Skip to content
Jeremy Longworth edited this page Sep 12, 2026 · 4 revisions

ChefSkills Wiki

ChefSkills is an open-source culinary Agent Skills framework for building AI agents that reason about cooking, ingredients, techniques, routing, state, safety, and evaluation.

The current public release is v0.1.0-public-preview. The initial public alpha prerelease is v0.1.0-alpha.

Repository: github.com/jeremylongworth-source/ChefSkills

Choose a path

You want to... Start here
Install a skill for GitHub Copilot GitHub Copilot Setup
Clone the repository and run checks Getting Started
Understand the framework Architecture
Choose a skill or route a request Skill Catalog and Routing and State Model
Review evidence and scoring Evaluation Workflow
Review safety boundaries and source use Safety and Source Checks
Propose or review a change Contributing
Maintain releases and repository controls Maintainer Operations

At a glance

ChefSkills separates culinary behavior into five layers:

  • focused skills under skills/
  • installable YAML bundles under skillsets/
  • routing rules under router/
  • cooking-state concepts under state/
  • scenarios, reports, scorecards, and safety gates under evaluation/

The project is a behavior framework, not a recipe corpus. Its purpose is to make culinary AI more inspectable, easier to test, and clearer about uncertainty.

Current evidence

The repository currently contains 9 evaluation reports covering 38 fixtures with no recorded blockers. These before/after outputs are medium-confidence local reviewer simulations. They are useful for regression tracking, but they are not live benchmark results, external validation, certification, or a guarantee of future model behavior.

A provider-neutral live-output harness is available. The first foundation smoke packet is pending external capture and scoring; see Evaluation Workflow.

Boundaries

Food safety is a hard gate. ChefSkills does not certify legal, medical, regulatory, commercial-kitchen, allergen, nutrition, or food-safety compliance. Safety-sensitive work requires current authoritative sources, conservative uncertainty handling, and escalation to the relevant authority or qualified professional when appropriate.

Reference map

Clone this wiki locally