icp-evidence-analysis
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources which could contain malicious instructions.
- Ingestion points: The skill accepts company URLs,
source_manifest.json, and public web pages (product, pricing, filings) as inputs for analysis (SKILL.md). - Boundary markers: There are no explicit instructions to wrap or delimit external content with markers to prevent the agent from mistaking data for instructions.
- Capability inventory: The skill utilizes web-fetching tools and local file read/write operations for
icp_hypotheses.json. - Sanitization: The skill mandates strict source attribution and fact-checking, requiring assertions to be classified as
company_asserted,buyer_behavior, orindependent, which helps mitigate the impact of malicious data but does not remove the ingestion surface.
Audit Metadata