Skip to main content

AI Assistant Integration

CVE Lite CLI ships with a built-in command that installs skill files for five AI coding assistants. Once installed, your AI assistant knows how to run a scan, read the results, and produce a prioritized remediation plan — without any extra prompting on your part.

Installing skills

Run this once in your project root:

cve-lite install-skill

Output:

CVE Lite CLI skills installed:

✓ Claude Code .claude/commands/cve-lite.md
✓ Codex CLI AGENTS.md (section written)
✓ Gemini CLI GEMINI.md (section written)
✓ Cursor .cursor/rules/cve-lite.mdc
✓ GitHub Copilot .github/copilot-instructions.md (section written)

Commit these files to your repo to share them with your team.

Running it twice is safe — existing sections are replaced in place, surrounding content is preserved.

What gets installed

ToolFileBehavior
Claude Code.claude/commands/cve-lite.mdCreated or overwritten
Codex CLIAGENTS.md## CVE Lite CLI section written
Gemini CLIGEMINI.md## CVE Lite CLI section written
Cursor.cursor/rules/cve-lite.mdcCreated or overwritten with front-matter
GitHub Copilot.github/copilot-instructions.md## CVE Lite CLI section written

Commit these files to your repository. Every developer who clones the repo gets the skills automatically.


The AI-assisted security workflow

Once the skill is installed, the workflow is:

1. Scan and save results

cve-lite . --json

The scan results are saved to a timestamped file (cve-lite-scan-<timestamp>.json) in the current directory.

2. Ask your AI assistant to analyze findings

In Claude Code, invoke the skill with /cve-lite. In other tools the skill is picked up automatically when you ask about vulnerabilities.

The assistant will:

  • Read the JSON output
  • Prioritize findings by severity and relationship (direct before transitive)
  • Check whether vulnerable packages are actually imported in your source code
  • Identify patterns — for example, one parent dependency causing multiple transitive findings
  • Produce a concrete remediation plan with the exact commands to run

3. Apply fixes

The runnableFixCommand field in each finding contains the exact install command for your package manager. The assistant will surface these alongside its analysis.


What the skill teaches the assistant

The skill file covers four things:

Getting scan data

How to invoke cve-lite . --json and which fields matter in the output:

  • package, version — the vulnerable package
  • severitycritical | high | medium | low | unknown
  • relationshipdirect | transitive
  • firstFixedVersion — minimum safe version, if known
  • runnableFixCommand — exact install command to run, if available
  • recommendedAction — human-readable fix guidance
  • cves — CVE IDs
  • dependencyPaths — chains showing how the package is pulled in
  • usage.imported — whether the package is actually imported in source files
  • suggestedFixCommands — grouped, copy-ready fix commands at the top level

Prioritization rules

  1. Critical before high before medium before low
  2. Direct dependencies before transitive
  3. If usage.imported === false, flag as lower practical risk but do not dismiss
  4. If runnableFixCommand is present, that is the exact command to run

Codebase analysis

  • Cross-reference vulnerable packages against source file imports to confirm reachability
  • Check package.json version constraints for direct dependency findings
  • Use dependencyPaths to trace transitive chains and identify which parent package to upgrade
  • Look for patterns: a single parent responsible for multiple transitive findings

Output

The assistant produces a prioritized remediation list with severity, relationship, import status, and copy-ready fix commands for each finding — plus a summary of what remains after applying the suggested commands.


Why commit the skill files?

Skill files are plain text committed to your repository. Any developer who clones the repo and opens it in a supported AI assistant gets the context automatically — no manual setup, no per-developer configuration.

For files that support multiple sections (AGENTS.md, GEMINI.md, .github/copilot-instructions.md), the CVE Lite CLI section is added alongside any existing content. Your own guidance is preserved.