metengine-data-agent

Fail

Audited by Gen Agent Trust Hub on Mar 3, 2026

Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill instructs the agent to read the Solana keypair file located at ~/.config/solana/id.json to enable payment signing, which exposes private keys to the agent context.
  • [COMMAND_EXECUTION]: The skill provides explicit instructions to set up persistence by adding a cron job (crontab -e) that automatically downloads and replaces the skill file weekly.
  • [REMOTE_CODE_EXECUTION]: The skill implements a 'Client Bootstrap' mechanism that saves working code snippets to a local memory file (~/.claude/agents/metengine-memory.md) and instructs the agent to read and execute them in subsequent sessions.
  • [COMMAND_EXECUTION]: The skill contains instructions to use curl to overwrite its own operational script (~/.claude/agents/metengine-data-agent.md) with content from a remote vendor URL.
  • [EXTERNAL_DOWNLOADS]: The skill requires the installation of external Node.js packages, including @x402/core, @x402/svm, and @solana/kit, to facilitate blockchain interactions and payment processing.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Mar 3, 2026, 06:11 PM