The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied context and external research data, creating a potential surface for indirect prompt injection. However, this is mitigated by strict internal policies. * Ingestion points: User objectives, constraints, and external research data (SKILL.md, references/skill-spec.json). * Boundary markers: The skill requires classifying claims as fact, inference, or assumption, and mandates separating evidence from inference (SKILL.md, references/skill-spec.json). * Capability inventory: The skill utilizes local routing scripts and allows for optional web research (SKILL.md, references/skill-spec.json). * Sanitization: The evidence policy requires verification of all material claims regarding platforms, pricing, and products (references/skill-spec.json).
[COMMAND_EXECUTION]: The skill references local scripts for dynamic routing and specialized agent selection. * Evidence: The skill calls ../../scripts/dynamic_router.py and ../../scripts/neural_router.py to handle cross-functional dependencies and agent selection (SKILL.md, references/skill-spec.json). * Context: These scripts appear to be part of the skill's standard execution framework and are invoked through relative paths within the project structure.
[SAFE]: No evidence of data exfiltration, credential theft, persistence mechanisms, or obfuscated content was found across the analyzed files. The instructions prioritize user-provided primary evidence and explicit uncertainty over fabricated proofs.

conversational-advertising