Hum... this is new: "Your account has exceeded GitHub's API rate limit. Please try again later."
English
Marc-André Moreau
34.8K posts
Marc-André Moreau
@awakecoding
Remote desktop protocol expert, OSS contributor and Microsoft MVP. I love designing products with Rust, C# and PowerShell. Proud to be CTO at Devolutions. 🇨🇦
Lavaltrie, Quebec, Canada Katılım Haziran 2010
1.9K Takip Edilen5.9K Takipçiler
Marc-André Moreau retweetledi
Introducing Response Caching: save tons of money and time on tests and agent retries.
Blog post: openrouter.ai/announcements/…
Available for free. Learn more 👇
English
@davidfowl It looks great, now I guess the only missing thing would be a great agent skill to provide instructions on how to use the tool properly?
English
A CLI for helping agents do performance analysis on .NET projects github.com/adityamandalee…
#dotnet #performance
English
@techspence @fr0gger_ I second this, Delta-V is a must read
English
Project Hail Mary and Dark Matter from this list are amazing!! If you are into sci-fi highly recommend them!
BookNote@BookNoteApp
8 fiction books more addicting than scrolling: 1) Project Hail Mary by Andy Weir
English
Here's an idea: call the Copilot CLI or OpenCode CLI with an injected MCP configuration + test prompt, then evaluate the result from the "parent" LLM and see if it fits the expected outcome
Marc-André Moreau@awakecoding
How does one perform end-to-end testing of an MCP server? I mean with proper evals to see if sample prompts end up triggering the tools you'd expect, and with an expected non-deterministic outcome from the LLM that calls it
English
@grhmc @0xSiddhesh @glcst @tursodatabase Every second can have 80 bits of random blueberry muffins that you can then sort easily by baking time
English
@0xSiddhesh @awakecoding @glcst @tursodatabase How would you make blueberry muffins in this architecture?
English
UUIDv4 provides ~2^122 random values; even generating 1 billion per second nonstop would take ~85 years for a 50% chance of just one collision, per standard probability calculations...
Glauber Costa@glcst
what if we run out of uuids? I'm concerned.
English
@0xSiddhesh @grhmc @glcst @tursodatabase ULIDs use 48 bits for a UNIX timestamp in seconds, so collision checking only needs to be done for the same second. It would be easy to prune old ULIDs out of the possible collision window
English
@awakecoding @grhmc @glcst @tursodatabase Interesting take on ULIDs + short-window checks! time-sortability would actually be a bonus for db creation logs too.
For the collision window, how short are we thinking — secs or mins? A tiny in-memory or Turso table that auto-prunes old entries could keep overhead minimal tbh
English
@grhmc @glcst @0xSiddhesh @tursodatabase Yeah the more I think about it, the more ULID + collision checking at database creation time would be a neat way to ensure uniqueness of database ids even between customers, without having to check against *all* the database IDs every time: github.com/ulid/spec
English
@grhmc @glcst @0xSiddhesh @tursodatabase The ULID increases the possibility of collisions when generating at the *exact* same time, but if you do a collision check with a Turso database, this could work. The database wouldn't even need to hold all the ULIDs, just a short time window for collision checking
English
@PrzemyslawKlys Report this feedback on the forum, someone from the right team will take a look at it: forum.devolutions.net
English
@awakecoding i tried to use RDM on ipad without keyboard and its pretty much unusable. Windows app without keyboard was hard, but RDM was just impossible to use. Simple type arrow up and press enter in codex, 10 steps. Hope the interface could be improved for finger management. Right click, scroll was really hard.
English
@glcst @0xSiddhesh @tursodatabase Maybe it's one of those problems that is unlikely to happen, but if it happens, you'll be happy that it needs to be solved in the first place because of what it means? If it happens, set a UUID for each customer, and assume database UUIDs are no longer unique between customers
English
@awakecoding @0xSiddhesh @tursodatabase we don't encode it in the uuid. We just want to create 10 zillion databases !
unlimited databases per customer
English
@glcst @0xSiddhesh @tursodatabase Why does it need to be encoded in the UUID that it is a Turso database? Can't it just be one of the zillion non-conflicting UUIDs?
English
@0xSiddhesh ok but we want to use them to identify @tursodatabase cloud databases. There are too many. Each customer can create unlimited databases
English
@davidfowl @ibuildthecloud @browser_use The truly unresolved problem right now is a decent CDP for RDP, but for Computer Use, not Browser Use. There are many UI-framework specific solutions, but one that could adapt well to many with fallbacks to generic accessibility APIs would do. Windows desktop apps need this
English
@davidfowl @ibuildthecloud @browser_use Is it still for an actual browser, or is the suggestion here to stream CDP from browsers in sandboxed RDP sessions? If CDP can be adapted to go through named pipes or hacked together with socat to go through anything, it could be adapted to go over an RDP virtual channel
English
What you need is CDP as a Service. @browser_use is that what y'all are doing? I'm serious. An API to get a CDP connection. Then additionally get a VNC or RDP (RDP way better) to get an interactive view of the browser. If you have this, I can do a cool integration. If you don't have it, make it now please. cc @awakecoding
English
How does one perform end-to-end testing of an MCP server? I mean with proper evals to see if sample prompts end up triggering the tools you'd expect, and with an expected non-deterministic outcome from the LLM that calls it
English
@unixterminal But even for the SDKs, I wonder if there isn't a legal-friendly way to make it work - I'm thinking the equivalent of the MinGW Windows compatibility headers that are sufficient for linking programs, but have all the headers stripped down to just the interface definitions
English
@unixterminal There are only two command-line tools I don't have: ibtool and actool. GitHub Copilot still managed to make a mock in bash that doesn't really do the work, but still fools .NET into doing the build, which is enough for quick C# build testing
English
Why is it so hard to cross-compile .NET from Linux to macOS? "dotnet workload install macos" only works on a macOS host for me. I've got native replacements for most Apple build tools running natively on Linux, and I've got a Frankenstein partial build working with lots of hacks
English
@DenelonMs @lemielek That's what I figured, and it makes sense, no worries. It'll just go into the next build, I was one day late 🙂
English
@awakecoding @lemielek We push a "feature team" release to the team on Monday morning (typically). If it doesn't blow up on anybody it will move to GitHub the next week.
English
A new WinGet Preview with lots of goodness:
github.com/microsoft/wing…
English