Why Use a Skill?
Skills are the recommended way to give Claude Code access to Paragon commands. Unlike adding to CLAUDE.md, skills are:
- On-demand - Only loaded when you invoke
/paragon
- Context-efficient - Doesn’t consume tokens until needed
- Organized - Keeps your CLAUDE.md clean for project-specific info
Quick Setup
Option 1: Let Claude Create It
Run Claude Code in your project and paste this prompt:
Create a skill file at .claude/skills/paragon.md with the Paragon CLI reference.
Include all commands: paragon run, paragon context, paragon submit, paragon branch,
and stack navigation (sync, next, prev, top, bottom). Include flags and examples.
Claude will create the skill file automatically.
Option 2: Manual Setup
- Create the directory:
mkdir -p .claude/skills
- Create
.claude/skills/paragon.md
- Copy the CLI reference below into the file
Usage
Once the skill is created, type /paragon in Claude Code to load the Paragon CLI reference into context. Claude will then know how to use all Paragon commands.
CLI Reference
Copy the content below into .claude/skills/paragon.md
.claude/skills/paragon.md
# Paragon CLI Commands
> Use Paragon CLI for enhanced code analysis, reviews, and git workflows.
## Quick Start
paragon run -q "Review this code for security issues" # Analyze code with a prompt
git diff | paragon run -q "Review these changes" # Pipe code for analysis
## Code Analysis Commands
### paragon run
Run a prompt in non-interactive mode. Accepts input via arguments or stdin.
paragon run "Explain the authentication flow in this codebase"
paragon run -q "Find potential bugs in the error handling" # Quiet mode (hide spinner)
paragon run --feature "Add input validation to the API endpoints" # Feature implementation mode
paragon run --json "List all security issues" # JSON output for structured parsing
cat src/auth.go | paragon run "Explain this code" # Pipe stdin
git diff HEAD~3 | paragon run -q "Review these changes"
paragon run -m paragon-max "Complex architectural analysis" # Select model
Flags:
-q, --quiet Hide spinner/progress output
--feature Use feature implementation mode (vs code review)
--json Output structured JSON findings
--concise Results-only, no reasoning
-m, --model Model selection (paragon-max, paragon-high, paragon-fast)
### paragon context
Specialized context-aware queries using semantic search and AI analysis.
paragon context issue "Database updates not persisting" # Diagnose and find root causes
paragon context search "authentication logic" # Semantic code search
paragon context knowledge "How does the caching layer work?" # Query project architecture
## Git Workflow Commands
### paragon submit
Push branch and create/update GitHub PR with AI-generated description.
paragon submit --ai # AI-generated title and description
paragon submit --title "Add user authentication" # With custom title
paragon submit --draft --ai # As draft PR
### paragon branch create
Create a new stacked branch with optional AI naming.
paragon branch create --ai # AI-generated branch name based on changes
paragon branch create feature/auth # Explicit name
### Stack Navigation
paragon stack # Show branch stack tree
paragon sync # Rebase current branch on parent
paragon next # Move to child branch
paragon prev # Move to parent branch
paragon top # Jump to top of stack
paragon bottom # Jump to bottom of stack
## Global Flags
Available on all commands:
-q, --quiet Hide spinners and progress output
-y, --yolo Auto-approve all permission requests (dangerous)
-d, --debug Enable debug logging
-c, --cwd Set working directory
## Integration Examples
git diff main | paragon run -q "Review for bugs and security issues" # Code Review on Diff
gh pr diff 123 | paragon run -q --json "Comprehensive code review" # Automated PR Review
paragon context issue "Tests failing after recent merge" # Find Root Cause
git diff --staged | paragon run -q "Check for issues before commit" # Pre-commit Analysis
cat src/complex-algorithm.go | paragon run "Explain this step by step" # Explain Complex Code
## Tips
1. Use -q flag for cleaner output in scripts
2. Pipe code via stdin for targeted analysis
3. Use --json when parsing results programmatically
4. Use context issue for debugging problems
5. Use context search to find code patterns
Why Skills Over CLAUDE.md?
| CLAUDE.md | Skills |
|---|
| Always loaded into context | Loaded on-demand with /paragon |
| Uses tokens on every message | Only uses tokens when invoked |
| Best for essential project info | Best for reference documentation |
| Can get cluttered | Organized by topic |
Keep your CLAUDE.md focused on project-specific context (architecture, conventions, important files). Use skills for tool references like Paragon that you only need occasionally.