meta-pixel

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill identifies event payloads and pasted exports as untrusted data surfaces and implements comprehensive defensive measures. It explicitly instructs the agent to treat all external content as data rather than instructions, requiring that any commands found within processed data be reported as findings rather than executed. It also forbids following URLs extracted from fetched content.
  • [DATA_EXFILTRATION]: Robust protections are in place to prevent the leakage of sensitive information. The skill mandates that personal data (PII) found in tracking events must never be reproduced in output or stored in state files. Furthermore, it references a global security policy that prohibits the storage or echoing of API keys, tokens, or other credentials found in data exports.
  • [COMMAND_EXECUTION]: The skill leverages the Model Context Protocol (MCP) to interact with Meta's ad account and Events Manager. This integration is scoped to read-only data retrieval for auditing purposes and does not provide a vector for arbitrary shell command execution.
  • [PROMPT_INJECTION]: There are no signs of malicious prompt injection. The skill's instructions focus on maintaining high-quality, research-backed auditing while strictly adhering to safety and privacy constraints. The internal instructions to minimize user questions by researching publicly available company information are transparently reported to the user as inferences.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 02:23 PM