Cognotik | CodingMode

⚙️ OrchestrationConfig.json


{
  "mode": "CodingMode",
  "config": {
    "codeRuntime": "GroovyRuntime"
  },
  "auto_fix": true,
  "model_config": {
    "smart": "gpt-4-turbo"
  }
}

→

🧠 Live Transcript

// Assistant generating Groovy script...


def result = FileModificationTask.call([
  target_file: "src/App.kt",
  instructions: "Add logging"
], "Updating main loop")

println "Task Result: " + result

Output:
Task Result: ✔ File Updated Successfully

The Execution Loop

Unlike static planning modes, CodingMode operates by generating and executing code in real-time. This allows the agent to use standard programming constructs (loops, conditionals, try-catch) to handle task results.

Phase	Description
Planning	The `CodeAgent` analyzes the user prompt and history to generate a script.
Injection	Cognotik tasks are mapped to functions and injected into the runtime environment.
Execution	The script runs via `CodeRuntimes` (e.g., Groovy), capturing stdout and return values.
Reflection	Results are added to the conversation history, allowing the agent to iterate if errors occur.

Runtime Environment

The following symbols are automatically available within the generated scripts:

[TaskName] - Every configured task is available as a callable function.
workingDir - The absolute path to the current workspace.
smartModel / fastModel - Direct access to LLM clients for sub-reasoning.
task - The current SessionTask object for UI manipulation.

Kotlin Implementation Skeleton

The core logic from CodingMode.kt handles the bridge between the LLM and the local runtime.


// From CodingMode.kt
override fun handleUserMessage(userMessage: String, task: SessionTask) {
    val response = plan(task) // Generates CodeAgent.CodeResult
    
    val tabs = TabbedDisplay(task)
    tabs["Code"] = renderMarkdown("```groovy\n${response.code}\n```")
    
    // Execute code and capture output/value
    val executionResult = response.result 
    output(executionResult, tabs, transcript, response)
}

Integration Guide

To use CodingMode in your application, define it in your OrchestrationConfig. This mode is ideal for complex engineering tasks where the agent needs to verify its own work.


val config = OrchestrationConfig(
    mode = CognitiveModeType.Coding,
    config = CodingMode.CodingModeConfig(
        codeRuntime = CodeRuntimes.GroovyRuntime
    ),
    autoFix = true // Allows the agent to loop on execution errors
)