skills/mukul975/anthropic-cybersecurity-skills/hunting-for-data-exfiltration-indicators/Gen Agent Trust Hub
hunting-for-data-exfiltration-indicators
Pass
Audited by Gen Agent Trust Hub on Mar 15, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill provides defensive tools and documentation for identifying data exfiltration indicators. The included Python scripts perform local analysis of log files using standard statistical methods (e.g., Shannon entropy) and do not exhibit any malicious behavior.
- [REMOTE_CODE_EXECUTION]: No patterns for remote code execution, such as 'curl | bash' or unsafe use of 'eval()', were found in the provided scripts.
- [DATA_EXPOSURE_AND_EXFILTRATION]: The scripts only perform local file reads on user-provided log paths and do not access sensitive system files (e.g., SSH keys, AWS credentials) or make outbound network requests.
- [INDIRECT_PROMPT_INJECTION]: While the scripts ingest untrusted external data (network logs), they do not execute content from these logs or use them in a way that could influence agent behavior. The parsing logic is strictly limited to extracting specific fields for statistical analysis.
- [OBFUSCATION]: The code and documentation are clear and readable, with no evidence of Base64, hex-encoding, or hidden characters intended to mask malicious intent.
Audit Metadata