InfraLens is designed to run on your machine by default.

• Repository scans and evidence-graph analysis execute locally.
• Project contents are not transmitted to InfraLens-operated cloud services by the default demo or local workbench path.
• Hosted static demos, if deployed, should use bundled synthetic estates only—not user uploads.

Treat any future remote deployment as an explicit operational choice with its own threat model.

Preview software may still contain sharp edges. Security reports for 0.1.x are welcome and prioritised over feature work when impact is clear.

The scanner reads potentially hostile repositories and manifests. Relevant threats include:

• Path traversal and symlink escape outside the scan root
• Recursive symlink / zip-bomb style denial of service
• Oversized files, deep trees, and YAML alias bombs
• Malicious labels or HTML/SVG content reaching the UI
• Secret leakage into scan bundles or reports
• Unsafe subprocess invocation or plugin execution
• Over-privileged live cloud/cluster access

Controls in the current design include:

• Bounded file size, depth, and document counts
• Symlink escape and recursion guards during tree walk
• Adapter isolation (one adapter failure does not abort the scan)
• Secret redaction before persistence; no raw secrets in bundles by policy
• Bundle structural validation and content hashing
• UI rendering of labels as text; project-local SVG icons (not fetched per scan)
• Live connectors disabled by default, credential-reference based, and intended to be read-only

Deeper notes: docs/security/threat-model.md and docs/security/scanner-security.md.

InfraLens is not marketed as a fully sandboxed multi-tenant scanner. Do not point it at untrusted repositories on shared production hosts without additional isolation.

• Prefer imported inventory snapshots for demos and CI.
• Live AWS / Kubernetes / OpenShift discovery requires an explicit credential reference and operator-provided credentials outside the bundle.
• Connectors must use least-privilege, read-only permissions.
• Never commit credentials into the repository, fixtures, or generated bundles.
• requires-credentials in the capability matrix means an operational dependency—not a licence lock.

Discovery adapters and the scanner pipeline should redact secret-shaped keys and values before writing evidence bundles or reports. If you find a redaction miss (for example a token surviving into fixtures/scans/*.json or HTML reports), treat it as a vulnerability.

Please use GitHub private vulnerability reporting for this repository (Security → Report a vulnerability).

Do not open a public issue for active vulnerabilities. Do not email personal maintainer addresses for security reports.

Include:

• Affected version / commit
• Reproduction steps or a minimal fixture
• Impact assessment (secret leak, RCE, path escape, DoS, etc.)
• Any known workarounds

We will acknowledge reports as promptly as a preview-stage volunteer project allows and coordinate disclosure once a fix or mitigation is available.