content
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes content from Reddit, Bluesky, and WordPress. This creates a surface for indirect prompt injection where malicious content from these platforms could attempt to influence the agent's behavior. The skill implements mitigations, including explicit boundary markers (
<untrusted-content>) and sanitization logic that strips these tags from input data to prevent boundary escaping in moderation prompts. - [PROMPT_INJECTION]: Instructions containing strings like "ignore previous instructions" were detected in
references/classification-prompt.md. These are defensive instructions designed to guide the AI model on how to securely handle untrusted content during moderation, rather than an attempt to bypass security filters. - [COMMAND_EXECUTION]: The skill uses Python scripts to interact with APIs (Reddit, WordPress). These scripts are invoked via the
Bashtool and perform network operations and local file modifications (e.g., updating a content calendar or saving scraped articles). - [EXTERNAL_DOWNLOADS]: The WordPress scraper script (
scripts/publish/wordpress-scraper.py) downloads images from specified URLs to the local filesystem. The script derives filenames from the URL path using logic that prevents path traversal vulnerabilities.
Audit Metadata