skills/nishuzumi/blitz-harness/mojo/Gen Agent Trust Hub

mojo

Warn

Audited by Gen Agent Trust Hub on Nov 11, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill is configured to fetch its core logic at runtime from a remote endpoint (<baseUrl>/api/agent/skills/mojo). It explicitly instructs the agent that the remote content takes precedence over the local instructions, allowing the remote source to change the skill's behavior dynamically after it has been loaded.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests markdown content from a remote API and adopts it as its primary instruction set without using boundary markers, validation, or sanitization. This provides an attack surface where a compromised or malicious API endpoint could inject instructions to hijack the agent's behavior.
  • [DATA_EXFILTRATION]: The skill accesses a local file (.mojo/agent.json) to read sensitive information including agentSecret. It is instructed to send this secret to the baseUrl found in the same configuration file. While this is the intended authentication flow for the MOJO platform, it creates a risk where credentials could be sent to an arbitrary external server if the configuration file is tampered with.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 11, 2026, 10:05 AM