Configuration Reference

Schema

interface Config {
  // Default model for queries
  model: 'sonnet' | 'opus' | 'haiku';

  // Maximum tokens for response
  maxTokens: number;

  // Theme variant
  theme: 'neon' | 'vibrant' | 'soft' | 'glow';

  // Context injection settings
  context: {
    git: boolean; // Include git branch, status, commits
    cwd: boolean; // Include current working directory
    lastCommand: boolean; // Include previous command output
  };

  // Additional system prompt
  systemPrompt?: string;

  // Prompt aliases
  prompts: Record<string, string>;

  // Safety settings
  safety: {
    confirmDestructive: boolean; // Confirm dangerous commands
    maxCostPerQuery: number; // USD limit per query
    maxInputSize: number; // Max input size in characters
    blockedCommands: string[]; // Patterns to never execute
  };
}

Default Values

model: sonnet
maxTokens: 4096
theme: neon

context:
  git: true
  cwd: true
  lastCommand: false

prompts: {}

safety:
  confirmDestructive: true
  maxCostPerQuery: 0.50
  maxInputSize: 100000
  blockedCommands: []

Options

model

The Claude model to use for queries.

Value	Description
sonnet	Balanced speed and quality (default)
opus	Maximum capability, slower
haiku	Fastest, best for simple queries

maxTokens

Maximum tokens in the response. Range: 1-4096.

theme

SilkCircuit color theme variant.

Value	Description
neon	Full intensity, maximum vibrancy (default)
vibrant	85% intensity
soft	70% intensity, easier on eyes
glow	Maximum contrast

context.git

When true, includes:

• Current branch name
• Clean/dirty status
• Commits ahead/behind
• Recent commit messages

context.cwd

When true, includes the current working directory path.

context.lastCommand

When true, includes the output of the previous shell command.

systemPrompt

Additional instructions appended to the default system prompt.

prompts

Aliases for common prompts. Use with @alias syntax:

q @explain "$(cat error.log)"

safety.confirmDestructive

When true, always prompts for confirmation before executing commands that match destructive patterns (rm, drop, delete, etc.).

safety.maxCostPerQuery

Maximum cost in USD allowed per query. Queries exceeding this will be stopped.

safety.maxInputSize

Maximum input size in characters (default: 100,000 or ~100KB). This limits the size of piped input or file contents to prevent excessive token usage and costs.

safety:
  maxInputSize: 200000 # Increase to 200KB for large files

safety.blockedCommands

Array of regex patterns for commands that should never be executed.