linking-errors-to-traces
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill uses user-supplied inputs, such as session IDs and trace IDs, to dynamically construct SQL queries for the
posthog:execute-sqltool, creating a vulnerability surface for injection attacks. - Ingestion points: User-provided occurrence UUIDs, session IDs, and trace IDs are identified as primary inputs in
SKILL.mdand processed via templates in the reference files. - Boundary markers: While
references/finding-exceptions.mdandreferences/finding-trace-logs.mdinclude natural language instructions for the agent to "validate" IDs or ensure they are hex/UUID-shaped, the SQL templates themselves use simple string interpolation (e.g.,WHERE $session_id = '<session_uuid>') without structural delimiters or enforcement. - Capability inventory: The skill possesses significant capabilities through the
posthog:execute-sqltool (used inreferences/finding-exceptions.mdandreferences/finding-trace-logs.md) and theposthog:query-logstool, which can access and aggregate sensitive platform data. - Sanitization: The instructions lack evidence of parameterized queries or automated sanitization, relying entirely on the agent's ability to follow natural language validation guidelines before inserting data into the SQL strings.
Audit Metadata