h3-prompt-writing
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user requests, creating a potential surface for indirect prompt injection where malicious instructions could be embedded in the input data.\n
- Ingestion points: The skill transforms user-supplied "multimodal requests" into H3 prompt structures as outlined in
SKILL.md.\n - Boundary markers: It uses specific tags such as
<d>,<Subject N>, and[Shot N]which help distinguish user data from the agent's instructions.\n - Capability inventory: The skill reads its own reference files but does not exhibit high-privilege capabilities such as network access, arbitrary file system writes, or subprocess execution.\n
- Sanitization: The instructions mandate that user dialogue and scene text be preserved verbatim, which effectively bypasses sanitization for those specific input fields.
Audit Metadata