list-enrichment
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from multiple external backends (Signal Engine, HubSpot, Gong, Clay, ZoomInfo) and writes it into local JSON files within the campaign workspace. This enriched data is intended for consumption by subsequent skills like
email-prompt-buildingandemail-generation. If external sources return malicious payloads (e.g., hidden instructions in news signals or research results), they could influence downstream agent behavior. - Ingestion points: External data from enrichment backends and local account files in
campaigns/{slug}/accounts/. - Boundary markers: No specific delimiters or "ignore instructions" warnings are defined for the enrichment data fields.
- Capability inventory: File system write access to
campaigns/, network connectivity to third-party APIs, and the ability to spawn subprocesses. - Sanitization: The skill does not describe any validation or escaping of the data retrieved from external backends before storage.
- [COMMAND_EXECUTION]: The enrichment workflow includes a step to invoke an "Account Researcher" as a "Claude Code agent subprocess" for complex research tasks. This involves executing secondary agent logic to handle specific data points that cannot be resolved by standard API lookups.
Audit Metadata