The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements strong defensive measures against indirect prompt injection. It explicitly warns that event properties (like channel_source or email_subject) are captured via public tokens and could be forged by external actors to steer the agent's behavior.
[COMMAND_EXECUTION]: The skill uses the execute-sql tool to perform read-only analytics on the events table. The SQL queries are structured to filter specifically for the $conversation_* event family using startsWith, which prevents accidental access to unrelated telemetry data.
[DATA_EXFILTRATION]: Findings are shared internally through the emit_report tool. The skill instructions specifically prohibit re-surfacing individual ticket content, focusing instead on aggregate operational metrics, which minimizes the risk of sensitive data exposure.
[SAFE]: The skill's behavior is consistent with its stated purpose as a PostHog product monitoring tool. All external resource references (event schemas and tools) align with the vendor's own infrastructure.

signals-scout-conversations