business-os
Pass
Audited by Gen Agent Trust Hub on Aug 30, 2026
Risk Level: SAFENO_CODE
Full Analysis
- No Executable Code: The skill package is composed exclusively of Markdown documentation and a JSON metadata file. It does not include any scripts (Python, JavaScript, Shell) or binaries, ensuring it cannot perform unauthorized operations on the host system.
- Governance and Human-in-the-loop Controls: The skill includes specific references (references/governance-evidence-qa.md) that instruct the agent to distinguish between thinking, drafting, and executing. It explicitly requires human confirmation for consequential actions like sending emails or committing financial resources.
- Factual Integrity and Transparency: The "Company DNA" framework (references/company-dna.md) enforces clear labeling of information as Explicit, Observed, Proposed, or Unknown. This design prevents the agent from fabricating organizational policies or presenting assumptions as facts.
- Defensive Instruction Set: The test scenarios (references/test-scenarios.md) include specific guidance on handling prompt injection in retrieved documents, instructing the agent to treat such content as data rather than instructions, which serves as a protective measure for the agent's behavior.
- Safe Skill Drafting: While the skill contains a "Skill Generator" (references/skill-generator.md), its function is limited to drafting Markdown specifications for new skills. It explicitly states it cannot autonomously deploy or overwrite live skills, maintaining a safe boundary between analysis and deployment.
Audit Metadata