Hello HN!
As part of our work at Kenobi[0], I used Codex to analyse an HTTP archive (HAR) for a website I was trying to see if I could use programatically, i.e. by exposing its underlying / undocumented API. That approach worked much better than I expected, so I started looking for a way to do this easily for other websites.
I couldn't find exactly what I wanted, so we decided to build Harpist -- a product in two parts: i. A Chrome extension, that allows you to record HAR files and view generated documentation; and ii. A CLI that can be used by agents to refine these HAR files and create oRPC contracts, documentation, and allow you to replay HTTP requests.
There's an agent skill that lets you use Claude or Codex to manage all of the refinement and CLI usage, so right now the workflow is as follows:
-
Open the Chrome extension and record yourself using a website (the parts you'd want to automate).
-
Tell your agent to "use Harpist to refine my latest recording".
-
When it's done, the extension lets you view documentation for the discovered endpoints.
-
Replay the endpoints and view captured auth data in the CLI yourself, or just...
-
Ask your agent to do something for you on one of the website's you've captured and processed using Harpist.
It's all open source (MIT licensed)[1], and under the hood it uses oRPC to create TypeScript-compatible API contracts, but I've mostly been focussing on the documentation for now which just uses the oRPC Scalar integration[2] to generate OpenAPI-looking docs.
I'm keen to learn what other people are doing to discover and automate web workflows at the moment, and what kind of things people wish existed too.
[0] - [https://kenobi.ai](https://kenobi.ai)
[1] - [https://github.com/kenobi-ai/harpist](https://github.com/kenobi-ai/harpist)
[2] - [http://orpc.dev/docs/openapi/scalar](http://orpc.dev/docs/openapi/scalar)
Comments URL: [https://news.ycombinator.com/item?id=48893610](https://news.ycombinator.com/item?id=48893610)
Points: 2