py-ai
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill acts as a security-focused guide for AI development. It systematically addresses common attack vectors like prompt injection and data exfiltration by recommending architectural boundaries and validation layers.
- [INDIRECT_PROMPT_INJECTION]: The documentation explicitly identifies the risks associated with processing untrusted external data (RAG) and tool outputs. It provides remediation guidance, including the use of clear boundary markers, system policies that cannot be overridden by context, and adversarial testing.
- [COMMAND_EXECUTION]: Example 2 provides a secure implementation for a file tool executor. It uses
pathlib.Path.resolve(strict=True)andrelative_to()checks to effectively prevent path traversal attacks, ensuring the agent cannot access files outside the intended root directory. - [CREDENTIALS_UNSAFE]: The skill enforces best practices for secret management, warning against hardcoded keys and instructing developers to use environment variables or secret managers. It also emphasizes the importance of redacting sensitive data (tokens, PII, auth headers) from logs and traces.
- [EXTERNAL_DOWNLOADS]: The skill utilizes well-known, industry-standard Python libraries for AI development. It provides instructions for downloading models from Hugging Face, a well-known service, and explicitly recommends setting
trust_remote_code=Falseto prevent the execution of arbitrary code from model repositories.
Audit Metadata