I build local-first AI tools and share what holds up in real use. My current Field Lab focus is turning authorized local repositories into installable Agent Skills, with clear provenance, boundaries, and validation.
- Try the free repo preflight: repo-distillation-preflight checks license signals and repository structure; it does not generate a Skill.
- See a real run: Input, command, output, and limits.
- Continue with Agentic Distiller: Product details.
Field Lab is where I build and document practical AI workflows for builders. I publish the inputs, outputs, failure cases, and limits so the work can be checked.
- ContextCore — User-owned agent context with MCP, approval, audit, and backup.
- evidence-first-research — Source-tracked workflows with explicit uncertainty and review gates.
- divergent-worlds — Forecasting and simulation experiments.
- Local-first software and user-owned data
- Explicit permissions and no silent writes
- Reproducible evidence and calibrated uncertainty