warden

Main command for running code analysis with AI-powered skills

warden

The main Warden command runs code analysis on your repository using AI-powered skills. It can analyze specific files, git changes, or use triggers configured in warden.toml.

Usage

warden [targets...] [options]

Targets

Warden accepts multiple target types to specify what code to analyze:

files · string

Specific file paths to analyze

warden src/auth.ts
warden src/auth.ts src/utils.ts

globs · string

Glob patterns to match multiple files (must be quoted)

warden "src/**/*.ts"
warden "src/**/*.{ts,tsx}"

git-ref · string

Git references to analyze changes since a commit or between refs

warden HEAD~3           # Changes in last 3 commits
warden main..feature    # Changes between branches
warden HEAD~1..HEAD     # Most recent commit

(none) · -

When no targets are provided, Warden uses triggers from warden.toml to analyze uncommitted changes

warden  # Analyze uncommitted changes

For ambiguous targets (no path separator, no extension), Warden checks if a file exists at that path before treating it as a git ref. Use --git to force git ref interpretation.

Options

Skill Selection

--skill · string

Run only the specified skill instead of all configured skills

warden --skill security-review
warden src/auth.ts --skill bug-detection

Configuration

--config · string · default: "./warden.toml"

Path to the Warden configuration file

warden --config .warden/custom.toml

-m, --model · string

Model to use for analysis (fallback when not set in config)

warden --model claude-sonnet-4.5
warden -m claude-opus-4

Output Control

--json · boolean · default: "false"

Output results in JSON format instead of human-readable format

warden --json
warden --json > results.json

-o, --output · string

Write full run output to a JSONL file for later replay

warden -o run.jsonl
warden --output .warden/logs/$(date +%s).jsonl

Filtering

--fail-on · string

Exit with code 1 if findings are at or above the specified severity level

Values: off, critical, high, medium, low, info

warden --fail-on high      # Fail on high or critical
warden --fail-on medium    # Fail on medium, high, or critical

--report-on · string

Only show findings at or above the specified severity in output

Values: off, critical, high, medium, low, info

warden --report-on high    # Only show high and critical

--min-confidence · string · default: "medium"

Only show findings at or above the specified confidence level

Values: off, high, medium, low

warden --min-confidence high  # Only high confidence findings

Execution Control

--fix · boolean · default: "false"

Automatically apply all suggested fixes without prompting

warden --fix

--parallel · number · default: "4"

Maximum number of concurrent trigger/skill executions

warden --parallel 8
warden --parallel 1  # Sequential execution

-x, --fail-fast · boolean · default: "false"

Stop execution after the first finding is detected

warden --fail-fast
warden -x

Git Integration

--staged · boolean · default: "false"

Analyze only staged changes (git diff —cached)

warden --staged

--git · boolean · default: "false"

Force ambiguous targets to be interpreted as git refs instead of files

warden --git main  # Treat 'main' as branch, not file

--offline · boolean · default: "false"

Use cached remote skills without attempting network access

warden --offline

Verbosity

--quiet · boolean · default: "false"

Show errors and final summary only

warden --quiet

-v, --verbose · flag

Show real-time findings and analysis details. Use -vv for debug information including token counts and latencies

warden -v      # Verbose output
warden -vv     # Debug output

--debug · boolean · default: "false"

Enable debug output (equivalent to -vv)

warden --debug

--log · boolean · default: "false"

Use log output mode (no animations, timestamped lines)

warden --log

--color / --no-color · boolean

Force color output on or off (overrides auto-detection)

warden --color     # Force color
warden --no-color  # Disable color

Help

--help, -h · boolean

Show help message with all commands and options

warden --help
warden -h

--version, -V · boolean

Show version number

warden --version
warden -V

Exit Codes

0 · Success

Analysis completed successfully with no failures

1 · Error

• Configuration error
• Invalid arguments
• Skill execution error
• Findings met --fail-on threshold

130 · Interrupted

User interrupted execution with Ctrl+C

Examples

Analyze Specific Files

# Single file
warden src/auth.ts

# Multiple files
warden src/auth.ts src/api/endpoints.ts

# With specific skill
warden src/auth.ts --skill security-review

Analyze with Glob Patterns

# All TypeScript files in src/
warden "src/**/*.ts"

# TypeScript and TSX files
warden "src/**/*.{ts,tsx}"

# Exclude tests
warden "src/**/*.ts" "!**/*.test.ts"

Analyze Git Changes

# Changes in last 3 commits
warden HEAD~3

# Changes between branches
warden main..feature

# Only staged changes
warden --staged

# Uncommitted changes
warden

Control Output

# JSON output
warden --json

# Save to JSONL file
warden -o results.jsonl

# Quiet mode (summary only)
warden --quiet

# Verbose with debug info
warden -vv

Fail on Severity

# Fail CI if high or critical findings
warden --fail-on high

# Show only critical findings
warden --report-on critical

# High confidence findings only
warden --min-confidence high

Automatic Fixes

# Apply all fixes automatically
warden --fix

# Interactive fix mode (default when fixes available)
warden

Remote Skills

# Use cached skills only (no network)
warden --offline

# Specify model
warden --model claude-sonnet-4.5

Environment Variables

WARDEN_ANTHROPIC_API_KEY · string · required

Anthropic API key for Claude model access. Can be set in .env.local

WARDEN_MODEL · string

Default model to use when not specified in config or CLI

WARDEN_SENTRY_DSN · string

Sentry DSN for error tracking (optional)

Notes

• Warden loads environment variables from .env and .env.local in the repository root
• When no targets are specified, Warden uses triggers defined in warden.toml
• Skills are discovered from .agents/skills/ and .claude/skills/ directories
• Logs are automatically saved to .warden/logs/ when a run completes
• Use warden logs to view saved runs

Related Commands

• warden init - Initialize Warden configuration
• warden add - Add skills to configuration
• warden logs - View saved run logs
• warden sync - Update remote skills