Local security scanning, finding review and checks before you push.
Quickstart · Daily workflow · Scanners · Release status · Security
CodeSecurity brings scanner results into one local workbench and your existing VS Code profile. Inspect a finding and its source, record a review decision, and check the committed history you intend to push. Ordinary scans do not invoke AI. An optional Advisor helps explain findings without deciding the gate.
0.1 source preview. Build the backend and extension from source. Mac ARM64 and Ubuntu AMD64 have scoped functional acceptance; Fedora qualification is partial, and Windows native and Remote WSL remain experimental. Marketplace publication is pending. See tested scope and known issues.
The Findings view, using synthetic demonstration data from the browser tests.
The real VS Code extension with synthetic demonstration finding data. Review findings in the sidebar, inspect their lifecycle and gate effect in a code hover, or navigate from Problems. The hover includes direct Web-detail and Advisor actions. Open the image for the full-size view.
| Task | In the workbench |
|---|---|
| Scan code and dependencies | Run eight pinned external tools for application security, secrets, infrastructure checks, vulnerabilities and an SBOM. Tool roles and limits. |
| Review findings | Filter results, inspect bounded code evidence, open the exact source in VS Code and record reasoned lifecycle decisions. |
| Manage repositories | Select independent repositories, authorize narrow source roots and keep each project's scans and policy separate. |
| Check outgoing history | Apply protection deliberately to a clone. Completed evidence and the configured policy determine whether its push proceeds. |
| Understand a result | Use optional local Ollama, or explicitly authorize a remote Advisor with a single-finding disclosure preview. Advice never changes the deterministic gate. |
The external tools are Trivy, Gitleaks, OpenGrep, Bearer, Grype, Checkov, TruffleHog and Syft. They retain their own licenses. Internal heuristic scanners are optional and disabled by default.
First follow platform setup. You need Python 3.12+, Git, Make and a local Linux Docker engine with Compose and Buildx. A fresh build requires network access and 32 GiB free on both source/state and Docker build storage; a restart with matching images has a separate 2 GiB floor. These are admission floors, not measured steady-state usage.
On macOS or Linux:
git clone https://github.com/lomolabs/lomo-codesecurity.git
cd lomo-codesecurity
python3 scripts/start-lomo.py --check --json
python3 scripts/start-lomo.py--check never downloads its storage-probe image. On a pristine engine,
engine_storage_probe_missing means normal startup needs to obtain that pinned
dependency first. For other failed checks, follow the report's recovery guidance.
The launcher builds and starts the backend. It does not install prerequisites, change your Docker configuration, install an extension or open an editor. Repeat the startup command to retain installation data and pairing and reuse matching builds. Keep the checkout separate from the repositories being scanned.
The default address is http://127.0.0.1:8080.
Choose another local port
if it is already in use.
Windows uses a WSL2 backend. Follow the separate Windows startup guide; native Windows VS Code and Remote WSL are different extension hosts.
Install the documented Node.js/npm prerequisites, then build the preview extension:
make vscode-packageIn your existing VS Code profile, run Extensions: Install from VSIX… and
select .kodesec-test/vscode/code-security-0.1.0.vsix. Open your project and run
CodeSecurity: Get Started. The extension ID is lomo-dev.code-security.
Installing it does not start the backend or replace your other extensions.
Review Workspace Trust normally. The first eligible profile uses the installation's first-owner connection flow. Additional profiles need owner approval in Settings → Trusted access → IDE access, using the displayed verification code and profile name. No access secret needs to be pasted into the editor. Browser-only setup is also available.
- Authorize the source. Select the repository and review its narrow source authorization. Selecting a child never grants its parent or siblings.
- Prepare scanners. Choose Tools → Prepare all scanners. The separate updater verifies data and rules without access to your source. Wait for the enabled scanners to be ready.
- Scan and review. Scan the repository, folder or supported file. Use Findings, Problems and Open Finding in Web to inspect results and coverage. A completed scan can correctly fail its policy gate.
- Choose protection. Review thresholds, then use Protect Pushes… to approve and apply protection to that clone. Check Outgoing Commits… checks the committed history you plan to push. A workspace scan alone does not install a hook.
The workflow guide explains the controls in detail.
Incomplete, cancelled, failed or stale evidence cannot authorize a protected
push. Local commits remain available. Hooks can be bypassed with --no-verify
or another clone; use remote branch controls for unavoidable enforcement.
The preview maintains scanner data through reviewed source updates. When definitions expire, obtain refreshed source, run normal startup to rebuild changed inputs, and prepare again. Downloading an expired definition again does not renew it. Preparation and expiry.
Preserve the installation's owner-private runtime state and Docker named volumes when updating. Startup and recovery and data persistence describe what to keep. Backup/restore helpers are not a validated disaster-recovery facility.
The default origin is loopback-only. Source access uses narrowly authorized read-only mounts or bounded snapshots. Runtime scanners operate offline against validated data; redacted findings and bounded snippets persist locally. Optional remote Advisor requests require separate authorization and a disclosure preview.
CodeSecurity is not a sandbox for hostile repositories. A passing scan does not guarantee secure code, and redaction cannot recognize every secret. Read the security model and platform limits.
For startup or runtime trouble, run make support-bundle in the backend checkout.
It collects bounded local diagnostics without uploading anything or restarting
services. Review the bundle before sharing.
Use Issues for ordinary
bugs and private vulnerability reporting
for security reports.
Start contributions with CONTRIBUTING.md, the development guide and architecture. The CI guide explains runner requirements, retained diagnostics and the intentional vulnerable test fixtures.
Project-authored source uses Apache-2.0. Third-party engines,
dependencies, rules and fixtures retain their own terms. See the
scanner license index and the Web build's
generated THIRD_PARTY_LICENSES.md. These notices do not establish rights
clearance for future distributions or grant trademark rights.

