菜单
Dead code detection strategies, safe removal processes, dependency pruning, and bloat metrics for systematically reducing codebase dead weight. Use when evaluating codebase health during architecture-audit, performing cleanup during cook, reviewing unused dependencies, removing commented-out code, or measuring code bloat.
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。
基于 SOC 职业分类
| name | code-pruning |
| description | Dead code detection strategies, safe removal processes, dependency pruning, and bloat metrics for systematically reducing codebase dead weight. Use when evaluating codebase health during architecture-audit, performing cleanup during cook, reviewing unused dependencies, removing commented-out code, or measuring code bloat. |
| Approach | How It Works | Confidence | Limitation |
|---|---|---|---|
| Static analysis | Parse AST, trace references | High for unused exports/functions | Misses dynamic dispatch, reflection |
| Dynamic analysis | Runtime instrumentation, coverage | High for executed paths | Misses rarely-used paths (disaster recovery, seasonal) |
| Combined (SCARF) | Static + dynamic + team review | Highest | Requires team coordination |
| Manual review | Human inspection, git blame | Variable | Doesn't scale; subjective |
| Phase | Action | Checkpoint |
|---|---|---|
| Survey | Run static analysis tools; identify candidates | Candidate list generated |
| Classify | Categorize each: dead, dormant, speculative, deprecated | Each candidate classified |
| Announce | Deprecation warnings; notify stakeholders | Team aware of pending removals |
| Remove | Delete with tests verifying no breakage | All tests pass after removal |
| Follow-up | Monitor production for regressions; verify no side effects | No production issues after 1 week |
For smaller teams: Compress Announce into a PR description or commit message. The key phases are Survey, Classify, and Remove-with-tests.
| Language | Tool | Notes |
|---|---|---|
| JS/TS | Knip | Supersedes ts-prune; handles re-exports, dependencies, and unused files |
| JS/TS | depcheck | Focused on unused npm dependencies |
| Python | Vulture | AST-based dead code detection; configure --min-confidence 80 |
| Python | autoflake | Removes unused imports and variables |
| Java | PMD | Static analysis rules for unused code |
| Go | deadcode | Official Go team tool (1.22+); superior to unused |
| Multi | SonarQube | Cross-language; includes unused code rules |
| False Positive | How to Verify |
|---|---|
| Peer dependency | Check if required by another installed package |
| CLI tool | Check npm scripts, Makefile, CI config |
| Plugin/loader | Check config files (webpack, babel, pytest, etc.) |
| Type-only import | Check .d.ts files and type annotations |
| Indicator | Healthy | Warning | Critical |
|---|---|---|---|
| Unreachable code ratio | <5% | 5-15% | >15% |
| Unused dependency count | 0-2 | 3-5 | >5 |
| Commented-out code blocks | 0 | 1-5 | >5 |
| Single-implementation interfaces | <3 | 3-10 | >10 |
| Evidence | Confidence | Action |
|---|---|---|
| Zero static references + zero runtime coverage | High | Delete (with tests) |
| Zero static refs but in error/recovery path | Medium | Verify with team before deleting |
| Referenced only by tests | Medium | Delete code and tests together |
| Referenced by commented-out code only | High | Delete both |
| Dynamic dispatch possible (reflection, eval) | Low | Instrument before deleting |
| Signal | Action |
|---|---|
| Zero imports in source code | Remove (check for false positives first) |
| Only in devDependencies, not used in scripts | Remove |
| Imported but functionality is unused | Replace with lighter alternative or remove |
| Peer dependency of another package | Keep — removing breaks the dependent |
| Last updated >2 years, no maintained fork | Plan migration to maintained alternative |
Antipattern catalog with categorized patterns, symptoms, examples, and fixes. Use when reviewing code for quality issues, detecting bad patterns during plan-audit or scope, evaluating whether a pattern is an antipattern or acceptable tradeoff, or fixing existing code smells. Covers surprise, misuse, complexity, and premature antipattern categories with severity classification.
Naming decisions for variables, functions, classes, constants, and APIs. Use when choosing names, reviewing naming quality, refactoring unclear names, or establishing naming conventions. Covers descriptive naming, naming as documentation, language-specific conventions, and common naming antipatterns.
Plan quality auditing with completeness scorecard, antipattern risk detection, and readiness checks. Use when auditing implementation plans, evaluating plan completeness, checking error handling strategy, assessing naming and readability pre-checks, reviewing testing strategy, evaluating tradeoff documentation, or running plan-audit on a software project. Covers the 10-point completeness scorecard, code quality pre-checks, antipattern risk assessment, and build-readiness decision.
Code quality pillars, goals, abstraction layers, and tradeoff thinking. Use when evaluating code quality, setting quality goals, choosing abstraction levels, making design tradeoffs, or auditing code against quality pillars. Covers readability, modularity, testability, reusability, and the principle of least astonishment.
Code readability patterns covering comments, function decomposition, nesting, naming conventions, and code structure. Use when evaluating whether code is easy to understand, deciding when and how to write comments, decomposing functions, reducing nesting depth, improving code structure, or reviewing code for readability. Covers comment strategy, function length, abstraction level mixing, nesting reduction, anonymous function readability, and making bad code visually obvious.
Code review process, effective feedback, PR best practices, team working agreements, review automation, AI-augmented review, and review antipatterns. Use when reviewing code, preparing PRs, giving review feedback, setting up review processes, writing review comments, or conducting code reviews during serve phase.
