question is what can you do 
My daily weblog did not have emojis for posts 0 - 104, so I asked Claude to add them:
can you add emojis to the titles of all the posts that do not already have them?
It churned for 5m or so, and now they are all there.
This is the kind of thing Claude is really good at, and shows the power of using a local Agent to work with local files.
1 Like
Claude has a new feature similar to the ChatGPT Code Interpreter:
1 Like
Claude Sonnet 4.5
… is now the default model in Claude Code.
There is also a VS Code extension in the works.
It may be a topic on itself, but pairing Claude with github/spec-kit: 
Toolkit to help you get started with Spec-Driven Development has been giving me some good results. I’ve been doing some warmup with it by recreating old python projects I have done in the past that I am able to clearly recreate the specifications and then see what the AI creates compared to what I thought it should do.
Den, the author, does a video every couple days on new features (348) Den Delimarsky - YouTube.
It feels weird actually having requirements first to drive the design instead of dev first, specifications second at my day job.
Neat, I just tried spec-kit this morning on a new Go project. Still some learning to do as my current spec-driven workflow is just update README.md or other docs, and tell Claude to implement the changes in the docs.
Thanks for the link to the videos – will check those out.
Claude plugins
You can now install plugins in Claude Code:
This is one example of a site that has a lot of plugins:
There are ~113 plugins in the above marketplace! A number of the plugins include agents.
Plugins are installed globally for a user in the ~/.claude/ directory. The ~/.claude/settings.json file lists which plugins from a marketplace are enabled.
Claude Code Commands Directory
The below directory is the source for the marketplace in the previous post – lots here …
A peak into Claude’s internal skills
I turns out that much of how Claude interacts with various documentation formats is public Markdown and Python.
I’ve been using claude-code now for about a week. These resources in this thread are great!
For Yocto tasks, Claude seems to need quite a lot of direction and/or memories to be provided to it in order for it to be productive, I’m slowly working towards being able to have it build a plan to implement a change I want, implement test scripts to verify the change, implement the change, and run the test scripts via runqemu. I don’t yet think Claude is faster than I am at doing Yocto things, but it is much faster at writing test scripting, which I didn’t used to do much of as I would manually test things.
For programming tasks, I’ve found Claude to be significantly faster than I am. Spending an hour creating a good and detailed plan, letting Claude implement the plan, and then having me do a code review results in MUCH more code being written in any given time frame as compared to me writing the code myself. And Claude is also good about writing unit tests, although I’m still working on convincing it that a failing unit test is never a reason to remove a test, but instead it’s a reason to investigate the failure more deeply.
One downside I’ve found is that I don’t take as many notes when I’m using Claude. Probably a good solution to this would be to archive and catalog my claude-code interactions so that they’re easily searchable. I need to investigate this further.
1 Like
I’ve tried to flip the script on this and adopt a document-driven development approach. I update the docs (README.md, whatever), and tell Claude to implement the changes in the docs. And then, after implementation, I reverse the process and instruct Claude to ensure the documents are up to date.
These are two simple Claude commands I use in about every repo:
update.md
Implement any code changes described in the documentation (\*.md) changes (git diff).
Note for this command, the documentation is driving the code changes, not the
other way around.
update-docs.md
Update any relevant documentation for the current code changes (git diff)
The changelog should describe changes as benefits to the user. It should not
include technical details of the change, that is what git log is for.
It’s not perfect, and I should probably learn to plan better, etc. However, I feel like this gets me out of the prompt-it ‘til you make-it spiral. And in the end, my docs are done.
I’ve also looked at spec-kit, but have not had time to process what it does – offhand, it seems a little heavy for most things I do.
Thanks for sharing what you are learning!
1 Like
In trying to make plugins for Claude Code using Claude Code, it has become clear that Anthropic’s documentation is created intending only to be consumed by humans. They post it online (Home - Claude Docs) but it’s chock full of javascript and hence very hard for Claude to actually parse (I’m unclear if a MCP server exists to provide access, I haven’t found one yet).
I have found that if I clone the actual codebase for Claude Code (GitHub - anthropics/claude-code: Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.) and then ask Claude to go create documentation for itself on how to make plugins based on the claude-code git repo, that the resulting documentation is actually really good and super easy for Claude to utilize to quickly craft new plugins!
1 Like
A post was split to a new topic: AI-friendly APIs have Markdown documentation available
I created my own plugins marketplace – tried to use Claude to create this given the command files already existed – amazing how painful it was to do something relatively simple – just create a couple meta files.
It got so many things wrong – I think in this case it would have been faster to simply read the documentation.
I’ve noticed before that if I ask Claude a question about itself, it will often give vague answers, so these days I simply go to Perplexity if I have a question about Claude. Maybe this will improve in the future with access to the md docs, but it seems like they purposely dumb it down about anything regarding itself.
Unfortunately, claude-code is not open source. The claude-code github repo is an output from their actual codebase, the actual code for claude-code is not public. Contrast this with some of Anthropic’s competitors who do have publicly available apps which are covered under open source licenses (like codex-cli, gemini-cli, octofriend, and qwen-code).