blog-cannibalization

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection vulnerability surface due to the ingestion of untrusted content.
  • Ingestion points: The skill scans and reads external/local blog content files with extensions **/*.md, **/*.mdx, and **/*.html during the local mode workflow within SKILL.md.
  • Boundary markers: There are no explicit boundary markers or delimiters defined to encapsulate the ingested file contents, nor instructions telling the model to ignore any embedded directives.
  • Capability inventory: The capabilities are limited to reading local files and invoking an external CLI wrapper for DataForSEO API calls; there are no file-writing or arbitrary shell command execution capabilities.
  • Sanitization: The text is tokenized, lowercased, and filtered for stop words during keyword extraction, but portions of the text (titles and headings) are directly interpolated into the final output report without sanitization against malicious prompts.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 04:57 AM
Security Audit — agent-trust-hub — blog-cannibalization