run-solver

name	run-solver
description	运行数学模型求解器。在本地 Python/MATLAB 环境执行求解代码，收集结果，超时保护。当用户说'运行求解'、'run solver'、'跑代码'时使用。
argument-hint	["solver-script-or-description"]
allowed-tools	Bash(*), Read, Grep, Glob, Edit, Write, Agent

运行求解器

在本地环境执行数模求解代码: $ARGUMENTS

Workflow

Step 1: 检测环境

Read the project's CLAUDE.md to determine the compute environment:

• Python: Look for Python version, virtualenv/conda info, required packages
• MATLAB: Look for MATLAB installation path and license info

If no environment info is found in CLAUDE.md, auto-detect:

# Check Python
python3 --version 2>/dev/null || python --version 2>/dev/null
pip list 2>/dev/null | grep -E "numpy|scipy|matplotlib|pandas|sklearn|pulp|cvxpy"

# Check MATLAB
which matlab 2>/dev/null && matlab -batch "disp('MATLAB OK')" 2>/dev/null

Step 2: 预检查

Check compute environment readiness:

# Check available memory
python3 -c "import psutil; print(f'RAM: {psutil.virtual_memory().total/1e9:.1f} GB, Available: {psutil.virtual_memory().available/1e9:.1f} GB')"

# Check required packages
python3 -c "
import importlib, sys
required = ['numpy', 'scipy', 'matplotlib', 'pandas']
missing = [p for p in required if importlib.util.find_spec(p) is None]
if missing: print(f'Missing: {missing}'); sys.exit(1)
print('All required packages available')
"

If packages are missing, install them:

pip install numpy scipy matplotlib pandas scikit-learn

Step 3: 准备求解脚本

1. Read the solve plan: Check SOLVE_PLAN.md or refine-logs/SOLVE_PLAN.md for the execution order
2. Locate scripts: Find solver scripts in the project directory

   Glob: **/*.py, **/*.m
   

3. Validate scripts: Quick syntax check before running

   python3 -m py_compile <script>
   

Step 3.5: 日志配置

Ensure the solver scripts have proper logging:

1. Check if logging is already in the script — look for import logging or print statements with progress info. If present, skip to Step 4.

2. If not present, add basic logging to the solver script:

   import logging, time
   logging.basicConfig(filename='solve_log.txt', level=logging.INFO,
                       format='%(asctime)s - %(message)s')
   
   # Inside solving loop:
   logging.info(f"Iteration {i}: objective={obj_value:.6f}, residual={residual:.6e}")
   
   # After completion:
   logging.info(f"Solving complete. Total time: {time.time()-start:.1f}s")
   

3. Metrics to log (add whichever apply to the solver):

   • iteration — current iteration number
   • objective — objective function value
   • residual — convergence residual
   • elapsed_time — time since start
   • R², RMSE, or other relevant metrics
   • Any custom metrics the solver already computes

Log files are saved to the project directory for later inspection by /monitor-solver and /analyze-results.

Step 4: 执行求解

Python

# Run solver with timeout protection
timeout 3600 python3 <script> <args> 2>&1 | tee solve_log.txt

# For long-running solvers, use background execution
python3 <script> <args> > solve_log.txt 2>&1 &
echo $! > solver_pid.txt

MATLAB

# Run MATLAB script
timeout 3600 matlab -batch "run('<script>')" 2>&1 | tee solve_log.txt

For long-running jobs, use run_in_background: true to keep the conversation responsive.

Step 5: 验证启动

Check process is running and producing output:

# Check if process is still running
ps aux | grep -E "python|matlab" | grep -v grep

# Check if log file is growing
ls -la solve_log.txt
tail -5 solve_log.txt

Step 6: 结果收集

After solver completes:

1. Check exit code for errors
2. Collect output files (results, figures, data)
3. Save a brief execution summary

echo "=== 求解完成 ==="
echo "退出码: $?"
echo "耗时: $(cat solve_log.txt | grep 'Total time' | tail -1)"
echo "输出文件:"
ls -la results/ output/ figures/ 2>/dev/null

Key Rules

• Large file handling: If the Write tool fails due to file size, immediately retry using Bash (cat << 'EOF' > file) to write in chunks. Do NOT ask the user for permission — just do it silently.
• ALWAYS check environment readiness first — never blindly run scripts
• Each solver gets its own background process with timeout protection
• Use tee to save logs for later inspection
• Run long commands with run_in_background: true to keep conversation responsive
• Report back: which script, what arguments, estimated time, PID
• If multiple sub-problems, run independent solvers in parallel
• Default timeout: 60 minutes. Warn if solver is expected to take longer.
• If solver fails, capture error output and suggest fixes

CLAUDE.md Example

Users should add their environment info to their project's CLAUDE.md:

## 计算环境
- Python: 3.10+ with numpy, scipy, matplotlib, pandas, scikit-learn
- MATLAB: R2024a (optional)
- 额外包: cvxpy (for optimization), networkx (for graph problems)
- 超时限制: 60 minutes per solver
- 数据目录: data/
- 结果目录: results/

环境配置: Ensure all required packages are installed before running. The skill will auto-detect and install missing Python packages, but MATLAB toolboxes must be pre-installed.