ChilliCream Agent Skills

Agent skills for the ChilliCream ecosystem - HotChocolate, Nitro, Green Donut, Mocha, and friends.

Skills follow the Agent Skills open standard and work with Claude Code, the Claude API, and any other compatible agent runtime.

What lives here

Each skill is a folder under skills/ containing a SKILL.md. The frontmatter says what the skill does and when an agent should fire it; the body teaches the agent how to perform the task — e.g. authoring a HotChocolate resolver, wiring a Green Donut DataLoader, extending Banana Cake Pop, or implementing a Mocha transport.

skills/
  <skill-name>/
    SKILL.md
    scripts/        # optional
    references/     # optional
    assets/         # optional

Available skills

prototype-feature

Build a clickable local-only frontend prototype before backend or schema work exists. The skill keeps the prototype aligned with the existing app, uses realistic product-domain mock data, preserves one-entity component boundaries, and models user actions as mutation-shaped local hooks.

Fire it with phrases like "prototype this feature", "mock up this page", "wireframe this view", "just build the UI", "no backend yet", or "no GraphQL yet".

prototype-to-contract

Convert a local-only prototype into GraphQL-client-backed component fragments and a backend contract. The skill replaces rendered mock fields with colocated fragments, turns local action hooks into schema mutations, preserves data-masked component boundaries, and writes schema contract SDL for the backend team.

Fire it with phrases like "prototype to contract", "mock to contract", "turn this prototype into a contract", "fragmentize this page", "wire this mocked page to GraphQL", or "spec this feature for the backend".

graphql-schema-design

Design and review GraphQL schema changes. The skill acts as a senior API architect that produces SDL proposals and design feedback — it does not write implementation code. It runs in two modes:

• Design mode (default) — given a feature, use case, or need, it walks through an iterative process to produce a thoroughly vetted SDL proposal you explicitly approve, covering types, mutations, queries, connections, enums, and error handling.
• Review mode — given a schema diff, it audits the changes for naming, nullability, evolvability, and client-friendliness.

The rules are framework-agnostic GraphQL design conventions, illustrated with a Book/Author domain. Fire it with phrases like "design a mutation", "new type", "review schema diff", or /graphql-schema-design.

hotchocolate-best-practices

Implementation best practices for HotChocolate 16.6+ servers — the implementation-side sibling of graphql-schema-design. The skill teaches the implementation-first style (attributes + source generator) and enforces the core rules: all data access flows through source-generated DataLoaders (relations and root byId fields), projections travel via QueryContext<T> with explicitly pinned keys, cursor pagination with a key-tailed order and no offset paging, restrictive opt-in filtering and sorting, mutation conventions, and explicit module registration.

Detailed rules live in on-demand references — DataLoaders, pagination, resolvers & type extensions, filtering, sorting, mutations, server setup, and subgraph-specific practices (lookups, gateway-enforced cost limits) that load only when the project is a source schema in a composite setup.

Fire it when writing or reviewing C# in any HotChocolate project — resolvers, [QueryType]/[ObjectType<T>]/[DataLoader] classes, paging, or GraphQL performance work in .NET.

nitro-mail

Operate Nitro's local-first agent mailbox. Covers registering agent identities, sending and replying to mail, managing inboxes and threads, and waiting for new coordination messages.

nitro-task

Manage Nitro's local-first, dependency-aware agent task tracker. Covers task lifecycle, dependencies, comments, JSONL synchronization, and non-interactive agent workflows.

nitro-task-orchestrator

Coordinate a large Nitro task backlog through implementation and review waves. Use it for autonomous multi-task delivery, not individual small changes.

nitro-task-planner

Turn feature briefs and feedback into an implementation-ready Nitro task graph, then hand it to the orchestration workflow.

nitro-wayfinder

Plan a large, initially fuzzy feature with Nitro tasks as persistent decision memory. It guides charting, research, prototyping, and gradual conversion to implementation tickets.

Installing

Claude Code (as a plugin)

/plugin marketplace add ChilliCream/agent-skills
/plugin install chillicream-skills@chillicream-agent-skills

skills CLI (via npx)

npx skills add ChilliCream/agent-skills

This pulls the skills into your agent's skills directory. Works with any Agent Skills-compatible runtime.

Anywhere else

Copy skills/<name>/ into whatever directory your agent loads skills from. The format is portable across compliant clients.

Authoring a new skill

The fastest path: open this folder in Claude Code and say "I want to write a new skill for X." The bundled local skill-author skill will auto-load and walk you through the conventions.

Otherwise, read AGENTS.md for the rules and the checklist.