meta-ads
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSDYNAMIC_CONTEXT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted campaign data from JSON configuration files, specifically natural language strings for ad copy and headlines. This creates a surface for indirect instructions that could influence agent behavior, although the impact is restricted to the Meta Ads platform domain. \n
- Ingestion points: The
scripts/campaign.mjsscript reads configuration data from user-specified file paths via the--configflag. \n - Boundary markers: While the skill uses structured JSON, it lacks explicit delimiters or instructions to the agent to ignore potentially malicious instructions within the interpolated ad copy fields. \n
- Capability inventory: The skill has the capability to perform network requests to Meta's Marketing API and read local files for creative assets. \n
- Sanitization: Data is handled using standard JSON serialization and URL encoding techniques.\n- [EXTERNAL_DOWNLOADS]: The skill performs network operations targeting
graph.facebook.comto interact with Meta's official Marketing API. As this targets a well-known technology service for its primary operational purpose, it is considered safe.\n- [DYNAMIC_CONTEXT_INJECTION]: The skill utilizes a dynamic context injection pattern (!command) inSKILL.mdto execute a local script for loading project-specific metadata. This is a benign use of the platform's initialization capabilities.
Audit Metadata