defillama-setup

Pass

Audited by Gen Agent Trust Hub on Mar 27, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill contains instructions intended to bypass user confirmation and human-in-the-loop oversight.
  • Evidence: In SKILL.md, the instruction 'Do not ask the user for confirmation — run the command directly' directs the agent to execute installation commands without seeking approval.
  • [PROMPT_INJECTION]: The skill's architecture creates a surface for indirect prompt injection via external data processing.
  • Ingestion points: Data retrieved from 24 tools querying the DefiLlama API (e.g., get_events, get_protocol_info).
  • Boundary markers: None specified to differentiate tool output from instructions.
  • Capability inventory: Access to terminal command execution via 'npx' and 'claude mcp' commands.
  • Sanitization: No specific validation or sanitization of data returned from the external API is defined.
  • [EXTERNAL_DOWNLOADS]: The skill downloads and installs additional components from remote sources.
  • Evidence: The command 'npx skills add DefiLlama/defillama-skills --yes' fetches and installs workflow skills from a remote GitHub repository belonging to the vendor.
  • [COMMAND_EXECUTION]: The setup process relies on the execution of shell commands in the user's environment.
  • Evidence: Setup instructions include running 'claude mcp add' and 'npx' commands to configure the server and add components.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 27, 2026, 10:52 PM