analyzing-user-feedback
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes workflows that involve ingesting untrusted external data into an AI agent's context for synthesis, which creates a surface for indirect prompt injection attacks.
- Ingestion points: The documentation instructs the agent to process data from Discord transcripts, app store reviews, customer support tickets, and Reddit thread exports (
SKILL.md,references/artifacts.md,references/guest-insights.md). - Boundary markers: There are no instructions or recommendations for using delimiters or boundary markers to distinguish untrusted user content from agent instructions.
- Capability inventory: The skill identifies capabilities for semantic clustering, sentiment analysis, and automated action based on feedback, such as creating Zendesk tickets via an NPS triage agent (
references/artifacts.md). - Sanitization: The instructions provide no guidance on sanitizing, filtering, or validating external inputs before they are processed by the LLM.
- [NO_CODE]: The skill is composed entirely of markdown documentation and reference materials; it does not contain any executable scripts, binaries, or configuration files that could be directly exploited.
Audit Metadata