openagent
Warn
Audited by Gen Agent Trust Hub on Sep 13, 2026
Risk Level: MEDIUMPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill instructs the agent to use
sudofor the5dive agent importcommand to provision new agents. While this is part of the agent lifecycle management, the use ofsudoinvolves acquiring administrative privileges. - [COMMAND_EXECUTION]: The skill executes multiple shell commands using the
npxutility and the5diveCLI for validating personas, rendering cards, and importing agents. - [EXTERNAL_DOWNLOADS]: The skill uses
curlto fetch images from Pollinations.ai, an external image generation service. It also fetches the@5dive/openagentpackage from the npm registry usingnpx. - [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided identity data (name, role, behavior) and incorporates it into a persona file. \n
- Ingestion points: User identity details gathered in chat are written to a
<id>.persona.yamlfile. \n - Boundary markers: Identity data is encapsulated within a structured YAML schema. \n
- Capability inventory: The skill can execute shell commands via
npxandsudo 5dive, and perform network operations viacurl. \n - Sanitization: The skill includes instructions to validate the persona file against a schema and enforce a pattern for the identity handle.
Audit Metadata