How to build an intelligent agent that can automatically score submitted Python files and reports?

Requirements are as follows:
Generate personalized and constructive improvement suggestions for specific issues in the code and reports.
For experimental reports: Conduct report scoring and analysis, including content completeness, quality of results and discussions, logic and clarity, etc.
For Python code: Conduct code scoring and inspection, including syntax and runtime checks, functional correctness evaluation, code quality analysis, etc.
Require the generation of an overall scoring summary, providing separate scores (on a 100-point scale) for the code and the report, along with a list of key deduction points and highlights.