experiment-code — security audits

Flagged by auditors

Independent security partners scan this skill's files for malicious content, prompt injection and risky patterns. Verdicts below come from the skills.sh audit program; each partner links to its full report.

• Gen Agent Trust Hub

  fail

  This skill is designed to automate machine learning research by generating and executing Python code. Its primary security risk is the execution of model-generated code through subprocess calls based on untrusted user input ($1). A malicious research plan or idea could manipulate the agent into generating and running harmful scripts, potentially leading to unauthorized system access or data exfiltration.

  Risk: HIGHCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONFeb 22, 2026Full report

• Socket

  pass

  No alerts

  Mar 18, 2026Full report

• Snyk

  pass

  Risk: LOW · No issues

  Risk: LOWFeb 22, 2026Full report

• Runlayer

  pass

  2/3 files flagged

  Risk: LOWFeb 22, 2026Full report