audit-dependency-versions
について
このClaudeスキルは、プロジェクトの依存関係を監査し、バージョンの陳腐化、セキュリティ脆弱性、互換性の問題をチェックします。ロックファイルを分析し、アップグレードパスを計画し、破壊的変更を評価します。リリース前、メンテナンス期間中、セキュリティ勧告後、あるいはプロジェクトを引き継ぐ際に使用し、依存関係の健全性を確保します。
クイックインストール
Claude Code
推奨npx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/audit-dependency-versionsこのコマンドをClaude Codeにコピー&ペーストしてスキルをインストールします
ドキュメント
Audit Dependency Versions
Audit project dependencies for version staleness, known security vulnerabilities, and compatibility issues. This skill inventories all dependencies from lock files, checks each against the latest available version, classifies staleness levels, identifies security concerns, and produces a prioritized upgrade report with recommended actions.
Cuándo Usar
- Before a release to ensure dependencies are current and secure
- During periodic maintenance (monthly or quarterly dependency reviews)
- After receiving a security advisory affecting a project dependency
- When upgrading a project to a new language version (e.g., R 4.4 to 4.5)
- Before submitting a package to CRAN, npm, or crates.io
- When inheriting a project and assessing its dependency health
Entradas
- Requerido: Project root directory containing dependency/lock files
- Opcional: Ecosystem type if not auto-detectable (R, Node.js, Python, Rust)
- Opcional: Security-only mode flag (skip staleness, focus on CVEs)
- Opcional: Allowlist of dependencies to skip (known acceptable older versions)
- Opcional: Target date for compatibility (e.g., "must work with R 4.4.x")
Procedimiento
Paso 1: Inventory All Dependencies
Locate and parse dependency files to build a complete inventory.
R packages:
# Direct dependencies from DESCRIPTION
grep -A 100 "^Imports:" DESCRIPTION | grep -B 100 "^[A-Z]" | head -50
grep -A 100 "^Suggests:" DESCRIPTION | grep -B 100 "^[A-Z]" | head -50
# Pinned versions from renv.lock
cat renv.lock | grep -A 3 '"Package"'
Node.js:
# Direct dependencies
cat package.json | grep -A 100 '"dependencies"' | grep -B 100 "}"
cat package.json | grep -A 100 '"devDependencies"' | grep -B 100 "}"
# Pinned versions from lock file
cat package-lock.json | grep '"version"' | head -20
Python:
# From requirements or pyproject
cat requirements.txt
cat pyproject.toml | grep -A 50 "dependencies"
# Pinned versions
cat requirements.lock 2>/dev/null || pip freeze
Rust:
# From Cargo.toml
grep -A 50 "\[dependencies\]" Cargo.toml
# Pinned versions
cat Cargo.lock | grep -A 2 "name ="
Build an inventory table:
| Package | Pinned Version | Type | Ecosystem |
|---|---|---|---|
| dplyr | 1.1.4 | Import | R |
| testthat | 3.2.1 | Suggests | R |
| express | 4.18.2 | dependency | Node.js |
| pytest | 8.0.0 | dev | Python |
Esperado: Complete inventory of all direct and (optionally) transitive dependencies with pinned versions.
En caso de fallo: If lock files are missing, the project has reproducibility issues. Note this as a finding and inventory from the manifest file (DESCRIPTION, package.json) using declared version constraints instead of pinned versions.
Paso 2: Check Latest Available Versions
For each dependency, determine the latest available version.
R:
# Check available versions
available.packages()[c("dplyr", "testthat"), "Version"]
# Or via CLI
Rscript -e 'cat(available.packages()["dplyr", "Version"])'
Node.js:
# Check outdated packages
npm outdated --json
# Or individual package
npm view express version
Python:
# Check outdated
pip list --outdated --format=json
# Or individual
pip index versions requests 2>/dev/null
Rust:
# Check outdated
cargo outdated
# Or individual
cargo search serde --limit 1
Update the inventory with latest versions:
| Package | Pinned | Latest | Gap |
|---|---|---|---|
| dplyr | 1.1.4 | 1.1.6 | patch |
| ggplot2 | 3.4.0 | 3.5.1 | minor |
| Rcpp | 1.0.10 | 1.0.14 | patch |
| shiny | 1.7.4 | 1.9.1 | minor |
Esperado: Latest version identified for each dependency with the gap magnitude (patch/minor/major).
En caso de fallo: If a package registry is unreachable, note the dependency as "unable to check" and proceed with the rest. Do not block the entire audit on one unreachable registry.
Paso 3: Classify Staleness
Assign a staleness level to each dependency:
| Level | Definition | Action |
|---|---|---|
| Current | At latest version or within latest patch | No action needed |
| Patch behind | Same major.minor, older patch | Low priority upgrade, usually safe |
| Minor behind | Same major, older minor | Medium priority, review changelog for new features |
| Major behind | Older major version | High priority, likely breaking changes in upgrade |
| EOL / Archived | Package no longer maintained | Critical: find replacement or fork |
Produce a staleness summary:
### Staleness Summary
- **Current**: 12 packages (48%)
- **Patch behind**: 8 packages (32%)
- **Minor behind**: 3 packages (12%)
- **Major behind**: 1 package (4%)
- **EOL/Archived**: 1 package (4%)
**Overall health**: AMBER (major-behind and EOL packages present)
Color coding:
- GREEN: All packages current or patch-behind
- AMBER: Any minor-behind or one major-behind
- RED: Multiple major-behind or any EOL packages
Esperado: Every dependency classified by staleness with an overall health rating.
En caso de fallo: If version comparison logic is ambiguous (non-SemVer versions, date-based versions), classify conservatively as "minor behind" and note the non-standard versioning.
Paso 4: Check for Security Vulnerabilities
Run ecosystem-specific security audit tools:
R:
# No built-in audit tool; check manually
# Cross-reference with https://www.r-project.org/security.html
# Check GitHub advisories for each package
Node.js:
# Built-in audit
npm audit --json
# Severity levels: info, low, moderate, high, critical
npm audit --audit-level=moderate
Python:
# Using pip-audit
pip-audit --format=json
# Or safety
safety check --json
Rust:
# Using cargo-audit
cargo audit --json
Document findings:
### Security Findings
| Package | Version | CVE | Severity | Fixed In | Description |
|---|---|---|---|---|---|
| express | 4.18.2 | CVE-2024-XXXX | High | 4.19.0 | Path traversal in static file serving |
| lodash | 4.17.20 | CVE-2021-23337 | Critical | 4.17.21 | Command injection via template |
**Security status**: RED (1 critical, 1 high)
Esperado: Security vulnerabilities identified with CVE, severity, affected version, and fix version.
En caso de fallo: If no audit tool is available for the ecosystem, search GitHub Security Advisories manually for each dependency. Note that the audit is best-effort without tooling.
Paso 5: Plan Upgrade Path
Prioritize upgrades based on risk and impact:
### Upgrade Plan
#### Priority 1: Security Fixes (do immediately)
| Package | Current | Target | Risk | Notes |
|---|---|---|---|---|
| lodash | 4.17.20 | 4.17.21 | Low (patch) | Fixes CVE-2021-23337 |
| express | 4.18.2 | 4.19.0 | Low (minor) | Fixes CVE-2024-XXXX |
#### Priority 2: EOL Replacements (plan within 1 month)
| Package | Current | Replacement | Migration Effort |
|---|---|---|---|
| request | 2.88.2 | node-fetch 3.x | Medium (API change) |
#### Priority 3: Major Version Upgrades (plan for next release cycle)
| Package | Current | Target | Breaking Changes |
|---|---|---|---|
| webpack | 4.46.0 | 5.90.0 | Config format, plugin API |
#### Priority 4: Minor/Patch Updates (batch in maintenance window)
| Package | Current | Target | Notes |
|---|---|---|---|
| dplyr | 1.1.4 | 1.1.6 | Patch fixes only |
| ggplot2 | 3.4.0 | 3.5.1 | New geom functions added |
For each major upgrade, note known breaking changes by checking the dependency's changelog.
Esperado: Prioritized upgrade plan with security fixes first, then EOL replacements, major upgrades, and minor/patch batches.
En caso de fallo: If a dependency has no clear upgrade path (abandoned with no fork), document the risk and recommend: (1) vendoring the current version, (2) finding an alternative package, or (3) accepting the risk with monitoring.
Paso 6: Document Compatibility Risks
For each planned upgrade, assess compatibility:
### Compatibility Assessment
#### express 4.18.2 -> 4.19.0
- **API changes**: None (patch-level fix)
- **Node.js requirement**: Same (>=14)
- **Test impact**: Run full test suite; expect zero failures
- **Confidence**: HIGH
#### webpack 4.46.0 -> 5.90.0
- **API changes**: Config file format changed, several plugins removed
- **Node.js requirement**: >=10.13 (unchanged)
- **Test impact**: Build configuration must be rewritten; all tests need re-run
- **Confidence**: LOW (requires dedicated migration effort)
- **Migration guide**: https://webpack.js.org/migrate/5/
Write the complete audit report to DEPENDENCY-AUDIT.md or DEPENDENCY-AUDIT-2026-02-17.md.
Esperado: Compatibility risks documented for each significant upgrade. Complete audit report written.
En caso de fallo: If compatibility cannot be assessed without testing, recommend a branch-based upgrade approach: create a branch, apply the upgrade, run tests, and evaluate results before merging.
Validación
- All direct dependencies inventoried from lock/manifest files
- Latest available version checked for each dependency
- Staleness level assigned (current / patch / minor / major / EOL)
- Overall health rating calculated (GREEN / AMBER / RED)
- Security audit run with ecosystem-appropriate tooling
- All CVEs documented with severity, affected version, and fix version
- Upgrade plan prioritized: security > EOL > major > minor/patch
- Compatibility risks assessed for each major upgrade
- Audit report written to DEPENDENCY-AUDIT.md
- No dependencies left as "unable to check" without documented reason
Errores Comunes
-
Ignoring transitive dependencies: A project may have 10 direct dependencies but 200 transitive ones. Security vulnerabilities often hide in transitive dependencies. Use
npm lsorrenv::dependencies()to see the full tree. -
Upgrading everything at once: Batch-upgrading all dependencies in one commit makes it impossible to identify which upgrade caused a regression. Upgrade in logical groups (security first, then majors individually, then minors/patches as a batch).
-
Confusing "outdated" with "insecure": A package one major version behind with no CVEs is lower risk than a current package with a critical vulnerability. Always prioritize security over freshness.
-
Not reading changelogs: Blindly upgrading a major version without reading the changelog. Breaking changes in the dependency become breaking changes in your project.
-
Audit fatigue: Running audits but not acting on findings. Set a policy: security findings must be addressed within 1 sprint, EOL within 1 quarter.
-
Missing lock files: Projects without lock files have non-reproducible builds. If the audit reveals missing lock files, that is itself a critical finding to address before versioned upgrades.
-
Wrong R binary on hybrid systems: On WSL or Docker,
Rscriptmay resolve to a cross-platform wrapper instead of native R. Check withwhich Rscript && Rscript --version. Prefer the native R binary (e.g.,/usr/local/bin/Rscripton Linux/WSL) for reliability. See Setting Up Your Environment for R path configuration.
Habilidades Relacionadas
apply-semantic-versioning-- Version bumps may be triggered by dependency upgradesmanage-renv-dependencies-- R-specific dependency management with renvsecurity-audit-codebase-- Broader security audit that includes dependency vulnerabilitiesmanage-changelog-- Document dependency upgrades in the changelogplan-release-cycle-- Schedule dependency upgrades within the release timeline
GitHub リポジトリ
関連スキル
llamaguard
その他LlamaGuardは、暴力やヘイトスピーチなど6つの安全性カテゴリーにおいて、LLMの入力と出力をモデレートするMetaの70-80億パラメータモデルです。94〜95%の精度を提供し、vLLM、Hugging Face、Amazon SageMakerを使用してデプロイ可能です。このスキルを使用して、AIアプリケーションにコンテンツフィルタリングと安全策を簡単に統合できます。
cost-optimization
その他このClaudeスキルは、リソースの適正サイジング、タグ付け戦略、支出分析を通じて、開発者がクラウドコストを最適化することを支援します。AWS、Azure、GCPにわたるクラウド支出の削減とコストガバナンスの実施のためのフレームワークを提供します。インフラコストの分析、リソースの適正サイジング、または予算制約への対応が必要な際にご利用ください。
quantizing-models-bitsandbytes
その他このスキルは、bitsandbytesを使用してLLMを8ビットまたは4ビット精度に量子化し、精度の低下を最小限に抑えつつ50〜75%のメモリ削減を実現します。限られたGPUメモリでより大規模なモデルを実行したり、推論を高速化するのに理想的で、INT8、NF4、FP4などのフォーマットをサポートしています。HuggingFace Transformersと統合され、QLoRAトレーニングや8ビットオプティマイザーを可能にします。
dispatching-parallel-agents
その他このClaudeスキルは、複数のエージェントを配備し、3つ以上の独立した問題を並行して調査・修正します。共有状態や依存関係がなく解決可能な、無関係な障害が発生するシナリオ向けに設計されています。中核となる機能は並列問題解決であり、効率を最大化するために独立した問題領域ごとに1つのエージェントを割り当てます。
