The Agent Skills Directory

[SAFE]: No security vulnerabilities were detected. The skill uses standard PostHog monitoring tools and processes telemetry data according to its stated purpose of identifying AI performance regressions.- [INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it analyzes external telemetry data (LLM traces and evaluations). However, the skill's capabilities are limited to generating observability reports and maintaining internal state via a scoped scratchpad, and it includes explicit instructions to localize and sample data rather than blindly processing raw content. This risk is inherent to the scout's primary monitoring function and is managed by its restricted environment.
Ingestion points: Telemetry and trace data is ingested via execute-sql, query-llm-trace, and query-llm-traces-list (SKILL.md).
Boundary markers: Absent, but the skill follows a specific 'trend -> spike -> localize -> sample' discipline to filter noise.
Capability inventory: emit_report, edit_report, and scout-scratchpad-remember (SKILL.md).
Sanitization: Not explicitly described for telemetry data ingestion.

signals-scout-ai-observability