runekaagaard

OK thanks! Would love to read :)

What ... how ...? Just bought two yesterday for playing with local llm but thought 3.8 was totally out of reach!?

/sandbox AFAIK uses https://github.com/anthropic-experimental/sandbox-runtime under the hood.

It's still experimental and if you dive into the issues I would call its protection light. Many users experiences erratic issues with perms not being enforced, etc.

For me the largest limitation was that it's read-mode is deny-only, meaning that with an empty deny-list it can read all files on your laptop.

Restricting to specific domains have worked fine for me, but it can't block on specific ports, so you can't say for instance you may access these dev-server ports, but not dev-server ports belonging to another sandbox.

It feels as though the primary usecase is running inside an already network and filesystem sandboxed container.

... also interested. What would one build an Emacs TRAMP plugin for? :)

Configuring Claude Code ... the new init.el ;)

It's impossible to not get decision-fatique and just mash enter anyway after a couple of months with Claude not messing anything important up, so a sandboxed approach in YOLO mode feels much safer.

It takes the stress about needing to monitor all the agents all the time too, which is great and creates incentives to learn how to build longer tasks for CC with more feedback loops.

I'm on Ubuntu 22.04 and it was surprisingly pleasant to create a layered sandbox approach with bubblewrap and Landlock LSM: Landlock for filesystem restrictions (deny-first, only whitelisted paths accessible) and TCP port control (API, git, local dev servers), bubblewrap for mount namespace isolation (/tmp per-project, hiding secrets), and dnsmasq for DNS whitelisting (only essential domains resolve - everything else gets NXDOMAIN).

I like Claude Code in the terminal. For me it's so good it don't need IDE integration. I'm just using emacs and magit to navigate the code out of band.

Thats me to a tee. I also don't have a inner dialogue when thinking through a problem:)

Yeah, totally this. I've had so much fun with AoC, learning nim, elixir at the same time.

I would normally tap out around the same place on the first dynamic programming puzzle which just takes me so long to wrap my head around each time (tips anyone? :)).

I welcome these new changes, and what ever the format are very greatful for all his hard work!

I agree. The industry standard for a great, boring, durable and surprisingly cheap knife is the Victorinox Fibrox Chef's Knife 20 cm.

Nope, I've really, really tried to like modal editing, because the programmable command chaining is super cool, but even though I became proficient with it I never really enjoyed it.

Starting out emacs i got super fatigued with all the long pinky driven commands for mostly used commands. It felt usable after I added keybindings for commands like switch buffer, close buffer, duplicate line(s), move line(s), find in project, find file in project, indent (wrote my own sane (for me)) indention code). The windows/apple key is great for those things because they are not used by emacs.

On linux I settled on using emacs vanilla key commands for copy/paste/cut but that took a looong time to feel comfortable with and I still mess it up sometimes, also with the ctrl+shift-X version of them in the terminal. On iOS, using the apple key like for the rest of the system is sweet relief.

Heh - yeah have had trillion dollar ideas many times :)

Luckily i do, but i mean it triggers the api limit in 10 minutes amount of tokens

My sweet spot at the moment is Claude Desktop with mcp servers for editing and aider --watch for quick fixes. Claude Code uses way, way, way too many tokens on the large project i work most on.

Yeah remember when people were using Claude 3.7... so oldschool man

Heh. I'm giving Claude running on AWS Bedrock in a EU datacenter access to read small parts of my email (normally 1-3 email threads in a chat), compose drafts for approval and then send them in a separate step. I can read and approve all tool calls before they are executed.

Yeah, I too found giving LLMs access to my emails via notmuch [1] is super helpful. Connecting peripheral sources like email and Redmine while coding creates a compounding effect on LLM quality.

Enterprise OAuth2 is a pain though - makes sending/receiving email complicated and setup takes forever [2].

- [1] https://github.com/runekaagaard/mcp-notmuch-sendmail

- [2] https://github.com/simonrob/email-oauth2-proxy

I thinks a lot is timing and also that it's a pretty low bar to write your first mcp server:

    from mcp.server.fastmcp import FastMCP
    mcp = FastMCP("Basic Math Server")

    @mcp.tool()
    def multiply(a: int, b: int) -> int:
        return a * b

    mcp.run()

If you have a large MCP server with many tools the amount of text sent to the LLM can be significant too. I've found that Claude works great with an OpenAPI spec if you provide it with a way to look up details for individual paths and a custom message that explains the basics. For instance https://github.com/runekaagaard/mcp-redmine

It feels a bit like "cheating" that new x-string features are built-in only. It would be cool to be able to do:

    from foo import bar
    bar"zoop"

I'm getting great and stable results with 3.7 on Claude desktop and mcp servers.

It feels like an upgrade from 3.5

I've used it a lot. Great product, emulates subtle device issues very well that Chrome device view doesn't

Yeah, Claude can query and read the result sets. It does not send the entire database.

I use this every day: https://github.com/runekaagaard/mcp-alchemy

I'm often wondering about if an elixir-like language for Ada could make it more popular.

We've rolled our own liveview and have happily used it in production for several years now: https://hypergen.it/

Rust's borrow checker?

Looks very good, thx for sharing!

Can it be integrated with an existing large nginx config with multiple domains, server and client certificates, websockets, other custom settings and different apps deployed with ansible or does it need to run the nginx process by itself?

Beauty!