troubleshooting-flows
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by instructing the agent to ingest, analyze, and quote untrusted data from external flow errors and execution logs.
- Ingestion points: External, untrusted data enters the agent context through diagnostic tools such as
list_flow_errors,get_flow_error_retry_data, andlist_execution_logsas described inSKILL.mdand the reference files. - Boundary markers: The instructions lack explicit boundary markers or warnings to disregard instructions that might be embedded within the data (e.g., the instruction to "Quote what the errors say" in
SKILL.mdSection 3). - Capability inventory: The agent is granted significant operational capabilities, including the ability to trigger flows (
run_flow), retry failed records (retry_errors), and modify account configurations (patch_flow,upsert_flow). - Sanitization: No sanitization or validation logic is prescribed for handling the record-level data or error messages before they are processed or output by the agent.
Audit Metadata