Skip to content

Security: vaur94/oma

Security

SECURITY.md

Security Policy

Supported Versions

Version Supported
2.0.x
1.8.x ⚠️
< 1.8

We recommend running the latest release. Security fixes are backported only to the latest minor version of the current major version.

Reporting a Vulnerability

Please do not open a public issue for security reports.

Report vulnerabilities through one of these channels:

  1. GitHub Security Advisories (preferred): https://github.com/vaur94/oma/security/advisories/new

  2. Email: Send details to the maintainers listed in package.json.

Include as much detail as possible to help us reproduce and verify the issue:

  • Description of the vulnerability
  • Steps to reproduce
  • Affected version(s)
  • Any known mitigations

Response Timeline

Severity Initial Response Fix Timeline
Critical 24 hours 7 days
High 48 hours 14 days
Medium 5 business days 30 days
Low 10 business days Next release

Security Boundaries

OMA runs with the same permissions as the user who invokes it. Key boundaries:

  • File access: Controlled by the embedded permission engine. File read/write is scoped to the workspace unless explicitly authorized.
  • Shell execution: run_command tool executes on the host system. Restrict via commandSurface config.
  • Network: OMA makes outbound HTTPS requests to LLM providers only. No inbound listeners are opened.
  • Environment variables: API keys must be provided via environment — never hardcode credentials.
  • Config loading: Config files are read locally; never load config from untrusted paths.

Configuration Security Recommendations

{
  "commandSurface": {
    "allowDangerousCommands": false,
    "restrictedCommands": ["rm -rf"],
  },
}
  • Keep allowDangerousCommands disabled in production.
  • Use .oma/oma.jsonc for project-local config instead of global config when working with untrusted repositories.
  • Rotate API keys regularly.

Dependency Security

Run npm audit or bun audit periodically to check for known vulnerabilities in dependencies.

Credits

We follow a coordinated disclosure process. Credit will be given to reporters unless they request anonymity.

There aren’t any published security advisories