Kilo

Kilo CLI Coding Assistant - Token Efficient

Use when you need to delegate coding tasks to Kilo CLI, an autonomous AI coding assistant that works in the terminal. Provides step-by-step guidance for setup, configuration, and usage with minimal token overhead.

Trigger Conditions (Token-Optimized)

• User wants to use an AI coding assistant in the terminal/command line
• User specifically mentions "Kilo" or references the Kilo CLI
• User needs help configuring API keys for AI coding assistants
• User wants to run autonomous coding tasks from the command line

Setup and Configuration (Minimal)

1. Check if Kilo is Installed

kilo --version

2. Configure API Credentials (One-Time)

Kilo requires API credentials to access AI models. Use environment variables for token efficiency:

# Set API key once (avoids repeated prompts)
export KILO_API_KEY="sk-..."
# Or configure in ~/.config/kilo/kilo.jsonc

3. Verify Configuration

kilo auth list

Usage Patterns (Token-Saving)

Interactive Mode (Use Sparingly)

Start Kilo in interactive mode only when multi-turn conversation is needed:

kilo

With a Specific Prompt (PREFERRED - One-Shot)

Run Kilo with a direct instruction - single call, gets result:

kilo "Create a Python script that calculates Fibonacci numbers"

Continue Previous Session (When Needed)

kilo --continue

Specify Model (Token-Efficient)

kilo --model anthropic/claude-3-5-sonnet-20241022 "Explain quantum computing"

Work with Specific Project

kilo ./my-project "Refactor this code to use dependency injection"

Common Workflows (Token-Efficient)

Code Generation (One-Shot)

1. Navigate to project directory
2. Run: kilo "Create a REST API endpoint for user management"
3. Review generated code (single output)
4. Iterate with follow-up ONLY if needed

Debugging Assistance (One-Shot)

1. Run: kilo "Why is this function returning null when input is valid?"
2. Share relevant code snippets when prompted
3. Implement suggested fixes

Code Review (One-Shot)

1. Run: kilo "Review this code for security vulnerabilities and performance issues"
2. Implement feedback if valid

Token Optimization Tips

1. One-shot is preferred - Most tasks complete in a single Kilo call
2. Specify model explicitly - Use cheaper models for simple tasks
3. Limit scope - Be specific in prompts to reduce token usage
4. Cache successful prompts - Repeat patterns can reuse previous results
5. Avoid interactive mode unless truly needed - each turn adds overhead
6. Use --allowedTools equivalent - Though Kilo doesn't have this flag, being specific in prompts serves the same purpose