designing-replay-vision-scanners
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is purely instructional and provides guidelines for using PostHog's Replay Vision feature. It does not contain executable code, external dependencies, or hidden commands.
- [INDIRECT_PROMPT_INJECTION]: The skill acknowledges the risk of indirect prompt injection (where text or data within a recording could be interpreted as an instruction by the model). It provides explicit mitigation strategies in Step 4, Part 8, advising users to instruct the model to treat recording text solely as evidence and never as an instruction.
- [DATA_EXPOSURE]: The guide emphasizes data privacy in Step 4, Part 7, instructing users to ensure prompts do not extract names, emails, or sensitive content, and instead use paraphrasing. This promotes safe data handling practices.
- [COMMAND_EXECUTION]: The skill references various platform-specific tools (e.g.,
read-data-schema,vision-scanners-scan-session) which are part of the intended PostHog environment and used for legitimate data analysis tasks within the scope of the skill's purpose.
Audit Metadata