hunt-agent-native-memory
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides a Python-based 'Runnable Probe' script intended for security research. This script programmatically interacts with external endpoints to determine buffer overflow crash boundaries, which is a functional component of the vulnerability testing methodology described.\n- [DATA_EXFILTRATION]: The included probe script is designed to send payloads to user-configured target URLs (e.g.,
TARGET_UPLOAD). While this constitutes network activity, it is a localized tool for the researcher and does not exfiltrate data to a hardcoded malicious third party.\n- [INDIRECT_PROMPT_INJECTION]: The skill analyzes the vulnerability surface where agents pass unvalidated bytes to native binaries. It provides a structured approach for researchers to test these boundaries; although it discusses the ingestion of untrusted data, the provided tools themselves do not facilitate prompt-based exploitation of the agent itself.
Audit Metadata