Recently got an reminder from Antropic, that my tokens will be expired soon.
Your organization [organizationName] has $19.48 Anthropic API credits that will expire on July 06, 2025 UTC.
I completely forgot about this part, and as I use claude console daily, still the the usage not really significant. I planned to try claude code, this was a good push to try it.
As you can see, its a significant signals in the graph of usage, as you see claude mostly use sonnet and a little bit of haiku I assume to optimize boilerplate read, and do some initial context scanning.
You basically can control claude what is allowed to read and what is not, I see some example that its declared in initial CLAUDE.md, with all other instructions and project description, and there is another way to add .claude/settings.json permission
{
"permissions": {
"deny": [
"Read(drafts)",
"Edit(drafts)",
"Read(.next/**)",
"Edit(.next/**)",
"Read(.wrangler/**)",
"Edit(.wrangler/**)"
]
}
}
There is mixed feedback yet, is it good or not, to block some access, probably need to more experiment, as I understand claude also have access to .node_modules, and its ok to go and understand how some dependency work, it may load context with some "unnecessary" information, and burn significant amount of token.
Interesting behaviour appear when in project may be some error, and you push it to the problem, and claude cannot solve it, he try to go to denied directories.
I havent expect to bump into some limitation, but here it is, I added to my project some research from different chats, openai, gemini and claude, and gather information into stripe-chats.md for the future summarization and analysis, and while claude read the project context this error appear. I havent go deeper to this problem to understand, and ask claude to ignore this file.
One of my first biggest session cost $4.63
The project already ongoing, its on MVP stage currently, without too much of the features, yet but with production consideration of core one, main tech stack is Next.js and Supabase. Its a first time I run claude in the project, I havent set any significant limitation while initialize CLAUDE.md, as a first feature that we started to work was already in progress one, payment integration, claude is helpful but still require a lot of provision, when the feature rely on different APIs, and has a side effects.
As I final, when I spent all my tokens that are had to be experied, I bought a monthly subscription. I like to have such a tool in my daily development process and envirnoment, but it is still a one of the tool, I cannot replace most of the coding task with it yet, its still require verification during the process.
The next part is test YOLO mode for the entire project :) Im sure it will be fun, only blocker may be token limitation, may check Max plan thought.