Skip to content

Claude Code Skills Integration

This guide explains how to integrate clinvk with Claude Code Skills to extend Claude's capabilities with multi-backend AI support.

Why Use clinvk in Skills?

Claude Code Skills extend Claude's capabilities, but sometimes you need:

  • Other AI Backends: Gemini excels at data analysis, Codex at code generation
  • Multi-Model Collaboration: Complex tasks benefit from multiple perspectives
  • Parallel Processing: Run multiple AI tasks concurrently

Prerequisites

  1. clinvk installed and in your PATH
  2. At least one backend CLI installed (claude, codex, or gemini)

Basic Skill Example

Single Backend Call

Create a skill that uses Gemini for data analysis:

<!-- ~/.claude/skills/analyze-data/SKILL.md -->
# Data Analysis Skill

Analyzes data using Gemini CLI via clinvk.

## Usage
Run this skill when you need to analyze structured data.

## Script

!/bin/bash

DATA="$1"

clinvk -b gemini -o json --ephemeral "Analyze this data and provide insights: $DATA"

Using the Skill

In Claude Code, the skill can be invoked:

User: /analyze-data {"sales": [100, 150, 200], "months": ["Jan", "Feb", "Mar"]}

Multi-Model Review Skill

A more powerful skill that uses multiple backends for comprehensive code review:

<!-- ~/.claude/skills/multi-review/SKILL.md -->
# Multi-Model Code Review

Performs comprehensive code review using Claude (architecture),
Codex (performance), and Gemini (security).

## Usage
Provide a file path or code snippet for multi-perspective review.

## Script

!/bin/bash

CODE="$1"

echo "## Multi-Model Code Review Results" echo ""

echo "### Architecture Review (Claude)" clinvk -b claude --ephemeral "Review this code for architecture and design patterns: $CODE"

echo "" echo "### Performance Review (Codex)" clinvk -b codex --ephemeral "Review this code for performance issues and optimizations: $CODE"

echo "" echo "### Security Review (Gemini)" clinvk -b gemini --ephemeral "Review this code for security vulnerabilities: $CODE"

Parallel Review Skill

For faster multi-model review using parallel execution:

<!-- ~/.claude/skills/parallel-review/SKILL.md -->
# Parallel Multi-Model Review

Fast parallel code review using all backends simultaneously.

## Script

!/bin/bash

CODE="$1"

Create tasks file

cat > /tmp/review-tasks.json << EOF { "tasks": [ {"backend": "claude", "prompt": "Review for architecture and design: $CODE"}, {"backend": "codex", "prompt": "Review for performance issues: $CODE"}, {"backend": "gemini", "prompt": "Review for security vulnerabilities: $CODE"} ] } EOF

clinvk parallel -f /tmp/review-tasks.json --json | jq -r ' "## Architecture (Claude)\n" + .results[0].output + "\n\n" + "## Performance (Codex)\n" + .results[1].output + "\n\n" + "## Security (Gemini)\n" + .results[2].output '

Chain Execution Skill

A skill that pipelines output through multiple backends:

<!-- ~/.claude/skills/doc-pipeline/SKILL.md -->
# Documentation Pipeline

Generates polished documentation through a multi-step pipeline:
1. Claude analyzes code structure
2. Codex generates documentation
3. Gemini polishes and improves readability

## Script

!/bin/bash

CODE="$1"

Create pipeline file

cat > /tmp/doc-pipeline.json << EOF { "steps": [ { "name": "analyze", "backend": "claude", "prompt": "Analyze the structure and purpose of this code. List all functions, classes, and their relationships:\n$CODE" }, { "name": "document", "backend": "codex", "prompt": "Based on this analysis, generate comprehensive API documentation in Markdown format:\n{{previous}}" }, { "name": "polish", "backend": "gemini", "prompt": "Improve the readability and add helpful examples to this documentation:\n{{previous}}" } ] } EOF

clinvk chain -f /tmp/doc-pipeline.json --json | jq -r '.results[-1].output'

Advanced Patterns

Error Handling

#!/bin/bash
set -e

if ! OUTPUT=$(clinvk -b claude --ephemeral "$1" 2>&1); then
  echo "Error executing clinvk: $OUTPUT"
  exit 1
fi

echo "$OUTPUT"

Conditional Backend Selection

#!/bin/bash
TASK_TYPE="$1"
PROMPT="$2"

case "$TASK_TYPE" in
  "analyze")
    BACKEND="claude"
    ;;
  "generate")
    BACKEND="codex"
    ;;
  "research")
    BACKEND="gemini"
    ;;
  *)
    BACKEND="claude"
    ;;
esac

clinvk -b "$BACKEND" --ephemeral "$PROMPT"

Compare Backends

#!/bin/bash
# Get responses from all backends and compare
clinvk compare --all-backends "$1"

Best Practices

1. Use Ephemeral Mode

For stateless skill execution, always use --ephemeral:

clinvk -b claude --ephemeral "your prompt"

2. Choose the Right Backend

Task Type Recommended Backend Why
Code Review Claude Deep understanding, context
Code Generation Codex Optimized for code
Data Analysis Gemini Strong analytical capabilities
Documentation Any All perform well
Security Audit Claude + Gemini Different perspectives

3. Use JSON Output for Parsing

When you need to process the output:

clinvk -b claude -o json --ephemeral "..." | jq -r '.content'

4. Format Output for Claude

Structure skill output for Claude to process:

echo "## Skill Results"
echo ""
echo "### Summary"
clinvk -b gemini --ephemeral "Summarize: $INPUT"

Skill Directory Structure

~/.claude/skills/
├── analyze-data/
│   └── SKILL.md
├── multi-review/
│   └── SKILL.md
├── doc-pipeline/
│   └── SKILL.md
└── shared/
    └── clinvk-helpers.sh  # Shared functions

Troubleshooting

Backend Not Available

# Check available backends
clinvk config show | grep available

Check Version

clinvk version

Next Steps