workflows
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides mechanisms for ingesting untrusted external data which is then processed by AI agents and tool nodes. While this introduces a potential attack surface, the skill implements structural boundaries to mitigate risk.
- Ingestion points: Webhook payloads (SKILL.md) and CSV file uploads (csv-triggers.md, SKILL.md).
- Boundary markers: Present. The system uses strict JSON schemas (inputSchema/outputSchema) and JSONPath selectors (sourcePath) to isolate specific data fields, preventing unstructured raw data from being interpreted directly as instructions.
- Capability inventory: Supports LLM-based reasoning (agent nodes), external API calls (tool nodes), and deterministic logic (code nodes), as described in SKILL.md and account-agents.md.
- Sanitization: Relies on platform-level schema validation and sandboxed code execution environments.
- [DYNAMIC_EXECUTION]: The skill supports 'Code nodes' which execute Python scripts at runtime to perform data transformations.
- Evidence: Documented in SKILL.md and testing.md.
- Safety Details: The execution environment is heavily restricted to the Python standard library, prohibits network and file system access, and enforces a 1-second execution timeout to prevent resource exhaustion or malicious behavior.
Audit Metadata