The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it is designed to visit and analyze arbitrary external URLs provided by the user. Malicious instructions embedded in the analyzed page's HTML or metadata could attempt to influence the agent's behavior.
Ingestion points: SKILL.md (Step 1: Visual inspection of supplied Lexsis or custom-domain URLs).
Boundary markers: Absent in the high-level instructions, though the agent is directed to rely on verified evidence ledgers.
Capability inventory: Significant capabilities including lexsis_drafts.page_replace, lexsis_drafts.experiment_create, and lexsis_live_ops.scale_winner (SKILL.md).
Sanitization: JavaScript generation is sanitized by a platform-level compiler (references/animation-system.md).
[DYNAMIC_EXECUTION]: The skill is designed to author and submit custom JavaScript modules for storefront animations (managed motion).
Evidence: references/animation-system.md defines a contract for authoring application/lexsis-motion scripts.
Context: This is a core feature of the skill's purpose (UI design). The risk is mitigated by a mandatory server-side compilation step (lexsis_pages.compile) that parses the AST and rejects unsafe globals, raw loops, and arbitrary networking.

ab-test