Skip to content

Security: rootellect/rootellectmcp

Security

docs/security.md

Security Model

  • No unrestricted shell tool is exposed.
  • File writes go through PermissionService and FilesystemService.
  • .env, secrets, database, auth, payment, Prisma, and deployment files are blocked or require approval.
  • Git commits are blocked on main and master.
  • Secret scanning blocks probable tokens before commit.
  • Production deployment is disabled by default and requires explicit approval text.
  • Audit entries are append-only JSON Lines with redaction.

Known limitation: deterministic visual checks do not replace human visual review.

Version 2 security notes

  • Audit mode is read-only and does not write website files.
  • Agent results must include evidence from repository inspection, Playwright, axe-core, Lighthouse, browser logs, or deterministic rules.
  • Protected-file policy is not weakened by Version 2. Auth, payment, orders, database, environment, and production deployment paths remain blocked or approval-gated.
  • finalize_multi_agent_edit commits only after scope validation, secret scanning, validation, re-audit, and screenshot comparison pass critical gates.
  • Vercel preview creation is optional and configuration-dependent.
  • Production deployment remains stopped by policy and requires explicit separate approval through the existing production tool.

There aren't any published security advisories