hunt-html-injection
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill includes active payload examples such as
<img src=x onerror=alert(91234)>and<svg onload=alert(91234)>. These patterns are designed to demonstrate script execution and bypass content filters, which matches behavior-altering instruction patterns. - [INDIRECT_PROMPT_INJECTION]: The skill describes an attack surface where untrusted data is reflected into an HTML context, potentially leading to the execution of embedded instructions.
- Ingestion points: Input fields like search boxes, profile fields, and contact forms (SKILL.md).
- Boundary markers: Features a mandatory confirmation gate requiring the agent to ask for explicit user permission before performing any active probing.
- Capability inventory: Includes local file reading via shell commands like
cat scope.txt(SKILL.md). - Sanitization: No automatic sanitization of generated payloads is provided; the skill relies on manual human verification and authorization.
Audit Metadata