Please do NOT report security vulnerabilities through public GitHub issues.

Instead, use one of these channels:

1. GitHub Security Advisory (preferred): Go to the repository -> Security -> Advisories -> "Report a vulnerability"
2. Email: security@impertio.ai

• Description of the vulnerability
• Steps to reproduce (or proof-of-concept)
• Affected component (oa-cli, Visual Canvas, bridge server, etc.)
• Potential impact
• Any suggested mitigations

• Acknowledgment: Within 3 business days
• Initial assessment: Within 7 business days
• Fix or mitigation: Depends on severity — critical issues are prioritized

This security policy covers:

• oa-cli Python package (agent orchestration)
• Visual Canvas (TypeScript monorepo — frontend, backend, packages)
• Flask bridge server (bridge.py)
• VS Code extension (packages/vscode-extension)

Never commit credentials to the repository.

Open-Agents uses the following credential management patterns:

• Claude subscription: Managed by Claude Code CLI (~/.claude/)
• API keys: Stored in .env files (always in .gitignore)
• Agent workspaces: Temporary directories (/tmp/oa-agent-*) — never persisted to git

If you accidentally commit credentials:

1. Revoke the credential immediately (GitHub, Anthropic, etc.)
2. Use git filter-branch or GitHub's secret scanning remediation tools
3. Force-push a new clean commit

• Python (oa-cli): pip audit checks for known vulnerabilities
• Node.js (packages/): npm audit / pnpm audit for dependency security
• CI/CD: GitHub Actions runs security checks on every push

If you discover a vulnerability in a dependency:

1. Check if a patch is available
2. If yes: create a PR updating the dependency
3. If no: report directly to the dependency maintainers
4. If critical: open a GitHub Security Advisory describing the issue

• Regular dependency updates are made via dependabot (auto-PRs)
• Major updates are reviewed manually for breaking changes
• Security patches are prioritized

Open-Agents uses safe defaults:

If you use Open-Agents in production:

1. Keep agents isolated: Run on dedicated machines/containers
2. Monitor workspace cleanup: Use oa clean regularly to remove old workspaces
3. Audit agent logs: Review oa collect <name> and oa status regularly
4. Rotate credentials: If agents interact with external APIs, rotate tokens periodically
5. Network security: If using web UI, keep it on localhost or behind authentication
6. Update regularly: Stay on the latest version for security patches

For non-public inquiries: submit via GitHub Security Advisory (preferred) or email.

Open-Agents aims to comply with:

• OWASP Top 10 (mitigation where applicable)
• CWE/CAPEC common vulnerability patterns
• Python/Node.js security standards

Last updated: 2026-03-02