mitm-find-pii
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes 'log.txt', which contains raw network traffic data including API responses. This data originates from external servers and is untrusted. An attacker could craft a response containing hidden instructions or adversarial payloads designed to manipulate the AI agent's analysis or behavior. \n
- Ingestion points: The agent is directed to analyze the mitmproxy dump file 'log.txt' located in the current directory. \n
- Boundary markers: The skill lacks explicit delimiters or instructions to treat the log content as untrusted data, increasing the risk that the LLM might follow instructions embedded within the traffic. \n
- Capability inventory: The skill involves reading local files. While no network or write capabilities are explicitly defined in the SKILL.md, the agent's general environment may provide access to other tools. \n
- Sanitization: No input validation or sanitization is applied to the log file content before processing. While the skill requests a 'Redacted example' for output, this measure does not prevent injection attacks targeting the input phase.
Audit Metadata