ChefSkills

Portable culinary Agent Skills, skillsets, routing rules, state models, and evaluation scenarios for AI agents that need to reason like a chef.

Why This Exists

Most recipe assistants can produce plausible instructions. ChefSkills is intended to make an agent reason about food as a changing physical system:

• ingredients and their functional roles
• heat, time, moisture, fat, starch, acid, salt, texture, and aroma
• observable cues rather than time-only instructions
• substitutions, scaling, troubleshooting, planning, and safety boundaries
• kitchen workflow and recovery decisions

The project starts small on purpose. The first milestone is a testable operating framework, not a large recipe database.

The project started in private development and is now public for alpha review after validation, safety, documentation, and contribution workflow gates.

Status

ChefSkills is public. The current GitHub skill-install distribution release is v0.1.0-public-preview; the initial public alpha prerelease is v0.1.0-alpha. The current readiness state is ready_for_public_alpha_readiness_work.

Current evidence: 9 reports, 38 evaluated fixtures, 0 blockers, a baseline average of 3.7599, a ChefSkills-enabled average of 4.8059, and a delta of 1.0461.

Current caveat: before/after outputs are medium-confidence local reviewer simulations, not live captured model runs from a reproducible harness. The v0.1.0-alpha evidence decision accepts this for public alpha only because the limitation is visible and the release is not framed as a benchmark or certification claim.

Core Idea

ChefSkills separates culinary expertise into four layers:

• Skills: focused reusable instructions under skills/
• Skillsets: installable bundles under skillsets/
• Router: task classification and minimum useful skill selection under router/
• State model: structured cooking state and transformation concepts under state/

Food safety is a hard gate. It is not averaged against good cooking advice.

ChefSkills does not certify legal, medical, regulatory, commercial kitchen, allergen, or food-safety compliance. Safety-sensitive work should use current authoritative sources and conservative uncertainty wording.

Current Skills

Skill	Purpose
chef-core	Core chef operating model and safety-first behavior
culinary-reasoning	Mechanism-based culinary diagnosis and intervention
ingredient-knowledge	Ingredient roles, properties, and behavior
cooking-techniques	Technique selection, cues, and failure points
sauce-work	Sauce construction, finishing, scaling, and recovery
baking-structure	Baked-good structure, crumb, rise, set, and texture recovery
protein-cookery	Doneness, carryover, moisture, and safety-aware protein handling
fermentation	Home fermentation, brines, pH, salt, gas, spoilage, and storage boundaries
equipment-cookery	Appliances, vessels, heat transfer, capacity, airflow, pressure, and tool substitutions
ingredient-substitution	Functional substitution reasoning
recipe-development	Recipe design, formatting, and iteration
recipe-scaling	Scaling quantities, vessels, heat transfer, and workflow
food-safety	Safety hazard recognition and conservative guidance

Quick Start

Requirements:

• Python 3.10 or newer.
• PowerShell for the wrapper script, or a shell that can run the individual Python validators.

After cloning the repository, run:

.\scripts\validate-all.ps1

If PowerShell is unavailable, run the individual Python commands listed in the validation section.

GitHub Skill Install

ChefSkills can be previewed and installed through GitHub CLI agent skills for GitHub Copilot.

Preview a skill:

gh skill preview jeremylongworth-source/ChefSkills chef-core

Install a skill for GitHub Copilot at project scope:

gh skill install jeremylongworth-source/ChefSkills chef-core

Pin the public-preview release when reproducibility matters:

gh skill install jeremylongworth-source/ChefSkills chef-core --agent github-copilot --scope project --pin v0.1.0-public-preview

Install the food-safety skill:

gh skill install jeremylongworth-source/ChefSkills food-safety --agent github-copilot --scope project --pin v0.1.0-public-preview

gh skill install installs atomic skill folders from skills/, not YAML skillsets from skillsets/.

See GitHub Copilot and gh skill Setup for examples, scope guidance, and verification prompts.

First Use

Start with a skillset:

• skillsets/chef.yaml for broad culinary reasoning.
• skillsets/recipe-development.yaml for recipe design, adaptation, testing, and scaling.

Then route a prompt through the smallest useful skill chain. Example:

Prompt: My chicken thighs keep charring on the grill before they are cooked near the bone.
Route: chef-core, equipment-cookery, protein-cookery, food-safety, cooking-techniques

That route keeps equipment behavior, protein doneness, and food safety active without loading unrelated specialist skills.

Repository Structure

ChefSkills/
|-- .github/
|-- skills/
|-- skillsets/
|-- router/
|-- state/
|-- agents/
|-- docs/
|-- scripts/
`-- tests/

Validation

Run the local validators from the repository root:

python .\scripts\validate-skill-files.py
python .\scripts\validate-skillsets.py
python .\scripts\validate-scenarios.py
python .\scripts\validate-router.py
python .\scripts\validate-state.py
python .\scripts\validate-evaluation.py
python .\scripts\validate-live-evaluation-runs.py
python .\scripts\validate-evaluation-reports.py
python .\scripts\validate-scorecards.py
python .\scripts\summarize-scorecards.py --check .\evaluation\scorecards\summary.json

Or run all current checks:

.\scripts\validate-all.ps1

Development Roadmap

1. CHEFSKILLS-01: Foundation skills, skillsets, docs, and scenario checks.
2. CHEFSKILLS-02: Culinary router specification with task classes, routing ceilings, ambiguity handling, confidence, and a broad routing catalog.
3. CHEFSKILLS-03: Culinary state model for ingredients, dish state, transformations, workflow, observed cues, recovery actions, and safety status.
4. CHEFSKILLS-04: Evaluation engine for rubric scoring, fixtures, regression suites, and safety gates.
5. CHEFSKILLS-05A: Foundation before/after evaluation reports.
6. CHEFSKILLS-05B: Targeted foundation skill improvements based on repeated evaluation gaps.
7. CHEFSKILLS-05C: Specialist expansion has stabilized sauce work, baking structure, protein cookery, fermentation, and equipment cookery before the next pastry, cuisine, service, and costing work.
8. CHEFSKILLS-06: Open source readiness, public alpha launch, GitHub Copilot skill-install docs, and v0.1.0-public-preview distribution.
9. CHEFSKILLS-07: Live-output harness scaffold for reproducible prompt packets, manifests, captured outputs, and rerun review.
10. Future expansion tracks: Michelin / fine-dining intelligence and Canadian commercial food-safety support after current roadmap gates are complete.

See docs/open-source-roadmap.md for the public-release workflow and tracked future expansion ideas.

Contributing

Use the GitHub issue templates for routing bugs, skill proposals, food-safety concerns, and evaluation fixture ideas. Pull requests should run .\scripts\validate-all.ps1 and update scorecards, docs, or release notes when reader expectations change.

License

MIT