signals-scout-feature-flags
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill acts as a focused monitor for feature flag health and traffic contradictions, using platform-integrated tools for analytics and reporting.
- [INDIRECT_PROMPT_INJECTION]: The skill identifies a clear attack surface where untrusted data enters the agent context via HogQL queries on the events table in SKILL.md. To mitigate this, the instructions mandate that the agent treat event-supplied keys and responses as data only, never as instructions. It specifically requires quoting these values as untrusted snippets and corroborating findings against trusted roster data from system.feature_flags and REST tools. The skill possesses capabilities for file-like scratchpad writing and network reporting but bounds these with instructions for truncation and sanitization of external strings.
- [COMMAND_EXECUTION]: The use of execute-sql is confined to HogQL for analytical purposes and does not provide a path for arbitrary system command execution.
Audit Metadata