Dependencies Skill

Tools for dependency analysis data collection. Scripts provide data, AI provides intelligence.

Philosophy

The rich reports users value come from AI intelligence, not script logic:

•Scripts run Trivy, query Maven Central, parse POMs
•AI understands context, identifies legacy packages, assesses breaking changes
•AI writes contextual recommendations, not template fill-ins

Available Tools

Trivy Vulnerability Scanning

Via maven skill's scan.py:

bash

# Full output (can be large - 160KB+ for complex projects)
uv run {workspace}/.claude/skills/maven/scripts/scan.py scan \
  --path {project-path} \
  --json

# Compact output (recommended - deduplicates CVEs, ~90% smaller)
uv run {workspace}/.claude/skills/maven/scripts/scan.py scan \
  --path {project-path} \
  --compact \
  --json

Compact mode:

•Deduplicates CVEs by (CVE_ID, package_name)
•Shows full details only for CRITICAL/HIGH severity
•Includes versions_found[] and occurrence_count to preserve version spread info
•Reduces output from ~160KB to ~16KB for typical projects

POM Analysis

Via maven skill's scan.py:

bash

uv run {workspace}/.claude/skills/maven/scripts/scan.py analyze \
  --path {pom.xml-path} \
  --check-versions \
  --json

Version Checking

Via maven skill's check.py:

bash

uv run {workspace}/.claude/skills/maven/scripts/check.py check \
  -d {groupId:artifactId} \
  -v {version} \
  --json

Data Aggregation (JSON only)

Via report.py - aggregates data for CI/automation:

bash

uv run {workspace}/.claude/skills/dependencies/scripts/report.py \
  {project-path} \
  --json

Note: report.py provides JSON data aggregation. For rich markdown reports, use the /dependencies command which applies AI intelligence.

Usage

For Rich Reports (Recommended)

Use the /dependencies command:

code

/dependencies partition
/dependencies /path/to/project

The command orchestrates tools and applies AI intelligence for:

•Context-aware analysis
•Legacy package identification
•Breaking change assessment
•Downstream impact evaluation
•Risk-prioritized recommendations

For Raw Data (CI/Automation)

Use scripts directly for structured JSON:

bash

# Vulnerability data (compact mode recommended)
uv run .claude/skills/maven/scripts/scan.py scan --path ./project --compact --json

# Full vulnerability data (large output)
uv run .claude/skills/maven/scripts/scan.py scan --path ./project --json

# Version data
uv run .claude/skills/maven/scripts/check.py check -d "org.springframework:spring-core" -v "6.1.0" --json

Risk Framework

Risk is calculated by combining multiple factors, not just version bump type.

Risk Score = Category + Jump + CVE + Location

Factor	Modifier	Examples
Category
Framework	+2	Spring Boot, Spring Security, Quarkus
Serialization/Network/DB/Security/Cloud	+1	Jackson, Netty, Protobuf, Azure SDK
Utility/Testing	0	Commons-*, Guava, JUnit
Version Jump
Patch	0	x.y.Z changes
Minor	+1	x.Y.z changes
Major	+3	X.y.z changes
CVE Context
CRITICAL CVE	-1	Urgency justifies faster action
HIGH/MEDIUM/LOW CVE	0	Standard priority
No CVE (proactive)	+1	No urgency, more caution
Fix Location
Direct/proper fix	0	Update in this service
Temporary override	+1	Upstream library should fix

Risk Levels:

•LOW (0-1): Batch apply, standard validation
•MEDIUM (2-3): Individual commits, standard validation
•HIGH (4+): Research first, extended validation

See reference/risk-framework.md for complete details.

Report Structure

CRITICAL: Organize reports by update risk level, NOT by CVE severity.

Input → Output Transformation

Input (from Trivy)	Output (in report)
CVE Severity (CRITICAL/HIGH/MEDIUM/LOW)	Update Risk (LOW/MEDIUM/HIGH)
"How dangerous is this vulnerability?"	"How safe is this fix to apply?"

Required Output Sections

code

## Updates by Risk Level

### LOW Risk (Score 0-1)
[Batch apply - list updates with score breakdown]

### MEDIUM Risk (Score 2-3)
[Individual commits - list updates with score breakdown]

### HIGH Risk (Score 4+)
[Research first - list updates with score breakdown]

## For /remediate Command
[Machine-readable tables grouped by risk level]

Examples

•CRITICAL CVE + patch fix = LOW risk update (urgent AND safe)
•No CVE + major framework bump = HIGH risk update (not urgent, potentially breaking)

Anti-pattern: Do NOT organize sections by CVE severity (Critical Vulnerabilities, High Vulnerabilities, etc.). Users need to know what's safe to fix first, not what's most dangerous.

Workflow

code

┌─────────────────┐         ┌─────────────────┐
│  /dependencies  │────────▶│   /remediate    │
│                 │         │                 │
│  AI-driven      │         │  Apply updates  │
│  analysis       │         │  with validation│
└─────────────────┘         └─────────────────┘
        │                           │
        ▼                           ▼
   {workspace}/reports/       git commits +
   dependencies-{project}-    code changes
     {date}.md