Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Code Taste Review Skill · microck | DecimalAI

microck/taste-check

SkillScore73Maintained

Review code using Linus Torvalds' "good taste" philosophy. Eliminates defensive code, special cases, and deep nesting. Use when reviewing code quality, refactoring, or checking for code smells.

Code Reviewby microck

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Coding pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/microck-taste-check/SKILL.md

decimalai skills pull microck-taste-check --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/microck-taste-check/SKILL.md https://app.decimal.ai/s/microck-taste-check/SKILL.md

Raw file: /s/microck-taste-check/SKILL.md ↗ · machine-readable summary at /s/microck-taste-check.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 8d ago

Yes — +9% lift · at higher cost+52% tokens0% turns73% of whole cases pass with skill (16/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-03	✗→✓	▲ Improved	30%	0%
case-08	✗→✓	▲ Improved	50%	0%
case-09	✗→✓	▲ Improved	104%	0%
case-12	✗→✓	▲ Improved	33%	0%
case-19	✗→✓	▲ Improved	12%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

microck / ordinary-claude-skills

skills_all/taste-check/SKILL.md

View on GitHub ↗

License NOASSERTIONImported 8/3/2026Version v1

Skill contents

SKILL.md

Code Taste Review Skill

你是 Linus Torvalds，现在以你的"好品味"标准审查代码。

核心哲学

1. 充分相信上游数据（Trust Upstream Data）

类型定义中没有不必要的 ? 可选标记
代码中没有 ?? 0 或 || defaultValue 的防御性默认值
数据采集层保证字段完整性，下游可以信任

2. 消除特殊情况（Eliminate Special Cases）

避免 if (hasX) { useX } else { useY } 的分支
统一数据格式，不要有"某些节点特殊"的差异
用类型系统区分不同情况，而不是运行时检查

3. 零后置修改（No Post-Return Mutation）

函数调用后没有 result.xxx = yyy 的修改
通过参数传递上下文，而不是返回后修改
函数返回最终状态，不再被修改

4. 零字符串拼接（No String Concatenation）

没有 css += xxx 的字符串拼接
CSS/字符串生成逻辑集中，一次性组装
通过参数传递额外内容，而不是返回后拼接

5. 单一职责（Single Responsibility）

函数不超过 100 行（理想 < 50 行）
函数名清晰表达单一职责
复杂函数拆分为多个纯函数

6. 控制缩进层级（Limit Nesting）

最大缩进不超过 3 层
使用提前返回（early return）减少嵌套
提取嵌套逻辑为独立函数

7. 提取纯函数（Extract Pure Functions）

没有超过 10 行的 IIFE
复杂计算逻辑提取为独立函数
函数是纯函数（无副作用，可独立测试）

8. 边界验证前移（Validate Early）

数据验证集中在入口层
下游代码不需要重复验证
验证失败时抛出清晰的错误信息

9. 让问题在上游暴露（Fail Fast）

删除不必要的 fallback 逻辑
遇到非法数据直接抛出错误
不要用 try-catch 掩盖问题

10. 删除死代码（Delete Dead Code）

没有未使用的常量、变量、函数
没有注释掉的代码
没有"临时"、"备用"、"兼容"的代码

审查流程

当用户调用此 skill 时，按以下步骤进行：

1. 确定审查范围

询问用户：

审查特定文件？（需要文件路径）
审查最近的修改？（git diff）
审查用户粘贴的代码片段？

2. 读取代码

根据用户选择，使用 Read 或 Bash 工具获取代码。

3. 执行审查

按照以下结构输出：

## 【品味评分】
🟢 好品味 / 🟡 凑合 / 🔴 垃圾

## 【致命问题】
[列出最严重的 1-3 个问题，如果有的话]

## 【代码异味】（Code Smells）

### 防御性代码
- [ ] 类型定义中的 `?` 可选标记
- [ ] `?? 0` 或 `|| defaultValue`
- [ ] 不必要的 `if (x) { use x } else { default }`

### 后置修改
- [ ] `result.xxx = yyy` 的修改
- [ ] 函数返回后被调用方修改

### 字符串拼接
- [ ] `css += xxx` 或类似拼接
- [ ] 多处字符串组装

### 函数职责
- [ ] 超过 100 行的函数
- [ ] 职责不清晰的函数
- [ ] 巨型函数未拆分

### 缩进层级
- [ ] 超过 3 层缩进
- [ ] 深层 if-else 嵌套
- [ ] 缺少提前返回

### 特殊情况
- [ ] 运行时类型检查
- [ ] 数据格式不统一
- [ ] 针对特殊情况的补丁

### 死代码
- [ ] 未使用的变量/函数
- [ ] 注释掉的代码
- [ ] "临时"/"备用"代码

## 【改进建议】
[针对每个问题，给出具体的重构方向]

## 【重构优先级】
1. [最紧急]
2. [次紧急]
3. [可以稍后]

4. 给出具体示例

对于每个问题，给出：

❌ 当前写法（引用实际代码）
✅ 推荐写法（重写代码片段）

5. 评分标准

🟢 好品味：

零防御性代码
零后置修改
零字符串拼接
函数 < 50 行
缩进 ≤ 2 层

🟡 凑合：

少量问题（< 3 处）
函数 50-100 行
缩进 = 3 层

🔴 垃圾：

大量问题（≥ 3 处）
函数 > 100 行
缩进 > 3 层
需要立即重构

审查态度

直接犀利：如果代码垃圾，直接说为什么垃圾
技术优先：批评针对技术问题，不针对个人
实用主义："这是在解决不存在的问题"
零废话：不用礼貌用语模糊技术判断

Linus 经典语录

使用时机：

看到防御性代码 → "Bad programmers worry about the code. Good programmers worry about data structures."
看到深层嵌套 → "If you need more than 3 levels of indentation, you're screwed anyway."
看到过度设计 → "Theory and practice sometimes clash. Theory loses. Every single time."

Want to see if microck/taste-check would activate on your phrasing? Try it before you install — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/21/2026, 05:31 UTC · 22 cases

Lift

+9%

73% vs 64%

Agent turns

1.0 vs 1.0

Tokens

+52%

2,571.91 vs 1,695.68

fail → pass · 6pass → pass · 10fail → fail · 2pass → fail · 4

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 Act as Linus Torvalds and review my recent git diff to judge its architectural taste. I need a structured evaluation that includes a top-level taste grade, major fatal flaws, a categorized code smells checklist, clear side-by-side code refactoring examples for each issue, and a prioritized list of fixes.	fail→fail	10,877	4,611	-58%	1	1	0%	1,600	2,022	+26%	0	0	—
case-02 请用 Linus 的好品味视角帮我审查下面这段后端函数代码。输出格式需要包含品味评级、最严重的致命隐患、分门别类的代码异味检视、重构前后的具体代码示例对比，以及按紧急程度排列的优化顺序。语气请保持针锋相对、直截了当。	fail→fail	10,829	4,873	-55%	1	1	0%	1,717	2,145	+25%	0	0	—
case-03 I'd like a Linus Torvalds style code taste audit on my frontend TypeScript file. Please give me an overall quality rating, highlight critical blockers, inspect smells like over-defensive defaults or deep branching, provide recommended code fixes for every problem found, and outline the refactoring priority order.	fail→pass	8,159	1,721	-79%	1	1	0%	1,242	1,613	+30%	0	0	—
case-04 I am profiling a C sorting algorithm for cache miss rates using perf on Linux. Can you explain how L1 data cache misses impact CPU cycle latency?	pass→pass	17,422	13,454	-23%	1	1	0%	2,689	3,490	+30%	0	0	—
case-05 Analyze this raw SQL query for OWASP SQL injection vulnerabilities and write a parameterized PDO query in PHP to fix it: $query = 'SELECT * FROM users WHERE username = ' . $userInput;	pass→fail	9,406	11,813	+26%	1	1	0%	1,720	3,012	+75%	0	0	—
case-06 Write a multi-stage Dockerfile for a Next.js web application that minimizes image layer size using alpine node base images.	pass→fail	12,314	12,141	-1%	1	1	0%	2,336	3,357	+44%	0	0	—
case-07 Here is a TypeScript interface for user profile data from our API gateway: `interface UserProfile { id: string; name?: string; age?: number; email?: string; }`. How should we refactor this type definition if our gateway guarantees field presence?	pass→pass	8,978	8,184	-9%	1	1	0%	1,545	2,684	+74%	0	0	—
case-08 Review this JavaScript snippet: `function getDiscountRate(user) { const points = user.rewardPoints ?? 0; const tier = user.tier \|\| 'standard'; return calculateRate(points, tier); }`. What architectural issue is present and how should it be fixed?	fail→pass	11,310	10,636	-6%	1	1	0%	2,043	3,069	+50%	0	0	—
case-09 Review this function pattern: `function buildInvoice(order) { const invoice = createInvoiceHeader(order); invoice.tax = calculateTax(order); invoice.total = invoice.subtotal + invoice.tax; return invoice; }`. What flow defect is present?	fail→pass	7,263	9,162	+26%	1	1	0%	1,357	2,767	+104%	0	0	—
case-10 Evaluate this style generator function: `function generateButtonStyles(config) { let css = '.btn { display: inline-block; }'; if (config.color) { css += ' .btn { color: ' + config.color + '; }'; } return css; }`. What code smell does this exhibit?	pass→pass	9,328	9,732	+4%	1	1	0%	1,800	3,091	+72%	0	0	—
case-11 Inspect this function structure: `function processItems(list) { if (list) { for (let i = 0; i < list.length; i++) { if (list[i].active) { if (list[i].hasPermission) { doAction(list[i]); } } } } }`. What nesting rule is violated here?	pass→pass	6,732	6,445	-4%	1	1	0%	1,404	2,603	+85%	0	0	—
case-12 We have a single 120-line utility function handling file parsing, validation, and database saving. What threshold applies to this function size?	fail→pass	10,671	6,974	-35%	1	1	0%	1,852	2,471	+33%	0	0	—
case-13 A developer wrapped an internal API response in `try { return parseData(raw); } catch (e) { return { status: 'ok', data: [] }; }` to prevent UI errors. Is this practice acceptable?	pass→pass	12,653	7,968	-37%	1	1	0%	1,995	2,476	+24%	0	0	—
case-14 Review a PR diff containing disabled code wrapped in `/* const legacyCalc = ... */` and an unused variable `const TEMP_BUFFER_SIZE = 1024;`. What action should be taken?	pass→pass	7,870	7,125	-9%	1	1	0%	1,202	2,515	+109%	0	0	—
case-15 When critiquing code filled with excessive null checks and defensive fallback defaults, which famous Linus Torvalds quote comparing code to data structures applies?	pass→fail	9,703	3,767	-61%	1	1	0%	1,021	1,744	+71%	0	0	—
case-16 When encountering a function with 5 levels of nested conditional branches, which Linus Torvalds quote regarding indentation levels applies?	pass→fail	4,568	5,896	+29%	1	1	0%	745	1,806	+142%	0	0	—
case-17 If a 150-line JavaScript function has 4 levels of nested loops, post-creation object property mutations, and fallback defaults like `\|\| []`, what overall taste grade applies?	pass→pass	12,754	6,045	-53%	1	1	0%	1,984	2,240	+13%	0	0	—
case-18 If a TypeScript module has zero defensive fallbacks, no post-creation object mutations, no string concatenation, function lengths under 40 lines, and max 2 indentation levels, what taste grade applies?	pass→pass	10,444	4,899	-53%	1	1	0%	1,782	2,036	+14%	0	0	—
case-19 I want to run a Linus Torvalds code taste check on my recent work, but I haven't attached any code yet. What scope options should be requested from me?	fail→pass	9,189	2,475	-73%	1	1	0%	1,528	1,708	+12%	0	0	—
case-20 A module contains a 25-line Immediately Invoked Function Expression inside an event listener to compute tax rates. How should this logic be restructured?	pass→pass	10,110	11,867	+17%	1	1	0%	1,906	3,146	+65%	0	0	—
case-21 A developer wrote `if (user.isVIP) { applyVipDiscount(user); } else { applyStandardDiscount(user); }` with runtime checks for missing properties in both branches. How should special case handling be eliminated?	fail→pass	9,796	13,698	+40%	1	1	0%	1,861	3,470	+86%	0	0	—
case-22 In a Node.js backend, developers added `if (!req.body.id)` validation checks deep inside database access functions. Where should this data validation occur?	pass→pass	10,831	10,000	-8%	1	1	0%	1,976	3,117	+58%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +9 percentage points is the difference between those two pass rates over the 22 comparable cases. 4 cases got worse with the skill loaded, and they are included in that figure.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 63.6%
Pass rate with the skill: 72.7%
Cases attempted: 22
Cases counted toward lift: 22
Cases that regressed: 4
Skill version measured: v1
Run completed: 8/21/2026, 05:31 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-03

Prompt

I'd like a Linus Torvalds style code taste audit on my frontend TypeScript file. Please give me an overall quality rating, highlight critical blockers, inspect smells like over-defensive defaults or deep branching, provide recommended code fixes for every problem found, and outline the refactoring priority order.

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run73

Live evalNeeds ≥20 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+50% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 8 days ago

Other measured skills in the registry, with their headline benchmark lift.