duvo-cli
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions involve executing the
duvocommand-line utility. This is the primary purpose of the skill, and the commands are used to manage agents, runs, and other resources on the Duvo platform. - [EXTERNAL_DOWNLOADS]: The skill documents the installation of the
@duvoai/clipackage from the npm registry. As this is the official package provided by the vendor (duvoai) for the tool being documented, it is considered a legitimate and safe external reference. - [INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection because it reads and processes data from external sources (the Duvo API, such as case data or agent configurations).
- Ingestion points: Data enters the agent's context through commands like
duvo agents list,duvo runs get, andduvo cases list. - Boundary markers: The instructions emphasize using the
--jsonflag and parsing withjqto handle structured data, which provides a layer of separation between data and instructions. - Capability inventory: The agent can perform file operations (
duvo files ...), manage connections, and execute API requests. - Sanitization: There is no explicit mention of sanitizing the API output before processing, which is typical for a tool-documentation skill.
- [CREDENTIALS_UNSAFE]: The skill handles authentication via API keys and OAuth. It correctly instructs the user on how to provide keys (using placeholders like
<key>) and environment variables (DUVO_API_KEY). It does not contain any hardcoded secrets or sensitive credentials.
Audit Metadata