harness-engineering
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill provides educational content and diagnostic frameworks for managing AI agent reliability. It contains no executable code, remote script downloads, or network exfiltration patterns.
- [DATA_EXFILTRATION]: No sensitive file paths, credentials, or unauthorized network operations were identified. The references to external sources (e.g., Leanpub) are standard citations for the adapted material.
- [INDIRECT_PROMPT_INJECTION]: The skill actively describes security best practices to mitigate injection risks, such as treating model output as untrusted at boundaries and using structured output validation. It does not perform data ingestion itself.
- [OBFUSCATION]: The content is written in clear, standard Markdown with no hidden characters, encoded strings, or homoglyph substitutions.
- [COMMAND_EXECUTION]: The skill provides guidance on using tools and commands but does not contain any shell commands or dynamic context injection patterns (such as backtick-enclosed commands).
Audit Metadata