Content
96%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is an exemplary decision-procedure skill: lean, git-competent prose with copy-paste bash, exact menu and confirmation texts, and validation checkpoints at every destructive boundary (merge, worktree removal, branch deletion). Its only structural note is that it is a monolithic single file — entirely appropriate at ~200 lines, though the rationalizations table could be a reference file if the skill grows.
Suggestions
If the skill grows beyond its current size, move the 常见的合理化借口 table into a references/ file (e.g. references/rationalizations.md) and keep a one-line pointer in SKILL.md, preserving the lean overview pattern.
Consider mentioning the English equivalent of the core principle or key terms (merge, PR, worktree) alongside the Chinese, since the frontmatter tags are in English and cross-language matching may improve trigger and indexing fidelity.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body assumes Claude's competence throughout: no git tutorials, no explanation of what a worktree or PR is, and every code comment carries load ('现在就捕获 —— 此刻还在工作区里面。步骤 5 会切换目录，而清理（步骤 6）需要这个值'). The one duplication — the step-2 capture comment restated by the step-6 warning '不要在这里重新计算这些值' — is two sentences guarding against a real silent-failure mode, which earns its place rather than padding. This matches anchor 5 ('lean and efficient; every token earns its place'), not anchor 4, which requires over-explanation that could be trimmed — nothing here explains concepts Claude doesn't already know. | 5 / 5 |
Actionability | Fully executable throughout: copy-paste bash blocks ('git worktree remove "$WORKTREE_PATH"', 'git push -u origin <feature-branch>', the MAIN_ROOT capture with `git -C`), verbatim menu texts to display character-for-character ('照原文展示菜单'), an exact confirmation token ('输入 discard 以确认… 等待**这个精确的**确认词'), and a state → menu → cleanup decision table. The placeholders (<base-branch>, <feature-branch>, <测试命令>) are genuinely project-specific parameters, not pseudocode. This is the anchor-5 case — commands are copy-paste ready and the common cases (normal repo, named worktree, detached HEAD, refused deletion) are each covered with concrete handling. | 5 / 5 |
Workflow Clarity | A six-step sequence with validation at every destructive boundary: '如果测试失败，报告失败并停下——菜单是在测试全绿之后才出现的'; '先合并 —— 在删除任何东西之前先验证合并成功' followed by re-running tests on the merge result; merge-result test failure → stop and investigate with branch/worktree left in place; worktree removal refused → 3-option submenu instead of --force; and the discard path gated behind an exact confirmation word. The 快速参考 table and the step-2 environment matrix act as checklists. This matches anchor 5 ('explicit validation steps; feedback loops for error recovery; checklists') — the destructive-operation cap of 3 does not apply because validation and confirmation gates are pervasive. | 5 / 5 |
Progressive Disclosure | No bundle files exist (no references/, scripts/, or assets/), and the skill is a single self-contained file with clear, navigable section headers (概述, 步骤 1–6, 快速参考, 常见的合理化借口) and a quick-reference table summarizing all paths. The ~200-line body is a cohesive decision procedure that does not clearly need splitting, fitting anchor 4 ('good structure; most content is appropriately placed'). It is not anchor 5, which is defined by well-signaled one-level-deep references to separate files — none exist — and the under-50-line simple-skill exception does not apply at this length; the rationalizations table (10 rows) is the one block that could plausibly live in a reference file. | 4 / 5 |
Total | 19 / 20 Passed |