ads-agent

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [Command Execution] (LOW): The skill leverages an external MCP server (meta-ads) with 46 specialized tools for Facebook API interaction. While powerful, these are used within a defined framework for ad management.
  • [Prompt Injection] (LOW): Risk of Indirect Prompt Injection (Category 8):
  • Ingestion points: Reads sensitive configuration and historical data from .claude/ads-agent/config/ and .claude/ads-agent/history/.
  • Boundary markers: Absent; the agent does not appear to use delimiters to distinguish between system instructions and data read from these files.
  • Capability inventory: Extensive capabilities including update_adset (budget changes), create_campaign, and pause_ad via the meta-ads server.
  • Sanitization: Absent; no explicit validation of the content within the briefs or history files is mentioned.
  • [Data Exposure] (LOW): The agent accesses local files containing account identifiers and marketing strategies (briefs). While no credentials are hardcoded, these files contain proprietary business data.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:18 PM