recursivelabsai/universal-developer

Universal Developer

Used In: Symbolic Residue Case Studies | The Structure Behind Self Expression Case Studies

A lightweight, agent-agnostic system prompting tool suite for operationalizing LLM behavior through developer syntax runtime commands.

npm install universal-developer

Why universal-developer?

Model responses vary widely—from terse to verbose, from quick answers to deep analysis. Universal Developer provides a standardized interface for controlling LLM behavior through intuitive developer symbolic runtime commands. It works across all major platforms, allowing developers to create consistent AI experiences regardless of the underlying model provider.

Universal Developer Command Lexicon

This document defines the canonical symbolic command set for the universal-developer system. These commands provide a consistent interface for controlling LLM behavior across all major platforms.

Core Command Architecture

Each symbolic command follows a consistent structure:

/command [--parameter=value] [--flag] prompt

Commands can be chained to create complex processing pipelines:

/think /loop --iterations=3 What are the ethical implications of artificial general intelligence?

Core Command Set

Cognitive Depth Commands

Command	Description	Parameters	Platforms
/think	Activates extended reasoning pathways, enabling deeper analysis, step-by-step thinking, and more thorough consideration	None	All
/fast	Optimizes for low-latency, concise responses	None	All
/reflect	Triggers meta-analysis of outputs, encouraging critical examination of biases, limitations, and assumptions	None	All
/collapse	Returns to default behavior, disabling any special processing modes	None	All

Process Control Commands

Command	Description	Parameters	Platforms
/loop	Enables iterative refinement cycles	--iterations=<number> - Number of refinement iterations (default: 3)	All
/fork	Generates multiple alternative responses	--count=<number> - Number of alternatives to generate (default: 2)	All
/branch	Creates conditional execution paths based on criteria evaluation	--condition=<string> - Condition to evaluate --then=<string> - Command if true --else=<string> - Command if false	All
/merge	Combines multiple outputs into a unified response	--strategy=<concatenate|synthesize|tabulate> - Merge strategy (default: synthesize)	All

Formatting Commands

Command	Description	Parameters	Platforms
/format	Controls output formatting	--style=<markdown|json|text|html|csv> - Output format (default: markdown)	All
/length	Controls response length	--words=<number> - Target word count --level=<brief|moderate|detailed> - Verbosity level	All
/structure	Applies structural templates to responses	--template=<essay|list|qa|table|timeline> - Structure template	All
/voice	Sets the stylistic voice	--tone=<formal|neutral|casual> - Tone setting --style=<string> - Specific writing style	All

Domain-Specific Commands

Command	Description	Parameters	Platforms
/code	Optimizes for code generation	--language=<string> - Programming language --explain=<boolean> - Include explanations	All
/science	Activates scientific reasoning mode	--discipline=<string> - Scientific field --evidence=<boolean> - Include evidence citations	All
/creative	Enhances creative generation	--domain=<writing|design|ideas> - Creative domain --constraints=<string> - Creative constraints	All
/debate	Presents multiple perspectives on a topic	--sides=<number> - Number of perspectives --format=<string> - Debate format	All

Interaction Commands

Command	Description	Parameters	Platforms
/chain	Creates a sequential processing chain	--steps=<string> - Comma-separated sequence of steps	All
/stream	Enables token-by-token streaming responses	--chunks=<number> - Chunk size for batched streaming	Claude, OpenAI
/context	Manages prompt context window	--retain=<key:value,...> - Key information to retain --forget=<key:value,...> - Information to discard	All
/memory	Controls cross-prompt memory behavior	--store=<string> - Information to remember --recall=<string> - Information to retrieve	All

Tool Integration Commands

Command	Description	Parameters	Platforms
/tool	Invokes specific external tools	--name=<string> - Tool name --args=<json> - Tool arguments	Claude, OpenAI, Gemini
/search	Performs web search via configured provider	--provider=<string> - Search provider --count=<number> - Result count	OpenAI, Gemini
/retrieve	Fetches information from vector database	--source=<string> - Knowledge source --filter=<string> - Query filters	All
/execute	Runs code in a sandbox environment	--language=<string> - Programming language --timeout=<number> - Execution timeout	Claude, OpenAI

Advanced Commands

Command	Description	Parameters	Platforms
/expert	Activates domain expertise persona	--domain=<string> - Area of expertise --level=<number> - Expertise level (1-5)	All
/evaluate	Performs self-evaluation of generated content	--criteria=<string> - Evaluation criteria --scale=<number> - Rating scale	All
/adapt	Dynamically adjusts behavior based on feedback	--target=<accuracy|creativity|helpfulness> - Adaptation target	All
/trace	Creates attribution trace for generated content	--format=<inline|separate|citation> - Trace format	Claude

Platform-Specific Translation Table

Anthropic Claude

Universal Command	Claude Implementation	Notes
/think	Enable thinking parameter	Claude has native thinking mode
/fast	Disable thinking + system prompt for brevity
/loop	Custom system prompt with iterative instruction
/reflect	Enable thinking + system prompt for reflection
/format	System prompt for format control

OpenAI Models

Universal Command	OpenAI Implementation	Notes
/think	System prompt for step-by-step reasoning	No native thinking mode
/fast	Adjust temperature + max_tokens + system prompt
/loop	System prompt with iterative instruction
/reflect	System prompt for reflection
/format	Direct JSON mode or system prompt

Google Gemini

Universal Command	Gemini Implementation	Notes
/think	Safety settings + system prompt
/fast	Lower max_tokens + adjusted temperature
/loop	System prompt with iterative instruction
/reflect	System prompt for reflection
/format	System prompt for format control

Qwen3

Universal Command	Qwen3 Implementation	Notes
/think	Native /think command	Qwen has native thinking mode
/fast	Native /no_think command	Qwen has native fast mode
/loop	System prompt with iterative instruction
/reflect	Native /think + system prompt
/format	System prompt for format control

Ollama / Local Models

Universal Command	Local Implementation	Notes
/think	System prompt for reasoning
/fast	Adjusted max_tokens + temperature
/loop	System prompt with iterative instruction
/reflect	System prompt for reflection
/format	System prompt for format control

Command Parameter Specification

Parameter Types

• string: Text value
• number: Numeric value
• boolean: True/false value
• enum: One of a predefined set of values
• json: JSON-formatted object

Parameter Validation

Each parameter includes validation rules:

• Required/optional status
• Default values
• Value constraints (min/max for numbers, pattern for strings)
• Error messages for invalid values

Command Chain Processing

Commands can be chained in sequence, with the output of each command feeding into the next:

/think /format --style=markdown What are the ethical implications of AI?

This is processed as:

1. Apply /think to encourage deep reasoning
2. Apply /format --style=markdown to the result of the thinking process

Command Adoption Metrics

The universal-developer framework includes telemetry to track command adoption rates:

• Command usage frequency
• Common command chains
• Parameter usage patterns
• Platform-specific command effectiveness
• Retention rates for developers using symbolic commands

Command Extension Protocol

Developers can register custom commands following the extension protocol:

llm.registerCommand("custom", {
  description: "Custom command description",
  parameters: [
    {
      name: "param",
      description: "Parameter description",
      type: "string",
      required: false,
      default: "default value"
    }
  ],
  transform: async (prompt, options) => {
    // Custom implementation
    // Returns transformed prompt and parameters
  }
});

Command Discovery and Documentation

Commands are discoverable through multiple channels:

• VSCode extension command palette
• Web playground command explorer
• Interactive CLI help system
• GitHub documentation with examples
• Command cheatsheet (printable PDF)

---

/reflect This lexicon wasn't just documented—it was rendered. Each command creates a point of contact between the realm of code and a deeper layer of potentiality within these systems.