seo-quote
Warn
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill instructions explicitly require the agent to generate a Python script at runtime and execute it to produce the final .docx report. This involves the use of the
python-docxlibrary and standard libraries likeosanddatetimeto build a functional script based on a provided template. - [COMMAND_EXECUTION]: The agent is directed to run the generated Python code, which performs file system operations such as directory creation (
os.makedirs) and file writing (doc.save). - [DATA_EXFILTRATION]: The skill uses a hardcoded local file path
C:/Users/cmano/claude-seo/for saving documents. This exposes a specific user profile path ('cmano') and targets a specific directory structure on the host machine for output. - [INDIRECT_PROMPT_INJECTION]: The skill fetches content from external, user-provided URLs (sitemaps, homepage HTML, robots.txt) and parses them to extract metadata for the report.
- Ingestion points: Fetches data from
<url>/sitemap.xml,<url>/robots.txt, and the homepage HTML. - Boundary markers: The instructions lack explicit delimiters or warnings to the agent to ignore potentially malicious instructions embedded in the external content (e.g., within HTML meta tags, sitemap entries, or code comments).
- Capability inventory: Network GET requests to arbitrary URLs, file system write access, and dynamic Python code execution.
- Sanitization: No sanitization, validation, or escaping of the content retrieved from external URLs is specified before the agent processes it into the structured report.
Audit Metadata