# Vibe-Weaving 1001: Multithreaded Vibe-Coding
URL:: https://liminalwarmth.substack.com/p/vibe-weaving-1001-multithreaded-vibe
Author:: Liminal Warmth ❤️🔥
## AI-Generated Summary
Create a clear project structure with top-level rules and agents.md files so agents understand subsystems. Use specialized sub-agents (code-reviewer, debugger, docs-expert) and explicit workflows to plan, implement, and review work. Keep tasks small, write specific prompts, and iterate until reviewers approve to avoid regressions.
## Highlights
> Put a project-specific Claude.md or Agents.md rules file in your top level project directory. ([View Highlight](https://read.readwise.io/read/01k4x4bew2ccb243y1579d3r9r))
> Shorter is better than longer. ([View Highlight](https://read.readwise.io/read/01k4x4bk5cwf73f08dhwx7j45f))
> **Only add rules you** ***really*** **need.** ([View Highlight](https://read.readwise.io/read/01k4x4bnz11wyp41bdf7dscmqs))
> Provide high level systems overviews. ([View Highlight](https://read.readwise.io/read/01k4x4bsdxgxwfzszhvheqswyx))
> Add an explicit directory and file map with functionality labels. ([View Highlight](https://read.readwise.io/read/01k4x4c062v6nqknjtrk6sh9h3))
> Require test-driven development in your project. ([View Highlight](https://read.readwise.io/read/01k4x4c6dh1j20zxscyqq6d48x))
> Create specialized sub-agents for specific tasks. ([View Highlight](https://read.readwise.io/read/01k4x4cbx124m02jwznma1bkbv))
> code-reviewer: ([View Highlight](https://read.readwise.io/read/01k4x4cw8gwex7rsnm1ba7ysqm))
> debugger ([View Highlight](https://read.readwise.io/read/01k4x4d5nh044e8nx06akd72fh))
> **plan-designer**: ([View Highlight](https://read.readwise.io/read/01k4x4d9q8mfvzbsergs97r8dg))
> docs-expert: ([View Highlight](https://read.readwise.io/read/01k4x4ddavypr9nftc6et3teqj))
> code-auditor: ([View Highlight](https://read.readwise.io/read/01k4x4dfz7ccb7cg6m2b0dpf93))
> Add agents.md files in subfolders to describe complex systems. ([View Highlight](https://read.readwise.io/read/01k4x4ey2azgzbvhw51gjjmwy6))
> I use the docs-expert subagent to generate these and keep them updated whenever I'm doing extension or refactoring work in a submodule. ([View Highlight](https://read.readwise.io/read/01k4x4ffa84n10d9q67cz2hetb))
> **Do as much up-front architectural planning for the subsystems and features as you can, and put it in the top level rules file.** ([View Highlight](https://read.readwise.io/read/01k4x4f6at90qsfm3vq3yt9v65))
> Automate error detection and use formatting and linting hooks. ([View Highlight](https://read.readwise.io/read/01k4x4hbvtw409k9pw13c1abwt))
> Vibecoding Mindset
> Another issue I see with people trying and struggling to vibe code is that they often have some miscalibrated expectations for both the process and what the AI output ought to look like. ([View Highlight](https://read.readwise.io/read/01k4x4kbeyv4ffr2fd0h8q0jes))
> The reality is that once your project is set up correctly, many of the problems you face can be resolved by:
> 1. Planning your request a bit more carefully (more specific is better)
> 2. Changing the scope of the task (smaller is better)
> 3. Having more patience (it’s normal to take a few tries once complexity increases) ([View Highlight](https://read.readwise.io/read/01k4x4khkz5n61ynattfpmtgt2))
> This is why I say that it's good to have some technical skill, even if you are coming at this from a purely vibe-coded perspective. Developing good intuitions about how the systems that Claude is building work together and understanding the constituent pieces of the project you're trying to build, even if you're learning about them as they’re being built, is a great way to build up a mental model that will help you to help your agent be more successful. ([View Highlight](https://read.readwise.io/read/01k4x4q73ts83htyrc81v9e2h8))
> Finally, don't assume that you know more than it does. One of the highest alpha things I do is flip over to ChatGPT while Claude is coding, talk through a problem I'm having at a high level, and ask it for advice on novel approaches. ([View Highlight](https://read.readwise.io/read/01k4x4qhdadd4pz11j2kgh3snq))
> I almost never check the agent's actual code. I rely heavily on sub-agents' cross-reviews to flag problematic patterns or design issues. ([View Highlight](https://read.readwise.io/read/01k4x4v94fdknxe36e16gsgpfm))