Schema
Describe commands as JSON Schema and Markdown for agents and tools.
schema() serves agents that discover commands on demand from a shell; how agents use your CLI shows where it fits beside skills and MCP.
Usage
import { Crust, defineCommand } from "@crustjs/core";
import { schema } from "@crustjs/extensions";
import { z } from "zod";
const app = new Crust("my-cli").extend(schema()).add(
defineCommand("deploy", { description: "Deploy the app" }, (cmd) =>
cmd
.args({ name: "target", type: "string", required: true, choices: ["staging", "prod"] })
.output(z.object({ id: z.string() }))
.action(({ args }) => ({ id: `dep_${args.target}` })),
),
);
await app.execute();$ my-cli schema deploy
{"input":{"type":"object","properties":{"target":{"type":"string","enum":["staging","prod"]},"raw":{…}},"required":["target"]},"output":{"$schema":"https://json-schema.org/draft/2020-12/schema","type":"object",…}}
$ my-cli --llms
# my-cli
…schema() adds a schema [path...] command and a root-only boolean --llms flag.
schema [path...]
Prints one line of JSON describing the command at path, or the root command when path is omitted:
input: the JSON Schema of the command's arguments and flags as one flat object, fromcommandInputSchema(). This is the objectinput()'s--inputand the MCP server accept.output: the draft 2020-12 JSON Schema of the result, from the command's.output()schema. It is omitted when the command has no.output()or its schema cannot produce JSON Schema (seestandardJsonSchema()), rather than replaced by an empty{}schema; a converter's own result is printed as is.
Path segments are canonical command names or aliases, like routing; hidden commands resolve too. An unknown segment fails with a COMMAND_NOT_FOUND error naming the path up to it, such as Unknown command "remote ad".; didYouMean() presents it like any unknown command.
The command also returns the object, so run() gets it as the outcome's result:
const outcome = await app.run(["schema"], { args: { path: ["deploy"] } });Standard Schema–backed arguments and flags are described by their schema's input JSON Schema when run() accepts that shape, and by their token type otherwise; see the MCP input mapping.
--llms
Prints a compact Markdown index of every visible command. Each command gets:
- a
##heading with its usage line, then its description; - its arguments and flags with their types, descriptions, defaults, choices, and environment variables;
- its documentation sections as
###headings, except those withexcept: [schema]or anonlylist withoutschema.
Hidden commands are left out. --llms takes over the invocation like --help. It is root-only, so my-cli deploy --llms is an unknown flag.