ocean-genomics

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.72). SKILL.md’s workflow text configures an MCP client (gget-mcp) whose tool calls can fetch BLAST/PDB/sequence content from public databases at runtime, and that returned text is then read by the agent/LLM—an outsider-source via public web/API content.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill instructs the agent to fetch and run remote skill/MCP packages at runtime (e.g., github.com/longevity-genie/gget-mcp and https://skills.sh/broomva/ocean-genomics), which will download/execute external code and directly extend/control agent behavior.