Open-source tools and experiments for AI agents, plus games and other projects. The open-source ones live under my org Beholdr Labs or my own GitHub.
Configurable subagent routing for Claude Code and Codex. Cheaper worker models read large codebases, research the web, or apply mechanical edits, and hand short sourced briefs back to the primary model, which keeps the design, decisions, and review. In my evals it cut the bill by 40–48% on a 453k-line codebase with the same facts found, and cost more on small tasks, so the skill now declines those.
A cognitive assistant for your AI agent that makes complex tasks easier to get through. It keeps the context for you, turns a big task into one clear next step, and makes decisions and reviews lighter. Works for coding and everyday tasks.
Local, deterministic CV linting for PDF, DOCX, and Markdown. It checks structure, text extraction, formatting, and spelling, and gives AI agents source-located evidence to review a CV with. Runs as a CLI, an MCP server, or a Claude Desktop extension, and never calls a model or edits the document.
A real-time asteroid mining space strategy game in Godot, inspired by Fragile Allegiance.
