Coding Harness

A TypeScript CLI that wraps a LLM model agent loop with file system and weather tools. Give it a prompt, it decides which tools to call, executes them, and returns a final answer.

Built with Bun + Commander.

---

Prerequisites

• Bun v1.3+
• An OpenAI API key

---

Installation

git clone https://github.com/ayeangad/Coding-Harness.git
cd Coding-Harness
bun install

---

Configuration

1. Login with your provider and API key

bun run cli.ts providers login -p openai -a <your-api-key>

This saves your provider and API key to ~/.coding-harness/config.json.

2. (Optional) Set a default provider

bun run cli.ts providers setprovider -p openai

3. (Optional) Logout

Clears the saved config.

bun run cli.ts providers logout

Note: The agent loop currently reads the API key from the OPENAI_API_KEY environment variable. If you prefer to set it manually instead of using the login command:

export OPENAI_API_KEY=your_key_here

---

Commands

agent

Runs the GPT-4o (but can run any OpenAI model) agent loop with a prompt. The agent can use tools to read/write files and fetch weather.

bun run cli.ts agent -p "read config.json and summarize it"
bun run cli.ts agent --prompt "what is the weather in Mumbai?"
bun run cli.ts agent -p "create a file called notes.txt and write 3 ideas for a startup"

Available tools:

Tool	Description
readFile	Reads a file from disk
writeFile	Writes or appends content to a file
deleteFile	Deletes a file
getCurrentTemperature	Fetches current weather for a location

---

models

Lists supported LLM models from the model registry.

bun run cli.ts models              # list all models
bun run cli.ts models -m openai    # list OpenAI models only
bun run cli.ts models -m anthropic # list Anthropic models only

---

providers

Manage your provider config. Credentials are saved to ~/.coding-harness/config.json.

bun run cli.ts providers login -p openai -a <your-api-key>   # save provider + API key
bun run cli.ts providers setprovider -p openai               # set default provider only
bun run cli.ts providers logout                              # clear saved config

---

Project Structure

cli.ts                  # entry point, registers all commands
commands/
  agent.ts              # agent command
  agent-loop.ts         # core loop: prompt → tool calls → final output
  model.ts              # models command
  providers/
    index.ts            # registers provider subcommands
    login.ts            # saves provider + API key to ~/.coding-harness/config.json
    logout.ts           # clears config
    setProvider.ts      # updates default provider in config
data/
  llm-models.ts         # model registry (OpenAI + Anthropic)

---

Stack

• Bun — runtime and package manager
• Commander — CLI framework
• OpenAI Node SDK — GPT-4o agent via client.responses.create
• TypeScript