Builder
Select a class, fill out the form, and download valid JSON ready to PR intoai-atoms. New classes appear here automatically as the catalog grows.
A bindable identity — one persona bound to the prompts, skills, tools, policies, and hooks it runs with, plus execution preferences.
Prefer to start from a file? Every class has a minimal valid example at /schemas/examples/<class>.json.
Slug * kebab-case, becomes the file name
Name *
Description * what it does and when to invoke it
System prompt fragment *
Must be self-contained — the agent may load this skill without any other context.Applicable domains * comma-separated
Invocation commands slash commands that trigger this skill
First entry is the primary command. Additional entries are subcommands shown indented below it.
Tags comma-separated
Category shared vocabulary across every class
Lifecycle
Slug * kebab-case
Name *
Description *
Event *
Trigger type *
Language *
Blocking behavior
Side effects comma-separated
Supported platforms uncheck if not verified
Tags comma-separated
Category shared vocabulary across every class
Lifecycle
Slug * kebab-case
Name *
Description * what it does and when to inject it
Subtype *
Content * the injected text, verbatim
Applicable turns
Vendors comma-separated; "any" for vendor-neutral
Tags comma-separated
Category shared vocabulary across every class
Lifecycle
Slug * kebab-case
Name *
Description * who this is and when to activate it
Job to be done * the outcome this persona owns
Primary tasks one per line
Out of scope one per line
Domain slug, e.g. software-engineering
Voice and tone
Constraints one per line, as the model should read them
Tags comma-separated
Category shared vocabulary across every class
Lifecycle
Slug * kebab-case
Name *
Description * what it does, what it may touch, what it must not
Subtype *
Persona *
Bindings comma-separated atom ids; every id must exist in the catalog
Execution
Tags comma-separated
Category shared vocabulary across every class
Lifecycle
Slug * model name as the provider spells it, e.g. llama3.2
Name *
Description *
Vendor * who trained it
Family e.g. Llama 3
Task *
Capabilities comma-separated: tools, vision, thinking, embedding
Parameter sizes comma-separated: 1b, 3b, 8x7b
Provider *
Weights license SPDX id or the vendor's name for it
Tags comma-separated
Category shared vocabulary across every class
Lifecycle
Generated JSON
Copied!
Save as atom.json inatoms/skill/ and open a PR →convergent-systems-co/ai-atoms