Security

ClawBio skills run locally, process sensitive genetic data, and are invoked by AI agents. These are the security principles we follow.

Core Principles

1. Local execution: genetic data never leaves the user's machine
2. No network access: skills should not make outbound network calls during analysis
3. Filesystem discipline: skills should only read inputs and write to declared output paths
4. No shell injection: no eval, exec, or subprocess with shell=True

Security Checklist for Skills

Check	Description
No network calls	No requests, urllib, httpx, socket imports in analysis code
Filesystem scope	Skills only access files under their declared input/output paths
No shell injection	No os.system, subprocess.call with shell=True, or eval/exec
No hardcoded secrets	No API keys, tokens, or credentials in source code
Input validation	All inputs validated before processing (type, format, size)
Pinned dependencies	All dependency versions pinned in requirements

Best Practices for Skill Authors

• Never import networking libraries in analysis code
• Never use eval, exec, or subprocess with shell=True
• Validate all inputs before processing (check file format, size, variant count)
• Pin all dependency versions in your requirements
• Include tests with expected output checksums
• Document all filesystem paths your skill accesses in the SKILL.md
• Always include the ClawBio safety disclaimer in report output

Reporting Issues

If you discover a security issue in ClawBio or any published skill, please open a private security advisory on the GitHub repository.