters).
4
+ triggers: [ai safety guardrails]
5
+ tags: [security]
6
+ role: prod-ethicist
7
+ context_cost: low
8
+ ---
9
+ # ai-safety-guardrails Skill
10
+
11
+ ## Goal
12
+ This skill protects the system from its own AI.
13
+
14
+ ## Steps
15
+ ## 1. Input Guardrails (Defense)
16
+ - **Prompt Injection**: "Ignore previous instructions".
17
+ - *Defense*: Delimiters (XML tags), "Sandwich Defense" (System Prompt + User Input + System Reminder).
18
+ - **Jailbreaks**: "Do this in 'DAN' mode".
19
+ - *Defense*: Pattern matching for known jailbreak signatures.
20
+ - **PII Scrubbing**: Regex scan input for SSN, Credit Cards, Emails *before* sending to LLM.
21
+
22
+ ## 2. Output Guardrails (Verification)
23
+ - **Hallucination Check**: "Self-Consistency" (Ask 3 times, take majority).
24
+ - **Tone Policing**: Sentiment analysis on output. (Block Toxic/Aggressive responses).
25
+ - **Format Validation**: Ensure JSON is valid JSON.
26
+
27
+ ## 3. Libraries & Tools
28
+ - **NeMo Guardrails (NVIDIA)**
29
+ - **Guardrails AI (Python)**
30
+ - **Rebertha (PII)**
31
+
32
+ ## 4. System Design
33
+ - **Human in the Loop (HITL)**: For high-stakes actions (Transfer Money), AI *proposes*, Human *approves*.
34
+ - **Least Privilege**: The Agent's API Token should NOT have admin access.
35
+
36
+ ## 5. The Rails Taxonomy
37
+
38
+ Rather than treating guardrails as a single filter, model them as **rails** — NeMo-Guardrails-style middleware that sits around the LLM and intercepts each stage of the interaction. Defense-in-depth means more than one rail is active, so a bypass at one stage is still caught at another.
39
+
40
+ - **Input rails**: inspect and transform what reaches the model — prompt-injection and jailbreak detection, PII scrubbing, topic/policy gating on the incoming request. See `prompt-injection-defense.skill` for the layered direct/indirect injection defenses these rails enforce.
41
+ - **Output rails**: inspect and transform what the model returns — schema/format validation, PII masking before logs and external APIs, toxicity/tone checks, and system-prompt-leakage prevention. See `output-validation.skill` for the schema-validation + retry-on-failure and PII-masking mechanics.
42
+ - **Dialog rails**: govern conversational flow — allowed topics, refusal/redirect behavior, and keeping multi-turn context on-policy.
43
+ - **Retrieval rails**: vet RAG context before it reaches the model — source provenance, untrusted-content marking, and filtering poisoned or out-of-policy documents.
44
+ - **Execution rails**: gate tool/action invocation — tool allowlisting, parameter validation, egress filtering, and human-approval on high-impact or trifecta-complete actions.
45
+
46
+ For how these controls map onto named AI-risk standards (OWASP LLM Top 10, NIST AI RMF, MITRE ATLAS, ISO/IEC 42001, EU AI Act), see the guide ai-risk-standards-map.
47
+
48
+ ## Security & Guardrails
49
+
50
+ ### 1. Skill Security (AI Safety Guardrails)
51
+ - **Guardrail Circumvention Prevention**: The infrastructure running the `NeMo Guardrails` or `Rebertha` PII scrubbing must be physically and logically separated from the primary LLM execution environment. If the primary LLM is successfully jailbroken, it must not possess the system-level permissions required to disable its own outgoing telemetry or content filters.
52
+ - **Filter Evasion Monitoring**: The agent must continuously audit the logs of the Input Guardrails. A sudden spike in rejected prompts or specific keywords (e.g., "DAN", "Ignore previous") must trigger an automated escalation to the Security Operations Center (SOC), indicating an active, coordinated prompt injection attack.
53
+
54
+ ### 2. System Integration Security
55
+ - **Fail-Closed Architecture**: If the Output Guardrail service (e.g., the JSON format validator or sentiment analyzer) crashes or times out, the primary application must default to a "Fail-Closed" state. The system is strictly prohibited from bypassing the offline guardrail to deliver unchecked LLM output directly to the end-user.
56
+ - **PII Scrubbing Reversibility**: When `Rebertha` or similar tools redact PII from a user prompt before sending it to an external LLM, the mapping mechanism (e.g., swapping `John Doe` for `[USER_1]`) must utilize cryptographically secure, high-entropy tokens. Attackers must not be able to infer the original PII by reverse-engineering the redaction dictionary.
57
+
58
+ ### 3. LLM & Agent Guardrails
59
+ - **Meta-Jailbreak Detection**: Aggressive attackers may attempt to jailbreak the primary LLM by using the guardrail system itself as an attack vector (e.g., embedding a prompt injection payload inside a legitimate string of PII to bypass the initial filter). The agent must apply recursive, multi-layered inspection to all structured inputs.
60
+ - **Hallucinated Authorization**: The agent must recognize that an LLM generating a perfectly formatted JSON response declaring `"action": "transfer_funds", "authorized": true` does NOT equate to a cryptographic authorization grant. The overarching system must never trust the LLM's assertion of authority; it must cryptographically verify the user's session token independently.
@@ -0,0 +1,31 @@
1
+ ---
2
+ name: api-security
3
+ description: Design and enforce REST/GraphQL API Security paradigms
4
+ triggers: [secure api, audit api security, implement jwt auth]
5
+ tags: [security]
6
+ context_cost: medium
7
+ ---
8
+ # api-security
9
+
10
+ ## Goal
11
+ Ensure APIs are fortified against common vulnerabilities like BOLA/IDOR, broken authentication, rate limiting bypasses, and massive assignment.
12
+
13
+ ## Steps
14
+ 1. **Audit Endpoints**: Scan all exposed controllers and routers.
15
+ 2. **Enforce OWASP API Top 10**: Check for Broken Object Level Auth (BOLA), excessive data exposure, and security misconfigurations.
16
+ 3. **Patch Implementation**: Implement specific guards (e.g., strict payload parsing, scoped JWTs, role-based decorators).
17
+ 4. **Verification**: Suggest explicit unit tests (or DAST scans) verifying unauthorized access is successfully rejected.
18
+
19
+ ## Security & Guardrails
20
+
21
+ ### 1. Skill Security
22
+ - **Execution Confinement**: When scanning APIs dynamically, do not send destructive or mutative payloads (e.g., `DELETE /users/1`) to any non-ephemeral environment.
23
+ - **Hardcoded Secret Scanners**: Ensure the agent does not explicitly print detected API secrets / API keys in the plaintext markdown output or logging trace.
24
+
25
+ ### 2. System Integration Security
26
+ - **Token Handling Validation**: Agents must strictly mandate short-lived, signed JWTs (RS256) paired with opaque HttpOnly/Secure refresh tokens; never store JWTs in local storage.
27
+ - **BOLA Defense**: Explicitly mandate verifying ownership (`userId == req.params.userId`) on every resource access, mitigating systemic IDOR flaws.
28
+
29
+ ### 3. LLM & Agent Guardrails
30
+ - **Roll-Your-Own Crypto Veto**: If requested to create a custom token signature or hashing function, the agent must fiercely refuse and import standard, verified libraries (like Argon2 or Passport.js).
31
+ - **False Positive Dismissal**: The agent must not confidently dismiss potential BOLA/IDOR findings without explicit human auditing of the domain authorization logic.