newsletter-promo
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests external data from newsletter links or full text which is then interpolated into a prompt for building an HTML file. This represents a potential vector for indirect prompt injection if the newsletter content contains malicious instructions intended for the agent.
- Ingestion points: User-provided newsletter links or text in
Step 1. - Boundary markers: The prompt uses bracketed placeholders like
[HOOK]and[LABEL]but lacks explicit instructions to the agent to treat the newsletter content strictly as data. - Capability inventory: The agent is instructed to build a self-contained
index.htmlfile using Claude Code, which involves file writing and execution capabilities. - Sanitization: There is no mention of sanitizing or escaping the user-provided newsletter content before it is inserted into the HTML brief.
- [DYNAMIC_EXECUTION]: The skill facilitates the generation of executable HTML and JavaScript code at runtime. While the structure is based on a template, the content of the labels and the hook is derived from untrusted external sources, which are then used to build the final output file.
Audit Metadata