Tddbudy

This is an experiment in AI-assisted test-driven development.

User Stories

• As a developer, I want to write unit tests without writing any SUT code, so the AI can generate the implementation for me. (The SUT is defined implicitly by the assertion; the AI infers the intended behavior.)
• As a developer, I want the tool to automatically retry code generation when the AI produces non-compiling code or failing tests.
• As a developer, I want to configure a maximum number of retries per prompt to avoid infinite loops.
• As a developer, I want to see the build output and test results after each generation attempt.
• As a developer, I want a lightweight CLI that I can run locally.
• As a developer, I want the tool to be model-agnostic — compatible with OpenAI, Mistral, LLaMA, or any other provider.
• As a developer, I want each generation attempt archived (prompt, AI response, result) for traceability and debugging.

Desired Behavior

The CLI should accept the following inputs:

• A specs.swift file containing the test cases.
• An output filename for the generated implementation.
• A maximum number of allowed iterations (for retrying failed generations).

Proposed System Flow:

System.generateCodeFrom(specs) → (GeneratedCode, Stdout/Stderr)
    → LLM.send(specs) → GeneratedCode
    → Concatenator.concatenate(GeneratedCode, Specs) → Concatenated
    → CodeRunner.run(Concatenated) → Stdout/Stderr
    → Exit

More control is needed (input & output urls, max iteration count), so we'll probably need more components and a higher level coordinator:

Usage

Define your specs in a swift folder without implementing the sut.

//specs.swift
func test_adder() {
    let sut = Adder(a: 1, b: 2)
    assert(sut.result == 3)
}

Pass the path of the file to TestBuddy:

tddbuddy \
  --input adder.swift \
  --output adder-generated.swift \
  --iterations 5