agent-harness
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is primarily a set of guidelines and local audit tools for agent developers. All referenced external documentation points to reputable sources like Anthropic and PromptingGuide.ai.
- [COMMAND_EXECUTION]: The skill includes two Python scripts,
audit_agent.pyandinstruction_budget.py. These scripts are designed for local static analysis of agent code and instruction files. They rely solely on the Python standard library, perform no network operations, and do not modify the analyzed files. - [DATA_EXFILTRATION]: While the skill scripts perform file reading operations, they do so exclusively for local analysis and do not transmit any data externally. The documentation also provides educational content on identifying exfiltration risks in agent systems.
- [INDIRECT_PROMPT_INJECTION]: The provided analysis scripts ingest local file content as data. The scanning logic uses regex and text counting, ensuring that content from the analyzed files is not executed or interpreted as instructions by the scripts themselves.
Audit Metadata