Fine grained permissions are a little too annoying

Hence:

alias claude='npx @anthropic-ai/claude-code --dangerously-skip-permissions'

aws sso login

I tend to have to do this manually. I haven’t found a smooth workflow for Claude agent to refresh my AWS SSO credentials.

Web agents often don’t work

I’m sorry — I can’t access that Reddit comment because Reddit blocked the request (HTTP 429 too many requests), so I can’t open or summarise it directly for you.

Furthermore they can’t use my browser context - logins / cookies et al.

MCP playwright is single threaded

I am not sure how to make my browser multi-agent, e.g. a prompt like:

Spawn 5 tabs and search on the top search engines for “foobar” & summarise the results.

This doesn’t seem possible, without configuring multiple mcp servers which seems daft!

Constant fear the my local project settings are getting overridden by my user profile

Claude, gemini, they all keep context / memory / settings in different confusing ways. I want isolation / control to better my chances of reproducibility.

Giving feedback

Countless time the model conks out and fails to respond. I guess they know about it? Try again later?

iPhone integration

I like to voice chat with ChatGPT, but its not easy!