code-interpreter

Code Interpreter: Analyze, Generate, Execute in Sandbox, Interpret

A data analysis question arrives. The agent analyzes the question to identify needed columns (["id", "value"]), generates Python code, executes it in a sandboxed environment (logging code.split("\n").length lines), and interprets the result with a confidence score.

Workflow

question, dataset -> ci_analyze_question -> ci_generate_code -> ci_execute_sandbox -> ci_interpret_result

Workers

AnalyzeQuestionWorker (ci_analyze_question) -- Identifies required columns and produces an analysis map.

GenerateCodeWorker (ci_generate_code) -- Generates Python code, counting linesOfCode via code.split("\n").length.

ExecuteSandboxWorker (ci_execute_sandbox) -- Executes code in a sandboxed environment. Logs the line count.

InterpretResultWorker (ci_interpret_result) -- Interprets execution results with a confidence score.

Tests

36 tests cover question analysis, code generation, sandbox execution, and interpretation.

Further Reading

• RUNNING.md -- how to build and run this example