repo-reproduction-skill

name	repo-reproduction-skill
description	This skill should be used when the user asks to analyze, reproduce, or run a computational math repository. Use when inspecting source code, detecting algorithms, planning runs, executing experiments, or collecting results from optimization code.

Repo Reproduction Skill

Analyzes a computational math repository, generates run plans, executes them, and collects results.

For end-to-end computational math research-code reproduction workflows, this Skill should be selected by computational_math_reproduction_workflow_skill rather than used as the first entrypoint.

Scripts

repo_analyzer

Analyzes source code directory. Input: --source <path>, --out <path>. Writes analysis summary to plan.md fields (source path, language, entrypoints, detected algorithms, risks) — does not write repo_analysis.json by default.

python -m skills.repo_reproduction_skill.scripts.repo_analyzer --source /path/to/repo --out /path/to/output

run_planner

Generates candidate run plans from the analysis. Input: --analysis <dict or json>, --out <path>. Outputs compact plan fields for plan.md — does not write run_plan.json by default.

python -m skills.repo_reproduction_skill.scripts.run_planner --analysis /path/to/analysis.json --out /path/to/output

executor

Executes an approved run plan. Input: --command <str>, --out <path>, --timeout <int>. Writes stdout/stderr to logs/run.log and returns exit code. Does not write run_plan.json or execution_log.jsonl by default.

python -m skills.repo_reproduction_skill.scripts.executor --command "python main.py" --out /path/to/output --timeout 300

result_collector

Scans executed output for result files and extracts metrics. Input: --source, --out. Writes results to results/ — does not write collected_results.json by default.

python -m skills.repo_reproduction_skill.scripts.result_collector --source /path/to/repo --out /path/to/output

Output Rules

• Do not write repo_analysis.json by default. Summarize analysis in conversation and in plan.md fields.
• Do not write run_plan.json by default. Use compact plan fields.
• Execution logs go to logs/run.log.
• Results go to results/.
• Figures go to figures/.
• Patches go to patches/.

Workflow

1. Run repo_analyzer to understand the repository.
2. Summarize findings in conversation and in plan.md fields.
3. Generate candidate command(s) and present to user for approval.
4. Run executor with the approved command.
5. Run result_collector to gather results under results/.