Skip to content
Back to skills

M15 Anti Pattern

ASecurity

Use when reviewing code for anti-patterns. Keywords: anti-pattern, common mistake, pitfall, code smell, bad practice, code review, is this an anti-pattern, better way to do this, common mistake to avoid, why is this bad, idiomatic way, beginner mistake, fighting borrow checker, clone everywhere, unwrap in production, should I refactor, 反模式, 常见错误, 代码异味, 最佳实践, 地道写法

  • 18 stars
  • 0 votes
  • 0 copies
  • 4 views
  • Added August 31, 2026
code-qualityrustrefactoringapiperformance

Works with

  • api

Security analysis

A100/100

Pro scans all 2 files and shows the line behind each finding

Scanned August 31, 2026

npx -y skills add k1-c/linear-tui --skill m15-anti-pattern --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of M15 Anti Pattern?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for M15 Anti Pattern
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/k1-c-m15-anti-pattern/badge)](https://www.skillsdirectory.com/skills/k1-c-m15-anti-pattern)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: m15-anti-pattern
description: "Use when reviewing code for anti-patterns. Keywords: anti-pattern, common mistake, pitfall, code smell, bad practice, code review, is this an anti-pattern, better way to do this, common mistake to avoid, why is this bad, idiomatic way, beginner mistake, fighting borrow checker, clone everywhere, unwrap in production, should I refactor, 反模式, 常见错误, 代码异味, 最佳实践, 地道写法"
user-invocable: false
---

# Anti-Patterns

> **Layer 2: Design Choices**

## Core Question

**Is this pattern hiding a design problem?**

When reviewing code:
- Is this solving the symptom or the cause?
- Is there a more idiomatic approach?
- Does this fight or flow with Rust?

---

## Anti-Pattern → Better Pattern

| Anti-Pattern | Why Bad | Better |
|--------------|---------|--------|
| `.clone()` everywhere | Hides ownership issues | Proper references or ownership |
| `.unwrap()` in production | Runtime panics | `?`, `expect`, or handling |
| `Rc` when single owner | Unnecessary overhead | Simple ownership |
| `unsafe` for convenience | UB risk | Find safe pattern |
| OOP via `Deref` | Misleading API | Composition, traits |
| Giant match arms | Unmaintainable | Extract to methods |
| `String` everywhere | Allocation waste | `&str`, `Cow<str>` |
| Ignoring `#[must_use]` | Lost errors | Handle or `let _ =` |

---

## Thinking Prompt

When seeing suspicious code:

1. **Is this symptom or cause?**
   - Clone to avoid borrow? → Ownership design issue
   - Unwrap "because it won't fail"? → Unhandled case

2. **What would idiomatic code look like?**
   - References instead of clones
   - Iterators instead of index loops
   - Pattern matching instead of flags

3. **Does this fight Rust?**
   - Fighting borrow checker → restructure
   - Excessive unsafe → find safe pattern

---

## Trace Up ↑

To design understanding:

```
"Why does my code have so many clones?"
    ↑ Ask: Is the ownership model correct?
    ↑ Check: m09-domain (data flow design)
    ↑ Check: m01-ownership (reference patterns)
```

| Anti-Pattern | Trace To | Question |
|--------------|----------|----------|
| Clone everywhere | m01-ownership | Who should own this data? |
| Unwrap everywhere | m06-error-handling | What's the error strategy? |
| Rc everywhere | m09-domain | Is ownership clear? |
| Fighting lifetimes | m09-domain | Should data structure change? |

---

## Trace Down ↓

To implementation (Layer 1):

```
"Replace clone with proper ownership"
    ↓ m01-ownership: Reference patterns
    ↓ m02-resource: Smart pointer if needed

"Replace unwrap with proper handling"
    ↓ m06-error-handling: ? operator
    ↓ m06-error-handling: expect with message
```

---

## Top 5 Beginner Mistakes

| Rank | Mistake | Fix |
|------|---------|-----|
| 1 | Clone to escape borrow checker | Use references |
| 2 | Unwrap in production | Propagate with `?` |
| 3 | String for everything | Use `&str` |
| 4 | Index loops | Use iterators |
| 5 | Fighting lifetimes | Restructure to own data |

## Code Smell → Refactoring

| Smell | Indicates | Refactoring |
|-------|-----------|-------------|
| Many `.clone()` | Ownership unclear | Clarify data flow |
| Many `.unwrap()` | Error handling missing | Add proper handling |
| Many `pub` fields | Encapsulation broken | Private + accessors |
| Deep nesting | Complex logic | Extract methods |
| Long functions | Multiple responsibilities | Split |
| Giant enums | Missing abstraction | Trait + types |

---

## Common Error Patterns

| Error | Anti-Pattern Cause | Fix |
|-------|-------------------|-----|
| E0382 use after move | Cloning vs ownership | Proper references |
| Panic in production | Unwrap everywhere | ?, matching |
| Slow performance | String for all text | &str, Cow |
| Borrow checker fights | Wrong structure | Restructure |
| Memory bloat | Rc/Arc everywhere | Simple ownership |

---

## Deprecated → Better

| Deprecated | Better |
|------------|--------|
| Index-based loops | `.iter()`, `.enumerate()` |
| `collect::<Vec<_>>()` then iterate | Chain iterators |
| Manual unsafe cell | `Cell`, `RefCell` |
| `mem::transmute` for casts | `as` or `TryFrom` |
| Custom linked list | `Vec`, `VecDeque` |
| `lazy_static!` | `std::sync::OnceLock` |

---

## Quick Review Checklist

- [ ] No `.clone()` without justification
- [ ] No `.unwrap()` in library code
- [ ] No `pub` fields with invariants
- [ ] No index loops when iterator works
- [ ] No `String` where `&str` suffices
- [ ] No ignored `#[must_use]` warnings
- [ ] No `unsafe` without SAFETY comment
- [ ] No giant functions (>50 lines)

---

## Related Skills

| When | See |
|------|-----|
| Ownership patterns | m01-ownership |
| Error handling | m06-error-handling |
| Mental models | m14-mental-model |
| Performance | m10-performance |

Files in this skill

  • SKILL.md4.7 KB
  • patterns/common-mistakes.md8.1 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.