AgentSkillsCN

review-julia

对 Julia 脚本执行代码审查流程，检查代码质量、类型稳定性、并行计算模式以及科学计算规范，生成报告但不直接修改原稿文件。

SKILL.md
--- frontmatter
name: review-julia
description: Run the Julia code review protocol on Julia scripts. Checks code quality, type stability, parallel computing patterns, and scientific computing standards. Produces a report without editing files.
disable-model-invocation: true
argument-hint: "[filename or 'all']"
allowed-tools: ["Read", "Grep", "Glob", "Write", "Task"]

Review Julia Code

Run a comprehensive Julia code review on the specified script(s). Do NOT edit any source files -- produce a report only.

Steps

  1. •

    Identify target: Use $ARGUMENTS to find the Julia file(s). If all, scan all .jl files in the project.

  2. •

    Read standards from .claude/rules/julia-code-conventions.md.

  3. •

    Check these categories:

    • •Module Structure: Proper module organization, exports, includes
    • •Parallel Computing: @everywhere annotations, pmap usage, worker data distribution
    • •Optimization: Convergence checks, multiple starting values, grid search patterns
    • •Type Stability: Concrete types in hot loops, @code_warntype recommendations
    • •Path Conventions: joinpath() usage, no hardcoded OS-specific separators
    • •Naming: snake_case functions, CamelCase types, paper notation alignment
    • •Common Pitfalls: Missing @everywhere, local minima, large closures in pmap
  4. •

    Save report to quality_reports/[script_name]_julia_review.md.

  5. •

    Present summary: Total issues, severity breakdown, top critical issues.

Important

  • •NEVER edit source files. Report only.
  • •Prioritize correctness and performance over style.
  • •For Julia code generation patterns (MLE, GMM, simulation), see /econometrics-julia.