⚠️ mlld is pre-release

Join the discord if you have questions. Report bugs on GitHub.

Prose Execution

mlld supports prose execution - the ability to define executable functions that invoke a prose interpreter (like OpenProse) via an LLM.

Prerequisites

Prose execution requires:

Claude Code - The model executor uses claude -p to run prompts
OpenProse plugin - Install with /plugin install open-prose@prose in Claude Code
Skill approval - The first time you run prose, Claude Code will prompt you to approve the OpenProse skills

Syntax

exe @function(params) = prose:@config { inline prose content }
exe @function(params) = prose:@config "path/to/file.prose"
exe @function(params) = prose:@config template "path/to/file.prose.att"

Quick Start

Use pre-built configs from @mlld/prose:

import { @opus } from @mlld/prose

exe @workflow(ctx) = prose:@opus {
  session "Process @ctx"
}

run @workflow("some context")

OpenProse requires Opus - it's the only model that can reliably interpret the prose syntax.

Configuration

The config object specifies how prose is executed:

import { @claude } from @mlld/claude

var @config = {
  model: @claude("prompt", "opus", @base),
  skills: ["open-prose:prose-boot", "open-prose:prose-compile", "open-prose:prose-run"]
}

Config Options

Option	Type	Default	Description
`model`	executable	required	Model executor (e.g., from `@mlld/claude`)
`skills`	array	OpenProse skills	Skills to invoke for prose execution
`skillPrompt`	string	auto-generated	Custom skill injection prompt
`skillPromptEnd`	string	auto-generated	Custom prompt suffix

Default Skills

When skills is not specified, mlld defaults to the OpenProse plugin skills:

open-prose:prose-boot
open-prose:prose-compile
open-prose:prose-run

Custom Interpreters

You can use any interpreter by specifying different skills:

import { @opus } from @mlld/claude

var @myInterpreter = {
  model: @opus,
  skills: ["my-dsl:boot", "my-dsl:run"]
}

exe @analyze(data) = prose:@myInterpreter {
  process @data
  output result
}

Content Sources

Inline Content

Variables are interpolated using @var syntax:

exe @summarize(text) = prose:@config {
  session "Summarize the following"
  input @text
  output summary
}

File Reference

Plain .prose files are loaded and interpolated:

exe @analyze(data) = prose:@config "./analysis.prose"

Template Files

Use .prose.att (ATT-style @var) or .prose.mtt (Mustache-style {{var}}):

import { @opus } from @mlld/prose

exe @reviewAtt(code) = prose:@opus "./review.prose.att"
exe @reviewMtt(code) = prose:@opus "./review.prose.mtt"

review.prose.att:

session "Code Review"
input @code
output review

review.prose.mtt:

session "Code Review"
input {{code}}
output review

Execution

Prose executables are invoked like any other mlld executable:

import { @opus } from @mlld/prose

exe @summarize(text) = prose:@opus { session "Summarize: @text" }

run @summarize("Long text to summarize...")

Prose executables can also be used in pipelines:

import { @opus } from @mlld/prose

exe @summarize(text) = prose:@opus { session "Summarize: @text" }
var @document = "This is a long document..."

show @document | @summarize()

Error Handling

If the OpenProse skills are not available (not installed or not approved), mlld throws:

Prose execution failed: OpenProse skills not available.
Skills must be installed AND approved.
Required skills: open-prose:prose-boot, open-prose:prose-compile, open-prose:prose-run

To fix:

Install the plugin: /plugin install open-prose@prose in Claude Code
Run a prose script - Claude Code will prompt for skill approval
Approve the skills when prompted