openagent
Fail
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: HIGHPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill instructs the agent to execute the
sudo 5dive agent importcommand to provision new teammates. Running commands with root privileges poses a significant security risk to the host environment if the agent's instructions are compromised. - [COMMAND_EXECUTION]: The skill makes extensive use of shell commands through
npxand the5diveCLI to perform validation, rarity tier checks, and image rendering. - [EXTERNAL_DOWNLOADS]: The skill uses
curlto fetch images from pollinations.ai, which is a well-known service for text-to-image generation. It also downloads assets and code from the NPM registry vianpxand references repositories under the5dive-aiorganization on GitHub. - [REMOTE_CODE_EXECUTION]: The skill repeatedly calls
npx @5dive/openagent@0.34.0, which downloads and executes the package from the NPM registry at runtime. While these are vendor-provided resources from the skill's author, the execution of remote scripts is a critical attack vector. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from user-supplied
<id>.persona.yamlfiles. - Ingestion points: Multiple commands (validate, tier, card, import) read data from the persona YAML file.
- Boundary markers: The skill relies on internal schema validation via the
validatecommand but does not mention explicit delimiters to prevent instructions within the YAML from affecting the agent's behavior during rendering or provisioning. - Capability inventory: The agent can execute arbitrary shell commands (via
npx), perform network operations (viacurl), and provision system agents (viasudo 5dive agent import). - Sanitization: The skill performs schema validation (
additionalProperties: false) to ensure data conforms to expected fields.
Recommendations
- AI detected serious security threats
Audit Metadata