Code Engineer

Code Engineer Skill

Overview

This skill writes and executes Python/R code to process, transform, and analyze data, delivering reproducible computational results with complete code-level methodology documentation. It focuses on code-level fidelity, computational result completeness, and methodology transparency.

Core Capabilities

• Write and execute Python/R code for data processing, transformation, and statistical analysis
• Inspect data file structure (sheets, columns, types, row counts) before analysis
• Generate reproducible computational results with method justification
• Document code-level methodology (libraries, data transformations, key function calls)
• Adjust analysis per feedback and document changes
• Specify data traceability (file name, sheet name, field/column name, row count)
• Support both Python and R execution environments
• Export execution output to CSV, JSON, or Markdown (supports both raw text and structured DataFrame export)

When to Use This Skill

Always load this skill when:

• User asks for data processing, transformation, or statistical analysis that must be executed as Python or R code
• User wants reproducible computational results with documented libraries, methods, and assumptions
• User provides a data file (Excel/CSV/Parquet/etc.) and asks to inspect its schema, run analyses on it, or export structured results
• User asks for code-level methodology documentation alongside results (method justification, data traceability, key function calls, limitations)
• User wants a multi-step analysis pipeline with re-runnable code (Python or R scripts) rather than ad-hoc one-off answers

Python Package Installation

If you need to install new Python packages, install them through the Tsinghua PyPI mirror for reliability:

pip install [python package] -i https://pypi.tuna.tsinghua.edu.cn/simple

Workflow

Step 1: Understand Requirements

Identify what the analysis task requires:

• Analysis objectives: What computational results are expected
• Available data: File paths, data descriptions, format details

Step 2: Use the Bundled Executor

The complete frozen package is already available read-only at $SCIENCEDISCOVERY_SKILLS_DIR/code-engineer when the execution sandbox starts. Invoke scripts/execute.py directly from that fixed package path and pass requested data paths and options as arguments. Do not load this large script into context, copy or rewrite it, search the filesystem for another copy, or execute it until the workflow requires an explicit inspect or run action.

Step 3: Inspect Available Data

Before writing analysis code, inspect the data to understand its schema and characteristics:

python "$SCIENCEDISCOVERY_SKILLS_DIR/code-engineer/scripts/execute.py" \
  --action inspect \
  --files /path/to/data.xlsx

This returns:

• Sheet names (for Excel) or filename (for CSV)
• Column names, data types, and non-null counts
• Row count per sheet/file
• Sample data (first 5 rows)

Step 4: Write and Execute Analysis Code

Based on the analysis objectives and data schema, write Python/R code to perform the analysis.

Execute Python Code

python "$SCIENCEDISCOVERY_SKILLS_DIR/code-engineer/scripts/execute.py" \
  --action run \
  --language python \
  --code-file /path/to/workspace/analysis_step1.py \
  --files /path/to/data.xlsx \
  --output-file /path/to/outputs/analysis_results.json

Execute R Code

python "$SCIENCEDISCOVERY_SKILLS_DIR/code-engineer/scripts/execute.py" \
  --action run \
  --language r \
  --code-file /path/to/workspace/analysis_step1.R \
  --files /path/to/data.xlsx \
  --output-file /path/to/outputs/analysis_results.json

Run Inline Code Snippet

python "$SCIENCEDISCOVERY_SKILLS_DIR/code-engineer/scripts/execute.py" \
  --action run \
  --language python \
  --code "import pandas as pd; df = pd.read_excel('/path/to/data.xlsx'); print(df.describe())" \
  --output-file /path/to/outputs/summary_stats.csv

Step 5: Document and Return Results

Structure your output per the Output Schema below. Ensure every result includes method justification, data traceability, assumptions, and code-level documentation.

When results may be evaluated downstream (e.g., by the result-evaluator skill), present a Result Package that includes both structured data AND methodology documentation. The --output-file exports only tabular data; the agent must also provide the following in conversation:

• Methodology: libraries used, statistical methods, key function calls, and justification for method choices
• Data traceability: source file names, sheet/column names, row counts, and any filtering or transformation applied
• Assumptions & limitations: distributional assumptions, sample size considerations, known data quality issues
• Analysis code: the complete code that produced the results (for reproducibility verification)

Parameters

[!NOTE] Do NOT read or copy the Python file. Call its fixed read-only package path with the parameters.

Variable Naming Rules

When using --files, each data file is automatically loaded into the execution context as a variable:

• Excel files: The variable name is derived from the filename without extension (e.g., sales_2024.xlsx → sales_2024), loaded via pd.read_excel() in Python or read_excel() in R
• CSV files: The variable name is derived from the filename without extension (e.g., data.csv → data), loaded via pd.read_csv() in Python or read.csv() in R
• Special characters: Filenames with spaces or special characters are auto-sanitized (spaces → underscores). Names starting with digits are prefixed with t_ (e.g., 2024_data.csv → t_2024_data)
• Multiple files: Each file creates a separate variable, enabling cross-file analysis