ai-wrapper-product
Warn
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: MEDIUMPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- PROMPT_INJECTION (MEDIUM): The skill demonstrates code patterns vulnerable to Indirect Prompt Injection where untrusted data is interpolated into prompt strings without sufficient isolation.
- Ingestion Points:
userInputingenerateContentand thepromptTemplates.emailWriterproperties inSKILL.md. - Boundary Markers: Absent. The templates do not use delimiters or structural markers to separate instructions from user-controlled content.
- Capability Inventory: The provided logic interacts with external LLM providers (
anthropic.messages.create), creating a path for exfiltration or behavior manipulation if the AI obeys instructions embedded in user input. - Sanitization: Only length validation is present; no escaping of control characters or content filtering is implemented.
- EXTERNAL_DOWNLOADS (LOW): The skill references the
@anthropic-ai/sdkpackage. - Trusted Source: The package originates from a trusted organization (Anthropic). Per [TRUST-SCOPE-RULE], this finding is downgraded to LOW.
Audit Metadata