savannahostrowski/ghost

Ghost 👻 is an experimental CLI that uses AI to generate GitHub Actions workflows, using OpenAI

Hey there👋! I'm Savannah!

A bit about me...

• 🛠 I work at Docker as a Staff Product Manager, leading product for the Docker runtime. I work on Docker engine, Docker CLI, Wasm and other OSS components.
• ✨ I previously worked on the Azure Developer CLI (azd) and before that, Pylance and Python in VS Code
• 🐍 I'm a CPython contributor!
• 👩‍💻 I love Python, Go, containers, DevOps, open source software and cats!
• 🐈 I have three cats named after Japanese foods - Nori, Miso & Momo

Upcoming conferences

• Container Plumbing Days (April 15) - attending on behalf of Docker
• Open Source Summit North America (April 16-18) - attending on behalf of Docker
• PyCon US (May 16-19)
• Microsoft Build (May 21-23) - attending on behalf of Docker

ghost's Issues

Investigate editor integrations

e.g. vscode

General clean up + refactoring

Stylize config command

Was a last minute addition, would be nice to make it look nicer

Play around with different prompts

There's probably room for improvement here

Error for maximum context length

Ran into it when running Ghost over: https://github.com/savannahostrowski/ghost-vscode

Make OpenAI model configurable

Currently using GPT3.5Turbo but it would be nice to make it configurable so that folks with access to GPT 4 can use that for improved responses.

Add a version command

Add a cobra command so that a user can run ghost version and have the version print out

Remove --config and --viper flags

Probably not needed

Refactor OpenAI API connection

Would be ideal to keep the same connection with the same context between refined prompts.

Investigate using streaming completion for GHA response

https://github.com/sashabaranov/go-openai