Stop fighting the Start9 build toolchain — it’s changed too many times. The clean solution is much simpler. "The real fix** — the dashboard now works on Start9 Pages without any build system at all:
I have build an html dashboard and I’m trying to fix CORS limits direct fetch to run properly over Tor on my start9 OS. The page works fine with Start9Pages but I’d like to sideload directly so as to fight the CORS-safe mempool endpoints.
Claude
The real solution for Start9/Tor is completely different from fighting CORS — the right architecture is a server-side proxy. The nginx container running on Start9 can proxy all API calls server-side, which means no browser CORS at all. The browser talks to your own Start9 nginx, which forwards requests to mempool.space from the server. Server-to-server calls are never subject to CORS.
So I have one problem: start-cli:
The error - start-cli package
error: ‘start-cli package’ requires a subcommand but one was not provided
[subcommands: action, attach, backup, cancel-install, host, install, installed-version, list, logs, rebuild, restart, set-outbound-gateway, start, stats, stop, uninstall, help]
Usage: start-cli package
For more information, try ‘–help’.
make: *** [btc-dashboard.s9pk] Error 2
What is the sub-command to build or create the s9pk file?
Yes, I think start-cli is only for 0.4.0. For 0.3.5.1 I always used the start-sdk. The docs are here. Might be good to point Claude to them and have it distill the important details to a document that it can reference.
Anyway, the specific command is “start-sdk pack” (though typically that is built into your makefile, and you just run “make” to build the .s9pk).
Thank you sir. After correcting my error I got the thing working properly.
Like I said the reason I wanted to sideload was to overcome the CORS-safe functions when running in Pages as http or in local mode. Sideload fixed that and now the functions work properly with https
This did work well actually but I’m now acutely aware of how Claude tokens are used so I need to re-adjust my thinking so I don’t use up tokens on wasteful/repetitive work and hit session limits. The /context command was an eye opener to be sure. Tokens are being used just to open a chat…you don’t have to type anything!
Also /review and /ultrareview defiantly saves me a lot now. Before I just kept telling Claude the output wasn’t working properly so the entire project was completely rebuilding itself every time eating tokens like a Monster.
This is a huge issue, and only going to get worse in the foreseeable future (all of the big providers are currently operating at a loss and will eventually need to jack prices much higher or risk going bankrupt waiting for more efficient models and hardware).
If you use LLMs heavily for dev work like I do (which you really have to if you don’t want to be left behind in the dust by devs who do use them), you end up learning when to reach for local models, when to switch down to less-capable cloud models, and when to reach for opus, in order to maximize your velocity with your budget. And another important skill (which is extremely difficult when you are on a roll) is knowing when to stop and wait for your token allotment to reset (and resist the urge to buy more tokens just to finish now). I think being a bitcoiner helps a bit with that last point, since Bitcoin is constatnly teaching us to lower our time preference.
