di-sukharev / ai-tdd Goto Github PK

Would be cool to make it possible in VSCode to:

1. Generate a unit test from debugger call stack or variable (everything for Arrange, Act and Assert (AAA) pattern is already there)
2. Slightly modify this test to satisfy new requirements
3. Generate code using ai-tdd based on generated unit test

I always receive the message Error: Request failed with status code 404
The api-key is correct.
When i am sending requests via Postman it is working.

i'm thinking of something like this:

1. you send a test suite filepath
2. lib saves the filepath and caches the file
3. you send the same filepath again (working on a particular test suite)
4. lib finds the difference between the cached file content and the new file content and send only diff to openAI api
5. GPT responses with a JSON something like {type: "replace", from: "some_code_from_cached_file_content", "to": "new_content"}, other types may be append (to add line after a line), insert with {startLine: <code_string>, endLine: <code_string>} (to add code between the lines)

I followed the straightforward README, and setup my config as such (API changed for obvious reason)

~/.aitdd/config

aitdd config set OPENAI_API_KEY sk-ManyLettersWithMixedUpperCaseAndLowerCase
aitdd config set RUN_TESTS "npm run test"

When I try to run it in the console:

➜  TiedSiren git:(main) aitdd run
│
└  'aitdd config set OPENAI_API_KEY sk-ManyLettersWithMixedUpperCaseAndLowerCase' name is invalid, it should be either 'AITDD CONFIG SET OPENAI_API_KEY SK-MANYLETTERSWITHMIXEDUPPERCASEANDLOWERCASE' or it doesn't exist.

│
└  Manually fix the '.env' file or global '~/.aitdd/config' config file.

I am not sure where I got it wrong. How can I fix it?

For non-trivial / implementations where they could be interpreted in many different ways, GPT4 will still miss in a zero-shot attempt. Perhaps you could run the test file and feed the results into a second execution of aitdd. Recursive AI coding—the dream!