Security
How Ensemble separates user authority, agent work, and OpenCode's native permissions. These pages are written for security reviewers and operators who need to assess the product before adopting it.
Security documents
| Document | What it covers |
|---|---|
| Threat model | Storage, filesystem and shell, tools, cross-workspace access, skills, approvals, model fallback, and logs, with code evidence |
| Permission-model review | Approvals, outside-roster contact, external directories, self-approval, and native permission preservation |
| Accepted risks | Dependency advisories and residual risks with review dates and compensating controls |
| Security policy | Private vulnerability reporting and the maintainer response process |
Reporting a vulnerability
Report vulnerabilities privately through the process in SECURITY.md. Never open a public issue, and never post credentials, tokens, conversation content, private paths, or unredacted logs. The support policy explains how to redact public evidence for ordinary support requests.
Boundaries
- Ensemble does not replace OpenCode's tool permissions. An Ensemble approval is a product workflow decision, not a grant of host permission.
- The leader coordinates work but receives no additional permissions.
- Agent skills are guidance, not tool grants.
- Diagnostics are local by default. Ensemble has no telemetry.
Related
- Privacy: data classes, retention, export, and delete.
- Operations: backup, restore, and incident procedures.
- Evidence index: security review records and verification evidence.