product-strategy-review
Pass
Audited by Gen Agent Trust Hub on Mar 3, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: No direct prompt injection or jailbreak attempts were found. The skill uses instructional flags like 'HOLD' and 'PROCEED' to manage internal workflow logic based on the presence of required inputs.- [DATA_EXFILTRATION]: The skill is designed to read sensitive internal context such as OKRs, vision documents, and team structures. This data is used solely for the local generation of the strategy review; no instructions for unauthorized external network transmission or hardcoded credentials were detected.- [EXTERNAL_DOWNLOADS]: The skill accepts external URLs via 'doc_link' and 'evidence_hub_link'. These are handled as data sources for analysis. There are no patterns suggesting the download and execution of remote scripts or binary assets.- [PROMPT_INJECTION]: Indirect Prompt Injection Surface: The skill ingests untrusted data from external strategy documents. While it lacks explicit sanitization or boundary markers (e.g., XML tags or delimiters) to isolate external content, this is an inherent property of the document-review use case and is managed by the highly structured output requirements of the PRISM framework. Evidence Chain: (1) Ingestion points: doc_link, evidence_hub_link; (2) Boundary markers: Absent; (3) Capability inventory: Generates Markdown and JSON text; (4) Sanitization: None specified.
Audit Metadata