The Agent Skills Directory

[SAFE]: A comprehensive analysis of the skill instructions revealed no evidence of malicious behavior, prompt injection, or obfuscation. The skill's functionality is consistent with its declared purpose of managing user interview topics.\n- [COMMAND_EXECUTION]: The skill utilizes HogQL (PostHog's SQL variant) to query person and event data for audience selection. It incorporates safety practices by instructing the agent to validate event names using the read-data-schema tool before constructing queries, and it enforces a 500-member cap on cohort resolution to maintain manageable execution limits.\n- [DATA_EXFILTRATION]: The skill identifies users by their email addresses or distinct IDs. This sensitive data is retrieved from the project's internal database and provided directly to PostHog's interview creation API. No external network requests or unauthorized data transfers were detected.\n- [PROMPT_INJECTION]: The skill instructions provide clear boundaries for the agent's role and focus on structured data collection. The use of user-provided 'agent context' and 'questions' represents a standard configuration surface for the research agent and does not include instructions designed to bypass or override the agent's core safety protocols.

planning-user-interviews