I use ChatGPT on the web to prompt codex because there are separate usage limits. I also prompt Claude Code the same way. It has better searching capabilities.
I want to give the web chat the thread, and copy-pasting doesn't work. I sometimes want:
1. prompts + thinking summaries
2. full detail w/ command outputs + diffs + subagents
Codex and Claude in their harnesses also want this just as much as I do for web.
I built an app where I can copy to clipboard the entire thread or just the last n messages, and then paste that so the web gets the full context. Via a local MCP Claude can see this same information from codex and vice versa.
It works, but the interface app for web isn't packaged yet because I am working on lots of projects and need to prioritise time and tokens. It does work rn for your harness agents in the linkedin repo.
Free and Apache 2.0, the interface would be too if people wanted it, I just don't want to go through the effort if no-one would find it useful.
Please let me know!
P.S. I've been working on much more interesting applications of using interfaces to make the complexity of formally checked software and maths more legible, but that's out of scope, it's just why I'm rationing my attention rn.
Here's a 5 min intro to that: https://www.youtube.com/watch?v=MBmVMMogXTA
Thanks!