turbo_promptable #

Object-Oriented Prompting framework for the turbo ecosystem. Define AI agent prompts, roles, workflows, and tools as type-safe Dart objects that serialize to JSON, YAML, Markdown, and XML.

Features #

• Type-safe workspace models: TRole, TPersona, TWorkflow, TStep, TInstruction, TInput, TOutput, TGoal, TEndGoal, TIssue, TContext, TTemplate, TTool, and more
• Spec models: TAbility, TFeature, TRequirement, TScenario, TJourney, TTask, TModule, TMockup, TPrototype
• Tool models: TApi, TCli, TScript, TMcp, and TToolSet, each carrying a list of TToolAbility
• Spawnable abstraction (TSpawnable, extended by TAgent) carrying id, allowedTools, yolo, model, and headless, with a spawn method that builds a launch command via a TCliTool for tools like Claude Code, Cursor, Windsurf, and custom CLIs
• Cross-referencing abstracts (TOfAbilities, TOfFeatures, TOfIssues, TOfJourneys, TOfMockups, TOfModules, TOfPrds, TOfProjects, TOfPrototypes, TOfScenarios) for composing specs
• JSON serialization on every model via json_serializable; YAML, Markdown, and XML output inherited from turbo_serializable
• Structured metadata (TMetaData) on every promptable

Installation #

dependencies:
  turbo_promptable: ^0.6.0

Usage #

import 'package:turbo_promptable/turbo_promptable.dart';

const workflow = TWorkflow(
  name: 'Review Workflow',
  endGoal: TEndGoal(
    'A reviewed, higher-quality codebase',
    name: 'Quality Review',
  ),
  steps: [
    TStep(
      name: 'Analyse',
      input: TInput(name: 'Source Code'),
      instructions: 'Analyse the provided source code for quality issues.',
      output: TOutput(
        name: 'Analysis Report',
        schema: 'markdown',
      ),
    ),
  ],
);

const persona = TPersona(
  name: 'Code Reviewer',
  expertise: 'Static analysis and code quality',
  identity: 'A meticulous reviewer focused on maintainability.',
);

void main() {
  print(workflow.toJson());
  print(persona.toJson());
}

Core Concepts #

TPromptable #

Every workspace model extends TPromptable, which itself extends TSerializable from turbo_serializable. Models have:

• name — required identifier
• metaData — optional TMetaData for frontmatter (description, tags, etc.) on models that declare it

Roles and Personas #

• TRole — a capability bundle with required expertise, plus optional instructions and tools
• TPersona — a TRole augmented with an identity string that describes the persona's character; construct via TPersona(...) or TPersona.fromRole(role: ..., identity: ...)

Workflows and Steps #

A TWorkflow contains an ordered list of TSteps and a required TEndGoal. Each TStep has a required TInput, a string instructions field, and a required TOutput.

Specs #

Spec models (TAbility, TFeature, TRequirement, TScenario, TJourney, TTask, TModule, TMockup, TPrototype) describe intended behaviour and deliverables. They cross-reference each other via the TOf* abstracts exported from workspace/abstracts/.

Tools #

Tool subclasses (TApi, TCli, TScript, TMcp) extend the shared TTool base, which carries a list of TToolAbilitys. TTool.asToolSet(...) converts a tool into a TToolSet.

Spawnable #

TSpawnable (extended by TAgent) carries id, allowedTools, yolo, model, and headless, and exposes a spawn method that builds a launch command via a TCliTool for orchestrating agent launches across tools like Claude Code, Cursor, Windsurf, and custom CLIs.

License #

MIT