I build repository intelligence for coding agents.
I’m the founder of Marcelle Labs and the creator of Vreko. Vreko turns repository structure, history, and constraints into durable, machine-readable evidence that helps agents make better engineering decisions.
The open substrate beneath that work is workspace.json—a portable specification for describing what agents need to understand about a codebase without repeatedly rediscovering it.
-
Vreko CLI
Local-first repository intelligence and agent-facing development workflows. -
agents-audit
The reference producer, validator, and CLI forworkspace.json. -
Codex MCP
Deterministic repository evidence and change-scope safeguards for Codex. -
SOPR MCP
Research and tooling for reducing agent tool-selection overhead through structured routing.
- Machine-readable repository context
- MCP servers and agent interoperability
- Evidence-driven change safeguards
- Repository history and co-change intelligence
- Agent memory, routing, and context efficiency
- TypeScript, Node.js, Next.js, VS Code, and GitHub Actions
I favor small, compatibility-first contributions that solve a maintainer’s existing problem before proposing a broader integration.
A recent example: my additive workspace.json compatibility change for Buildomator shipped in v3.4.6 with the original authorship and regression coverage preserved.
I’m currently exploring reporter and adapter boundaries where established analysis tools can produce portable repository evidence without taking a dependency on the full workspace.json ecosystem.
Agents should not have to rediscover the same repository facts every session—and safety-critical decisions should be grounded in inspectable evidence, not prompt folklore.