Skip to main content

​
Overview

The squash command is the primary interface for Splat’s AI-powered debugging. It executes your code, captures any errors that occur, and provides intelligent fixes based on comprehensive context analysis.

​
Basic Usage

splat squash "<command>"
Run your application exactly as you normally would, but prefix it with splat squash: 
splat squash "python3 main.py"

splat squash "npm start"

splat squash "python3 foo.py"

​
How It Works

When you run splat squash, the tool:
  1. Executes your command in a controlled subprocess environment
  2. Captures all output including stdout, stderr, and return codes
  3. Parses error tracebacks to identify affected files and line numbers
  4. Gathers context from your codebase (respecting .gitignore)
  5. Analyzes with AI using Groq’s fast inference models
  6. Provides actionable fixes with specific code suggestions

​
Example Output

When Splat detects an error, it provides structured analysis: 
$ splat squash "python3 test.py"

  Running: python3 test.py
  
  Error detected in test.py:2
  
  WHERE:
    File: test.py
    Line: 2
    
  WHAT:
    Error Type: SyntaxError
    Description: Unclosed parenthesis in print statement
    
  HOW:
    Suggested Fix:
    print("hello")

​
Error Analysis

Splat’s AI engine analyzes errors and returns:
  • WHERE: Repository path, filename, and exact line number
  • WHAT: Error type and clear explanation
  • HOW: Line where error originated and suggested code solution

​
Response Format

The AI response is structured as JSON internally: 
{
  "where": {
    "repository_path": "/absolute/path/to/repo",
    "file_name": "test.py",
    "line_number": "2"
  },
  "what": {
    "error_type": "SyntaxError",
    "description": "'(' was never closed in print statement"
  },
  "how": {
    "error_origination": "2",
    "suggested_code_solution": "print(\"hello\")"
  }
}

​
Context Gathering

Splat automatically:
  • Parses error stack traces to identify relevant files
  • Respects your .gitignore to avoid sensitive data
  • Uses repomix to package file context efficiently
  • Sends only relevant code to the AI model

​
Performance

Splat is optimized for speed:
  • Uses Groq (llama3-70b-8192) for near-instant inference
  • Captures output in real-time with threading
  • Minimizes context sent to AI while maximizing relevance

​
Use Cases

​
Quick Bug Fixes

splat squash "python3 buggy_script.py"
Get instant explanations and fixes for syntax errors, runtime errors, and exceptions.

​
Development Workflow

splat squash "python3 -m myapp.main"
Run your module entrypoint and debug any startup errors.

​
Testing

splat squash "pytest tests/"
Analyze test failures with full context.

​
Limitations

Splat currently works best with scripts launched from the root directory. Sub-module entrypoints may not provide full context.
Suggested code fixes do not currently conform to your project’s formatting configuration (e.g., Black, Prettier).

​
Next Steps

Build docs developers (and LLMs) love

Get started for freeTalk to us