The Agent Skills Directory

[SAFE]: The skill is well-documented and adheres strictly to its stated purpose of monitoring session recording health and identifying user friction patterns within the PostHog platform.
[INDIRECT_PROMPT_INJECTION]: The skill includes an explicit 'Untrusted data' section that correctly identifies session content (URLs, element text, console logs) as attacker-controllable input. It mandates sanitization, aggressive truncation, and treats these inputs strictly as data to be reported, never as instructions to be followed.
Ingestion points: Reads from raw_session_replay_events, events, and posthog.session_replay_features tables which contain end-user browser data.
Boundary markers: Uses explicit instructions to isolate session-derived text from the agent's reasoning process.
Capability inventory: Uses execute-sql for data retrieval and scout-emit-report for output.
Sanitization: Instructs the agent to truncate snippets and use sanitized, slugified identifiers for memory storage.
[COMMAND_EXECUTION]: The skill utilizes SQL queries for analysis. While powerful, these are limited to the specific analytics tables provided by the PostHog environment and follow standard data analysis patterns.
[DATA_EXFILTRATION]: There is no evidence of sensitive data being sent to unauthorized external domains. All reported data is handled through official reporting channels (emit_report).

signals-scout-session-replay