The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and analyze untrusted data from external sources, including user reviews, support tickets, and open-ended survey responses (verbatims). While the skill provides comprehensive guidance on sentiment analysis and thematic coding, it lacks specific instructions for the agent to treat these external inputs as untrusted or to use delimiters to prevent instructions embedded within user feedback from influencing agent behavior.
Ingestion points: Processes data from NPS verbatims, App Store reviews, support tickets (Zendesk/Intercom), and social media monitoring as described in reference/multi-channel-synthesis.md.
Boundary markers: The skill mentions using structured output schemas for LLM classification but does not provide explicit boundary markers or directives to ignore instructions within the processed text.
Capability inventory: The skill is designed to output reports and classification labels, potentially influencing downstream agents like Spark or Growth as outlined in the collaboration patterns in SKILL.md.
Sanitization: Includes a 'contamination gate' in SKILL.md to detect synthetic/AI-generated feedback, but this does not specifically address adversarial prompt injection attempts within legitimate-looking user feedback.

voice