The harder problem is conflicting design choices, or duplicating similar infra. It means I need to be much more involved in steering individual agents and planning up front (waterfall style), which limits the parallelism further
I did when just starting on a new project, it was working well when I had many new components to implement.
But as the project matured and stabilized every new feature is cross-cutting and it's impossible to parallelize the work without running into conflicts (design conflicts, where two agents would add similar overlapping mechanisms, and also the usual code conflicts, touching the same files). Also, with project maturity I'm much more concerned about keeping it stable and correct, which is hard to do with parallel agents running amok.
Is this 'task' feature really useful? I'd say applications like IDEs and text editors should not have automatic arbitrary execution of code in the first place. 'eval' should be blocked and extensions/plugins should have only very limited power to execute external logic (such as processes for LSP) or require allowlisting manually every process.
For Claude, it's enough to prompt "use tmux to test", that usually does the work out of the box. If colors are important I also add "use -e option with capture-pane to see colors". It just works. I used it regularly with Claude and my TUI. For other agents other than Claude I need to use a more specific set of instructions ("use send-keys, capture-pane and mouse control via tmux" etc.)
Since I have e2e tests, I only use the agent for: guiding it on how to write the e2e test ("use tmux to try the new UI and then write a test") or to evaluate its overall usability (fake user testing, before actual user testing): "use tmux to evaluate the feature X and compile a list of usability issues"
I'm also a long time emacs user (>15 years) but got tired of the endless config fiddling, with some packages breaking over emacs versions, other packages which were cool at the time slowly getting stale and need to switch to yet another similar incarnation of the same idea.
And most of all having to recover the config every time I use a new computer or just connect to a new VM.
I'm building an alternative, and I haven't opened emacs for a month now
I also got tired of the bloat and also prefer working from the terminal.
I'm building Fresh [0] [1] as an alternative to VSCode that runs in your terminal, with the main goal being ease-of-use out of the box (not a vi-clone modal editor), for example supports mouse, menu, command palette, etc out of the box. LSP as well. I'm focused on making it easy to use with minimum or zero configuration.
For a long time I watched TypeScript take over the JavaScript ecosystem as a reasonable way to add structure and safety. But even before TypeScript existed I was thinking about how to approach this differently, ripping out the "bad parts" of JavaScript instead of accommodating them with a complex and partially strict type system.
Also, recently I found out just how expensive (runtime-wise) it is to host a full-blown TypeScript host, such as Deno. This is very cool tech but kind of bloated when all you want is to run some JavaScript, e.g. as a plugin system. It's possible to just pre-compile (transpile) your .ts files and then ship only the JavaScript output and use a lightweight runtime (QuickJS or even the new Micro QuickJS), but then compilation becomes a neccessary step for what could've been purely dynamic, ad-hoc and quick plugins (think elisp in emacs).
minfern is my second shot at this, and I'm exploring just how much JavaScript idioms are "must have" for it to be useful.
LSP is working but requires polish. Go to definition, completion suggestions, inlay type hints and symbol highlighting are all working (probably not as pretty as they should)
Tree-sitter is supported and used internally for a subset of languages for syntax highlighting (other languages use textmate grammars). What exactly do you mean by support for tree-sitter parsers?
I wasn't aware of Debug Adapter Protocol! I'll take a look at that.
Fresh itself has a few dozen direct dependencies. About 33 of those already exist on Debian. It boils down to these missing dependencies:
- The entire oxc toolchain — oxc_allocator, oxc_ast, oxc_codegen, oxc_diagnostics, oxc_isolated_declarations, oxc_parser, oxc_semantic, oxc_span, oxc_transformer
- rquickjs, rquickjs-serde (the QuickJS plugin engine)
- ts-rs
- jsonc-parser
- rust-i18n (this one has an open ITP)
If we build Fresh without plugins enabled, it's just the last three.