Projects
All MIT-licensed. All on GitLab.
Developer Infrastructure
CLI tools that manage the working environment for AI-augmented development.
Spec management for AI coordination
Six-phase workflow state machine with human gates. Task DAGs, propagation chains, stakeholder dispositions, branched sessions. The same CLI interface for the human and the agent.
Skill registry manager for AI coding agents
Syncs reusable agent instructions from git-based registries. Bidirectional sync, multi-registry support, drift detection via Claude Code hook. Platform-agnostic.
Verified tool registry manager
Resolves tool versions across git-based registries. Generates mise configuration. Verifies checksums and signatures against upstream publishers. S-2 integrity: verification failure halts the install.
Tmux session manager for AI coding workflows
Organizes tmux sessions into verticals and remotes. Save and restore across reboots. Server isolation for demos and CI. Pre-create hooks for environment setup.
Context Engineering
Managing the context that makes AI effective on long-running work.
Evaluation
Benchmark harnesses for measuring AI on structured tasks.
AI on SysML v2 model comprehension
Reproducible evaluation across models, tool configurations, and corpus scales. Per-field structured scoring. Available as Python package, Docker container, and HuggingFace dataset.
GitLab Knowledge Graph for SDLC queries
SDLC query evaluation framework contributed to orbit/knowledge-graph. 62 fixtures across 6 ontology domains, 3 models, 5 tool description conditions. Runs against the orbit query compiler with a DuckDB simulation layer.
MBSE Toolchain
Parser, CLI, and grammar tools for model-based systems engineering languages. The industries that heavily adopt MBSE (defense, aerospace, automotive, medical devices) are typically slower to adopt modern AI tooling. We want to help them get there through open instrumentation.
SysML v2 CLI with MCP server
Indexes .sysml repositories into a persistent knowledge graph.
14 CLI commands and 10 MCP tools. Hybrid search (8 keyword signals + vector).
94% average token reduction vs raw file injection.
Tree-sitter grammar for SysML v2
The parsing foundation. Built by iterating against a corpus of real-world SysML v2 models. Bindings for Rust, C, Node.js, Python, Go, and Swift.
KeBNF grammar converter
Converts OMG KeBNF specifications to ANTLR4 and tree-sitter. Parses all 640 KerML + SysML v2 rules. Bridges the gap between OMG specs and working parsers.
kebnf's grammar conversion pipeline is designed to extend beyond SysML. These specification languages are on our horizon. If you work with any of them and want parser tooling, we'd like to talk.