The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's architecture includes an AI-interviewer mode and automated scoring that process untrusted text from survey respondents. * Ingestion points: Respondent input is ingested into the AI-interviewer logic in references/multimodal-ux.md and the scoring framework in references/scoring-framework.md. * Boundary markers: The suggested implementation uses tags and specific system instructions to delineate respondent input from the agent's internal reasoning and structured output. * Capability inventory: The skill itself does not perform sensitive network or file operations; dangerous actions are deferred to a user-controlled application stack (e.g., Supabase, Next.js). * Sanitization: The framework employs LLM-based logic to detect and flag violations and verify answer specificity, effectively filtering low-signal or non-compliant input.

rethink-surveys