I created a Claude Code plugin to protect files from unwanted edits. Just define the paths or block a whole directory tree from automatic edits. It's still early days, so report any issues and I'll get them fixed quickly.

github.com/kodroi/block

How to get #ClaudeCode to follow your coding standards?

1. Update editorconfig to treat all analyzer issues as errors
2. Add a stop hook to build the project

This way Claude will work until all the code matches your standards.

What analyzer are you missing to get this done?

Are you afraid to run Claude Code in full YOLO mode?

Use the Claude safety net and have a peaceful lunch while Claude gets your feature done!

github.com/kenryu42/cla...

Best way to get Claude Code to do what you want?

Create deterministic feedback loops.

Add a stop hook that runs your build and tests. That's just the start, turn up your static analysis and get the exact patterns you want - every time.

#ClaudeCode #AgenticCoding

Why is the standard way to look at unit test coverage a percentage, usually 80%?

Can’t we just define what should be tested with pattern analysis and architectural practices and monitor that?

#UnitTesting #TestAutomation

How to get the code you want out of a coding agent?

Add scaffolding.

Scripts that run when the code is changed to verify. Guards that prevent editing certain files.

With Claude Code, you can achieve this with hooks code.claude.com/docs/en/hook...

#ClaudeCode #AgenticCodding

Why is coding the best use case for AI workflows?

Feedback loops. Do tests pass and does it meet our coding standards?

They are deterministic questions, the same input has the same output. AI models are not deterministic but statistical. Different answer everytime.

Feedback loops change it.

How to make sure Claude Code is refactoring a legacy pattern away from everywhere?

Create custom code analyzer that will fail if the pattern is still used.

I've been creating them in .NET (with Claude). The patterns are treated as build errors --> Claude runs until all the patterns are fixed.

The most useful design principle when creating code with Claude Code:

The principle of least astonishment or the pricinple of least WTFs. en.wikipedia.org/wiki/Princip...

Prompt: "Review my changes for POLA violations"
Result: "Code that's easy to follow"

#claudecode

Claude Code can now start a with a fresh context for implementing with a click of a button. #claudecode

1. Use plan mode
2. Plan your feature
3. Select: Yes, clear context

Is Claude Code not following your architecture?

1. Create architecture tests
2. Create a stop hook to the tests
3. Claude code won't stop until the implementation matches your architecture
4. Less micro-managing, more value

#claudecode

I haven't written code in months, still did a major refactoring for a client. How?

Agentic coding with feedback loops.
1) Add a Claude Code stop hook which builds, runs the tests, checks code quality and tests architecture
2) Tell Claude code to refactor a class. It will run until the hooks pass.

AI without process redesign is just automation theater. Impressive demos, zero sustainable value. Do the unglamorous work first.

Your automation isn't failing because of bad technology. It's failing because your processes were never clearly defined in the first place. AI can't fix what you never understood.

We have an AI talent shortage, yes.

But the bigger gap? Business and technical teams who can't speak the same language.

AI fails when strategy and execution live in different worlds.

AI frees humans for creative work—but only if you're brave enough to let go of control. Most leaders say they want innovation while micromanaging every automated decision. Pick one.

Most AI failures aren't technical failures. They're cultural ones. You can't bolt innovation onto a culture that punishes experimentation.

Most AI projects die from data problems, not algorithm problems.

Poor quality. Unstructured. Inaccessible.

Before you buy another AI tool, ask: do we actually have the data foundation this needs to work?

95% of generative AI pilots fail to deliver measurable business impact.

Not because AI doesn't work. Because we're solving the wrong problems.

The question isn't "what can AI do?" It's "what problem are we actually trying to solve?"