Security
MODRISS combines user-authored models, generated files, external AI calls, and cloud deployment artifacts. Security controls therefore span the platform runtime and the projects it generates.
Platform Controls
- Protected REST commands require an
X-Auth-Tokensession token. - Project access is checked against owner, editor, and viewer membership.
- Model revisions prevent silent stale overwrites.
- PostgreSQL foreign keys and checks enforce key ownership and enum boundaries.
- Multipart uploads, generated files, file counts, and artifact size have configurable limits.
- Artifact paths reject absolute paths and directory traversal.
- MDE work has execution, queue, concurrency, and captured-output limits.
- Request logging includes a correlation ID.
Assistant Controls
- AI is disabled unless explicitly enabled.
- The assistant runs as one unified durable CIM/PIM workflow with a strict adaptive strategy, an obligation-gated conceptual path, an inspect/contract path, and an enforced read-only answer path. There are no delegated model workers and no client-selectable strategy.
- The assistant uses backend-owned metamodel contracts, selected model state, optional attachments, and durable recent conversation context rather than raw EVL files or unchecked database access.
- Conceptual objects remain private until mandatory-obligation evidence and live Ecore structure pass. All mutation paths are structurally validated, committed against the expected model revision, and audited through durable turns.
- EVL and stored semantic validation are never assistant apply/repair/commit gates.
- Destructive batches require explicit confirmation; committed checkpoints may include inverse patches for undo.
- Provider calls have timeouts, retries, rate limits, a circuit breaker, and an optional dedicated proxy.
- Provider credentials belong in environment variables or a secret manager, never committed files.
Deployment Hardening
The development Compose stack uses convenient local defaults. Before public deployment:
- Terminate TLS and restrict all exposed ports.
- Replace default database credentials and rotate secrets.
- Restrict CORS origins to deployed frontend origins.
- Restrict Actuator and database access.
- Configure backups, restoration tests, log retention, monitoring, and alerting.
- Define AI provider cost, privacy, retention, and acceptable-use policy.
- Review upload, artifact-editing, and download controls for the deployment threat model.
- Patch base images and dependencies through a managed release process.
Generated Project Security
Generated AWS projects include IAM rationale, security review, manual-action reports, tests, and protected regions. They still require human review.
Before deployment:
- Replace secret and owner placeholders.
- Review IAM permissions for least privilege.
- Implement protected business logic.
- Run generated security, contract, and integration tests.
- Inspect infrastructure templates, API contracts, event schemas, and workflow definitions.
- Deploy first to a dedicated non-production account.
Floci and LocalStack prove much of the packaging and integration behavior, but neither is a substitute for a real AWS security review.