Skills

Skills are agents that analyze your pull requests. Each skill has a specific purpose, a prompt that guides the analysis, and restrictions on what tools it can use.

Built-in Skills

security-review

Scans code changes for security vulnerabilities. This skill analyzes diffs for common security issues and reports findings with severity levels.

What It Checks

Injection Vulnerabilities — SQL injection, command injection, XSS, template injection, path traversal
Authentication & Authorization — Missing auth checks, improper session handling, authorization bypass, hardcoded credentials
Data Security — Sensitive data exposure (PII, API keys), insecure storage, missing encryption, logging secrets
Dependencies — Known vulnerable packages, insecure configurations
General Security — Insecure crypto, race conditions, information disclosure, missing input validation

Severity Levels

Level	Description
`critical`	Actively exploitable, high impact vulnerability
`high`	Exploitable with moderate effort
`medium`	Potential vulnerability, needs review
`low`	Minor security concern
`info`	Security-related observation

Usage

warden.toml

[[triggers]]
name = "Security Review"
event = "pull_request"
actions = ["opened", "synchronize"]
skill = "security-review"

CLI

# Run security review on uncommitted changes
npx warden --skill security-review

# Run on specific files
npx warden src/auth.ts --skill security-review

code-simplifier

Identifies opportunities to reduce code complexity. This skill analyzes diffs for patterns that could be simplified, refactored, or made more readable.

What It Checks

Nested Logic — Deeply nested conditionals, complex ternaries, callback pyramids
Redundant Code — Duplicate logic, unnecessary null checks, dead code
Long Functions — Functions that do too much, hard to test or maintain
Complex Expressions — Hard-to-read boolean logic, magic numbers, unclear naming
Modern Patterns — Opportunities to use optional chaining, nullish coalescing, destructuring

Severity Levels

Level	Description
`high`	Significant complexity that hurts maintainability
`medium`	Moderate complexity worth addressing
`low`	Minor improvements for readability
`info`	Style suggestions

Usage

warden.toml

[[triggers]]
name = "Code Simplifier"
event = "pull_request"
actions = ["opened", "synchronize"]
skill = "code-simplifier"

CLI

# Run code simplifier on uncommitted changes
npx warden --skill code-simplifier

# Run and auto-fix suggestions
npx warden --skill code-simplifier --fix

Custom Skills

Define your own skills in .warden/skills/ as TOML files.

Skill Definition

.warden/skills/code-review.toml

name = "code-review"
description = "General code quality review"

prompt = """
You are a code reviewer. Analyze the pull request for:
- Code clarity and readability
- Potential bugs or logic errors
- Performance concerns
- Best practices violations

Provide constructive feedback with specific suggestions.
"""

[tools]
allowed = ["Read", "Grep", "Glob"]
denied = ["Write", "Edit", "Bash"]

Skill Fields

Field	Required	Description
`name`	Yes	Unique identifier (kebab-case)
`description`	Yes	Human-readable purpose
`prompt`	Yes	System instructions for the agent
`tools.allowed`	No	Tools the skill can use
`tools.denied`	No	Tools the skill cannot use

Available Tools

Skills can use Claude Code's built-in tools:

Tool	Description
`Read`	Read file contents
`Grep`	Search file contents with regex
`Glob`	Find files by pattern
`WebFetch`	Fetch content from URLs (e.g., CVE databases)
`Write`	Write files (usually denied for review skills)
`Edit`	Edit files (usually denied for review skills)
`Bash`	Run shell commands (usually denied for security)

Using Custom Skills

Reference your custom skill by name in triggers:

warden.toml

[[triggers]]
name = "Code Review"
event = "pull_request"
actions = ["opened"]
skill = "code-review"  # Matches .warden/skills/code-review.toml

Output Format

Skills return findings in a structured format. Each finding includes:

Severity — How serious the issue is
Title — Brief description of the issue
Description — Detailed explanation
Location — File path and line numbers
Suggested Fix — Optional code suggestion

Warden translates these findings into GitHub PR reviews with inline comments on the relevant lines.