The OpenCode Agent Loop project takes security seriously. If you discover a security vulnerability, please report it privately. Do not report security vulnerabilities through public GitHub issues.

Send a detailed report to the maintainers by opening a draft security advisory on GitHub, or if that is unavailable, by contacting the project maintainers directly through a private channel.

Please include:

• A description of the vulnerability
• Steps to reproduce
• Affected versions
• Any potential mitigations you've identified

You should receive a response within 72 hours. If you don't, please follow up.

• Acknowledgment of your report within 3 business days
• An assessment of the vulnerability's severity and impact
• A timeline for a fix and release
• Credit in the release notes and SECURITY.md (unless you prefer to remain anonymous)

This security policy covers the opencode-agent-loop package and its official plugins and commands. It does not cover:

• Third-party model providers
• OpenCode itself
• Projects that use this package

We consider security research conducted under this policy to be:

• Authorized under the Computer Fraud and Abuse Act
• Exempt from DMCA anti-circumvention provisions
• Not a violation of our terms of service

You agree not to:

• Access or modify user data without permission
• Disrupt production services
• Exploit vulnerabilities beyond what is necessary to confirm the issue

This project implements the following security practices:

• All agents are read-only or edit-restricted by default
• Destructive git operations (push, reset, clean) are denied for all agents
• Agent-loop recursion is blocked for worker processes
• Privacy-aware model routing with data-policy classification
• Secret detection in staged changes before commit
• Path confinement through allowed-tool scoping