account-tiering-workflow
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, including company profiles and job listings, which is then processed by an LLM node to generate scoring rationales. This creates a surface for indirect prompt injection.\n
- Ingestion points: External data from Tech Stack, Hiring, and Company Profile enrichments, as well as account record fields mapped by the user.\n
- Boundary markers: The prompt instructions direct the LLM to use only provided evidence and report unobserved data points.\n
- Capability inventory: The skill includes the ability to write results back to the CRM using the
upsert-audiences-recordtool node.\n - Sanitization: The instructions do not specify any sanitization or filtering of external data before interpolation into the agent prompt.\n- [DYNAMIC_EXECUTION]: The skill provides Python code snippets for implementing specific logic in the platform's 'code nodes'.\n
- Evidence: Python
handlercode blocks provided in Step 3 for track resolution and tiering logic.\n- [COMMAND_EXECUTION]: The skill instructs the agent to run platform-specific CLI commands for environment verification and resource discovery.\n - Evidence: Commands such as
clay whoamiandclay workflows nodes --helplisted in the setup instructions.
Audit Metadata