LunaForge Code Context

Experimental local code-context and change-risk engine for VS Code and AI-agent governance.

LunaForge is being rebuilt around one narrow, verifiable workflow:

Map real internal dependencies → explain change impact → evaluate architecture policy → export evidence.

It is intended to become the code-context engine used by OpenSyber before an AI agent is allowed to modify a repository.

Current preview capabilities

• Reads actual TypeScript, JavaScript, and Python workspace files locally.
• Extracts internal module imports and exports.
• Resolves relative dependencies and Python module paths.
• Detects dependency cycles and coupling hotspots.
• Computes direct and transitive dependents for the current file.
• Evaluates configurable Guardian deny rules.
• Exports a versioned JSON evidence bundle.

Explicit limitations

LunaForge is an experimental preview. It does not currently claim:

• Whole-program semantic or call-graph analysis.
• Runtime tracing or service discovery.
• Autonomous code changes.
• Team collaboration or enterprise tenancy.
• SOC 2 certification, GDPR compliance certification, or end-to-end encryption.
• Verified performance on 10,000-file repositories.

The older experimental mode packages remain in the monorepo for research, but only Galaxy, CodeFlow, and Guardian are activated by the VS Code extension.

Commands

Command	Purpose
LunaForge: Open Code Context	Open the focused dashboard
LunaForge: Build Dependency Graph	Parse supported source files
LunaForge: Analyze Current File Impact	Explain direct and transitive blast radius
LunaForge: Run Guardian Policies	Check architecture boundary violations
LunaForge: Export Evidence JSON	Export graph and impact evidence

Guardian rule example

{
  "lunaforge.guardian.rules": [
    {
      "id": "domain-no-infra",
      "name": "Domain must not import infrastructure",
      "fromPattern": "src/domain/**",
      "toPattern": "src/infrastructure/**",
      "effect": "deny"
    }
  ]
}

Development

git clone https://github.com/finsavvyai/lunaforge.git
cd lunaforge
npm install
npm run build
npm run build:extension

CI/CD

LunaForge uses PushCI instead of GitHub Actions. After applying the overlay:

./scripts/pushci-run.sh run --dry-run
./scripts/pushci-run.sh run --parallel --security --verbose --all

The root pushci.yml validates the MCP package, website release artifacts, and Wrangler production bundle locally.

Strategic direction

LunaForge is not positioned as another general AI coding assistant. Its target role is model-independent context and risk intelligence:

1. An agent proposes a change.
2. LunaForge identifies affected code and policy boundaries.
3. OpenSyber allows, blocks, or requires approval.
4. LunaForge exports an evidence-backed change receipt.

For the complete capability inventory, current limitations, in-progress work, and future roadmap, see docs/CAPABILITIES-AND-ROADMAP.md.

License

MIT