Mitigrid is permanently frozen as an open-source PoC/research artifact; see the
final V0 stop decision. No
generalized runner exists, no Docker execution is authorized, and the retained
material is not a security-product release or a qualified security conclusion.
The preserved safety references are:
They remain methodology constraints only. The prototype does not accept arbitrary Docker Compose input and does not support remote targets, experiment-stage external networking, privileged containers, host-kernel exploits, container escapes, destructive payloads, Kubernetes, or physical OT systems. The fixed V0 Compose fixture will not run; its static and synthetic outputs are provisional development evidence, not qualified security conclusions.
A probe failure alone is never evidence that a defense worked. Successful patch or mitigation qualification would require valid infrastructure, a reproduced vulnerable control, a blocked declared security effect, and preserved legitimate functionality.
The OpenCEM material below describes the historical v0.1 implementation and remains applicable to its preserved source release.
Mitigrid has no public security-product release, supported runtime, or active development line. The frozen artifact does not receive V0 completion or engine development.
OpenCEM v0.1 is a separate historical source release, not an actively supported product line. A credible vulnerability affecting that preserved release will trigger an explicit decision about whether to create a narrowly scoped maintenance branch.
Do not report suspected vulnerabilities in a public issue.
Use GitHub private vulnerability reporting if it is enabled for the repository. If it is not enabled or you cannot access it, email the maintainer at ro.peneder@gmail.com.
Include:
- affected command, crate, schema, or workflow
- steps to reproduce
- expected and observed behavior
- any known impact on local files, generated reports, imported data, or command execution
Security-sensitive areas include:
- manifest parsing and schema validation
- archive packing and diffing
- report generation
- importers for external formats
- model-run command execution
OpenCEM can execute model-run commands declared by a local project manifest. Treat manifests from untrusted projects like executable code. opencem run <run-id> refuses to execute by default; use opencem run --dry-run <run-id> to inspect the command, cwd, inputs, and outputs, and use opencem run --allow-exec <run-id> only after trusting the project.
YAML project inputs are also inside the local-project trust boundary. This
includes opencem.yaml, project.yaml, generated .opencem/*.yaml import
indexes, domain-pack sidecars such as labtrace.yaml and thermal-basic.yaml,
and YAML fixtures imported from external tools. Do not parse untrusted YAML as a
privileged operation or service request. Review or sandbox untrusted projects
before running OpenCEM commands against them.
Artifact paths and generated or imported evidence paths must resolve to regular files under the canonical project root. Symlink artifacts are rejected instead of followed so untrusted manifests cannot cause ingest, check, report, pack, or import workflows to read files outside the project.
The project will coordinate credible reports privately before any public disclosure. A frozen Mitigrid PoC has no product release or advisory stream; any advisory decision for the preserved OpenCEM source release remains case-specific.