replay-ux-audit
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user session interaction timelines and customer feedback. While the risk is limited as the skill does not possess high-privilege capabilities like arbitrary code execution, an attacker could attempt to bias the UX audit results by crafting specific event patterns or feedback text.
- Ingestion points: Data enters the context via
Amplitude:get_amp_session_replay_info(interaction timelines) andAmplitude:use_amplitude_ai_feedback(customer feedback insights). - Boundary markers: Not present; there are no instructions to the agent to distinguish between data and potential instructions embedded within event labels or feedback content.
- Capability inventory: The skill's tools are restricted to the Amplitude ecosystem for data retrieval and analysis; no file-system writes, shell execution, or non-vendor network operations are permitted.
- Sanitization: No explicit sanitization or validation is applied to the retrieved interaction data before it is synthesized into the friction map.
Audit Metadata