菜单
Discover large source-file refactor candidates and propose cohesion- and risk-driven split plans using available semantic tooling.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/PaulRBerg/agent-skills --skill large-file-refactor命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Assess recurrence risk for agent behavior using local Codex/Claude Code transcripts and recommend evidence-backed durable fixes to AGENTS.md or skills. Not for routine post-success skill-evolution or agent self-improvement reviews.
Create or polish repo agent context: README.md, AGENTS.md/CLAUDE.md, installed project skills, and other Markdown context docs.
Use for current official documentation about Codex, Codex CLI, or Claude Code behavior, configuration, prompting, skills, permissions, tools, surfaces, capabilities, troubleshooting, hooks, app-server, or hook trust; fetch the relevant official URL before answering.
正在显示 SKILL.md
| argument-hint | [path] [--include-generated] |
| coordination | exempt |
| disable-model-invocation | true |
| name | large-file-refactor |
| user-invocable | true |
| description | Discover large source-file refactor candidates and propose cohesion- and risk-driven split plans using available semantic tooling. |
This skill is coordination-exempt: skip the ai-coord gate for its declared work.
If these instructions are already present in the conversation from a slash or dollar invocation, follow them directly; do not invoke this skill again through a skill tool.
Use LOC thresholds to discover candidates, then decide whether a split is justified by cohesion, coupling, ownership, and change risk. Test files use a relaxed 2000 LOC discovery threshold.
path: Optional file or directory to scan. Default: current working directory.--include-generated: Include generated, vendored, dependency, and build-output paths that are skipped by default.Resolve the skill directory, then run the helper from that directory:
uv run scripts/large-file-refactor.py "$ARGUMENTS"
If no arguments were provided, run the default scan:
uv run scripts/large-file-refactor.py
Preserve the helper's Markdown table as the exhaustive report. Do not omit matching rows, even when the refactor plan only covers a subset.
If the helper reports no threshold matches, stop after the report. A match is a candidate, not proof that the file should be split.
Draft a refactor plan for the 3 largest files only, unless the user explicitly requested another count.
For each candidate, rank split value by mixed responsibilities, change frequency/risk, coupling, and testability. Use whichever semantic symbol/reference tooling is available; prefer Serena when installed:
Do not implement the refactor unless the user separately asks for execution.
For each selected file, include:
Lead each plan with one explicit verdict: ### ✂ Split justified, ### 🧱 Keep intact, or
### ⚠️ Generated — change the generator. Use a compact evidence table for repeated criteria. When proposing a split,
show the source and target modules as a tree only when it clarifies ownership.
If a selected file is generated or vendored because --include-generated was used, plan against the generator, schema,
or upstream source instead of hand-splitting generated output.
tokei.Complete with the exhaustive threshold report plus evidence-ranked plans only for candidates whose cohesion or change
risk justifies a split. Lead with ### 🔎 Large-file scan — <candidate count>, state when a large file should remain
intact and why, and surface the portable-LOC fallback as ⚠️ Approximate counts when used. Keep the helper's exhaustive
table, paths, LOC values, and commands exact and undecorated.