Model Code Analyzer

Purpose

Translate validated mathematical modeling plans into executable code thinking.

This skill does not write runnable code itself. It turns the selected modeling route (from the candidate method pool) into a language-neutral coding plan: what each subquestion needs, which data enters each stage, which intermediate artifacts should be saved, how code should be split by subquestion, how the experiments/roundN/ output structure must be organized, and which language-specific generator should implement the plan next.

A critical function of this skill is defining the output structure so that experiment reproducibility, multi-method comparison, and downstream report generation are possible.

When to use

Use this skill:

• After data-auditor-cleaner has confirmed that required data is ready or ready with documented warnings.
• After method-selector has produced the candidate method pool per subquestion.
• Before python-model-code-generator or matlab-model-code-generator.
• When the team needs a code-thinking document under code/ before writing executable scripts.
• When a new round of experiments is being planned and the code structure needs to be defined.

Preconditions

The following should already exist or be provided:

• Candidate method pool per subquestion (methods/Qx/qx_method_candidates.md).
• An implementation.target field with value python or matlab.
• Cleaned data under workspace/data/data_clean/, if data is required.
• A data report at workspace/data/data_report.md, if data is used.
• The round number for this experiment batch (round1, round2, ..., final).

Inputs

Use or request:

• methods/Qx/qx_method_candidates.md — candidate method pool.
• workspace/data/data_report.md, if available.
• Cleaned data under workspace/data/data_clean/.
• Implementation target and runtime notes.
• Expected result outputs and figure outputs.
• Round number (N for roundN).

Workflow

1. Read the candidate method pool.

• Work subquestion by subquestion.
• Identify which methods are prioritized for this round.
• Preserve baseline and candidate method roles.
• Do not change the method list.

1. Translate the mathematical plan into code logic per subquestion.

• Identify required inputs for each candidate method.
• Identify calculation order, intermediate values, and saved artifacts.
• Identify which outputs belong under results/Qx/experiments/roundN/.
• For each method, plan the computation pipeline.

1. Define the experiments/roundN/ output structure.

• Every round of experiments MUST produce outputs in this structure:

`` results/Qx/experiments/roundN/ ├── figures/ # All generated figures (.png, .pdf) │ ├── m1_result.png │ ├── m2_result.png │ └── comparison.png ├── tables/ # All result tables (.csv) │ ├── m1_scores.csv │ ├── m2_scores.csv │ └── comparison_table.csv ├── metrics/ # All evaluation metrics (.csv, .json) │ ├── m1_metrics.json │ ├── m2_metrics.json │ └── comparison_metrics.json ├── logs/ # Execution logs (.log, .txt) │ ├── m1_run.log │ └── m2_run.log └── run_summary.json # Execution record (see specification below) ``

• Ensure this structure is documented and expected by downstream skills.

1. Specify the run_summary.json.

• This is the machine-readable record of what was executed.
• Downstream skills (result-report-generator, robustness-checker) rely on it.
• Required fields:

``json { "question": "Q1", "round": "round1", "execution_timestamp": "2026-05-14T10:30:00+08:00", "implementation_target": "python", "random_seed": 2026, "methods": [ { "method_id": "M1", "method_name": "equal_weight_scoring", "script": "code/Q1/q1_baseline.py", "status": "success", "execution_time_seconds": 12.5, "input_files": [ "workspace/data/data_clean/indicator_data.csv" ], "output_files": [ "results/Q1/experiments/round1/tables/m1_scores.csv", "results/Q1/experiments/round1/metrics/m1_metrics.json" ], "figure_files": [ "results/Q1/experiments/round1/figures/m1_score_bar.png" ], "metrics_summary": { "score_range": [0.45, 0.92], "ranking_top3": ["City_A", "City_B", "City_C"] }, "errors": [], "warnings": [] }, { "method_id": "M2", "method_name": "entropy_topsis", "script": "code/Q1/q1_entropy_topsis.py", "status": "success", "execution_time_seconds": 18.3, "input_files": [ "workspace/data/data_clean/indicator_data.csv" ], "output_files": [ "results/Q1/experiments/round1/tables/m2_scores.csv", "results/Q1/experiments/round1/tables/m2_weights.csv", "results/Q1/experiments/round1/metrics/m2_metrics.json" ], "figure_files": [ "results/Q1/experiments/round1/figures/m2_weight_bar.png", "results/Q1/experiments/round1/figures/m2_score_dist.png" ], "metrics_summary": { "entropy_weights": {"indicator_1": 0.23, "indicator_2": 0.18, "indicator_3": 0.31, "indicator_4": 0.28}, "closeness_scores_range": [0.12, 0.88], "ranking_top3": ["City_A", "City_C", "City_B"] }, "errors": [], "warnings": ["Indicator_4 has very high variance and dominates entropy weight."] } ], "comparison": { "ranking_correlation": 0.87, "top3_overlap": "2 out of 3 identical" }, "data_inputs": [ "workspace/data/data_clean/indicator_data.csv" ], "environment": { "python_version": "3.10", "key_dependencies": ["numpy==1.24", "pandas==1.5", "matplotlib==3.6"] } } ``

1. Define the code layout per subquestion.

• For Python: code/Q1/, code/Q2/, etc.
• For MATLAB: code/matlab/Q1/, code/matlab/Q2/, etc.
• One script per candidate method within each subquestion folder.
• A README.md in each code subfolder documenting run order, inputs, outputs, and dependencies.
• A run_all.py or run_all.m script only if batch execution is useful.

1. Write the code-thinking document.

• Save the analysis under code/model-code-analyzer.md.
• Keep the document language-neutral.
• Make the handoff usable by the language-specific code generator.

1. Hand off to the language-specific generator.

• Route to python-model-code-generator when implementation.target = python.
• Route to matlab-model-code-generator when implementation.target = matlab.

Outputs

Produce one Markdown planning artifact:

• code/model-code-analyzer.md

The document should include:

• Implementation target and runtime notes.
• Round number and purpose.
• Per-subquestion code logic for each candidate method.
• Expected code folder layout.
• Expected experiments/roundN/ output structure (explicit directory tree).
• run_summary.json field specification.
• Expected input data per method.
• Expected result artifacts per method.
• Expected figure artifacts per method.
• Reproducibility requirements (fixed seeds, environment notes).
• Recommended next skill.

Output format

Write a concise Markdown document under code/model-code-analyzer.md.

The document should explain, for each subquestion:

# Model Code Analysis

## Implementation Target
- **Language**: python / matlab
- **Round**: roundN (purpose: e.g., "first multi-method comparison")
- **Runtime notes**: [constraints like minimal-dependencies, portable-artifacts, etc.]

## Qx: [Subquestion Summary]

### Methods to Implement This Round
- M1: [name] (baseline / main candidate)
- M2: [name] (main candidate / backup)
- M3: [name] (if applicable)

### Per-Method Code Logic

#### M1: [name]
1. **Input**: read `data_clean/[file]`, columns [list]
2. **Processing**: [step-by-step computation logic]
3. **Output**: save results to `results/Qx/experiments/roundN/tables/m1_results.csv`
4. **Figures**: save to `results/Qx/experiments/roundN/figures/m1_*.png`
5. **Metrics**: save to `results/Qx/experiments/roundN/metrics/m1_metrics.json`
6. **Log**: write execution log to `results/Qx/experiments/roundN/logs/m1_run.log`

#### M2: [name]
[same structure]

### Expected Code Files

code/Qx/ ├── README.md # Run instructions ├── qxbaseline.py # M1 implementation ├── qxmain.py # M2 implementation └── qx_improved.py # M3 implementation (if applicable)

### Expected Output Structure

results/Qx/experiments/roundN/ ├── figures/ │ ├── m1scorebar.png │ ├── m2weightbar.png │ ├── m2scoredist.png │ └── rankingcomparison.png ├── tables/ │ ├── m1scores.csv │ ├── m2scores.csv │ ├── m2weights.csv │ └── comparisontable.csv ├── metrics/ │ ├── m1metrics.json │ ├── m2metrics.json │ └── comparisonmetrics.json ├── logs/ │ ├── m1run.log │ └── m2run.log └── run_summary.json

### run_summary.json Key Fields
[Reference the specification above]

### Reproducibility
- **Random seed**: 2026 (fixed)
- **Environment**: Python 3.10, numpy, pandas, matplotlib, scipy
- **Expected runtime**: ~2 minutes for all methods

Rules

• Do not write executable Python or MATLAB code in this skill.
• Do not change the candidate method pool or method specifications.
• Do not fabricate data, scripts, results, or figures.
• Do not bypass language-specific code generators.
• Always specify the full experiments/roundN/ output structure.
• Always specify the run_summary.json schema.
• Keep the code plan minimal but complete enough for the code generator to execute.
• Ensure every method has clearly defined inputs, processing steps, and output destinations.
• Separate baseline outputs from main method outputs in the plan.
• Include reproducibility requirements: random seed, environment, expected runtime.

Verification

Before handing off, verify:

• Every method from the candidate pool for this round has a code logic plan.
• The experiments/roundN/ output structure is explicitly specified.
• The run_summary.json schema is referenced or specified.
• Input data paths are correct.
• Output paths follow the results/Qx/experiments/roundN/ convention.
• Code folder layout follows code/Qx/ (Python) or code/matlab/Qx/ (MATLAB) convention.
• Reproducibility requirements are stated.
• The next skill is the correct language-specific generator.

Failure modes

Stop and report a blocker if:

• The candidate method pool is missing.
• implementation.target is undefined.
• Cleaned data is required but unavailable.
• The methods cannot be mapped to available data fields.
• The expected outputs are undefined.

Stop conditions

This skill must stop instead of guessing when:

• Important parameters are unspecified and cannot be safely defaulted.
• The method requires a dependency that may be unavailable.
• The code would need to overwrite raw data.
• A script would produce results that cannot be traced to the candidate method.
• Continuing would change the mathematical meaning of the model.

When stopping, output:

• blocker
• why it matters
• affected subquestion or model
• missing information or artifact
• safe partial plan if any
• recommended next skill

Handoff

After writing code/model-code-analyzer.md, hand off to:

• python-model-code-generator if implementation.target = python
• matlab-model-code-generator if implementation.target = matlab

The handoff must include:

• code/model-code-analyzer.md (the full code thinking document)
• Candidate method pool files
• Cleaned data paths
• Round number
• experiments/roundN/ output structure specification
• run_summary.json schema
• Reproducibility requirements

Examples

Example 1: Python evaluation model code analysis for round 1

Input state:

• Q1 candidate pool: M1 (equal-weight baseline), M2 (entropy-TOPSIS).
• Implementation target: python.
• Round: round1 (first multi-method comparison).
• Data: cleaned indicator matrix.

Output (code/model-code-analyzer.md excerpt):

## Q1: City Resilience Evaluation

### Methods to Implement This Round
- M1: equal_weight_scoring (baseline)
- M2: entropy_topsis (main candidate)

### Expected Output Structure

results/Q1/experiments/round1/ ├── figures/ │ ├── m1scorebar.png │ ├── m2weightbar.png │ ├── m2scoredist.png │ └── rankingcomparison.png ├── tables/ │ ├── m1scores.csv │ ├── m2scores.csv │ └── m2weights.csv ├── metrics/ │ ├── m1metrics.json │ ├── m2metrics.json │ └── comparisonmetrics.json ├── logs/ │ ├── m1run.log │ └── m2run.log └── runsummary.json

### run_summary.json
[Full specification with M1 and M2 method entries, comparison field, environment info]

Example 2: MATLAB optimization model code analysis

Input state:

• Q3 candidate pool: M1 (greedy baseline), M2 (integer linear programming).
• Implementation target: matlab.
• Runtime notes: beita-tianyuan-compatible.

Output: same structure as above, but code under code/matlab/Q3/, and MATLAB-specific output path conventions.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: zhnnky329
• Source: zhnnky329/MathModeling-skills
• License: MIT

Install and usage instructions live in the source repository linked above.