Skip to content

Security: mcp-tool-shop-org/role-os

Security

SECURITY.md

Security Policy

Supported Versions

Version Supported
2.x Yes
< 2.0 No

Reporting a Vulnerability

Email: 64996768+mcp-tool-shop@users.noreply.github.com

Include:

  • Description of the vulnerability
  • Steps to reproduce
  • Version affected
  • Potential impact

Response timeline

Action Target
Acknowledge report 48 hours
Assess severity 7 days
Release fix 30 days

Scope

Role OS is a CLI that scaffolds role contracts, routes work, and runs persistent multi-step missions in a repository. By default it operates on the local filesystem only.

  • Data touched (default): Local filesystem only — copies markdown templates into the .claude/ directory; creates packet, verdict, run, manifest, and hook files
  • No network egress by default — default operation makes no network requests
  • No secrets handling — does not read, store, or transmit credentials
  • No telemetry is collected or sent

Opt-in network paths

Three features perform network egress when explicitly enabled. All are off by default and fail open to local deterministic behavior:

Feature How enabled What is transmitted, and where
Citation gate (roleos verify-citations) Running the command (requires the external prism CLI) Citation identifiers/URLs from the artifact, resolved against public arXiv/Crossref APIs by prism
Specialist tier Registering a specialist with a backend_url in .role-os/specialists.json Dispatch prompts/step context POSTed to the configured backend_url (typically a local model endpoint)
Budget / conformance consult ROLEOS_BUDGET_CONSULT=1 / ROLEOS_CONFORMANCE_CONSULT=1 Step or tool-call context sent over HTTP to the configured local model endpoint for an advisory verdict

Vulnerability reports for these paths (the HTTP client in src/specialist/, the prism shell-out in src/verify-citations.mjs, and the consult endpoints) are in scope.

There aren't any published security advisories