Claude Code Forgets, Rushes and Designs Like a Template. These 3 GitHub Repos Fix It
Sunny
Claude Code is very good at writing code. It is less good at three things: remembering what you told it yesterday, slowing down to plan before it builds, and producing a website that doesn’t look like every other AI-built website. Three open-source projects on GitHub go after exactly those problems, and between them they have more than 536,000 stars.
They are claude-mem (memory), mattpocock/skills (process) and Impeccable (design). All three install as Claude Code plugins. Here’s what each one actually does, who it suits, and what to check before you let any of them near a work codebase.
The short version
Repository | What it fixes |
|---|---|
Claude Code forgetting your project between sessions. Apache-2.0, 116.9k stars | |
Claude Code jumping straight to code before the plan is clear. MIT, 307.2k stars | |
AI-built interfaces that all look the same. Apache-2.0, 112.0k stars |
Last checked: 18 November 2026, from each repository’s GitHub page.
1. claude-mem: stop re-explaining your project every morning
Every new Claude Code session starts cold. You explain the architecture again, the naming rules again, the decision you made last Tuesday and why. On a long project that repetition adds up, and the details you forget to repeat are the ones that cause bugs.
claude-mem handles the remembering for you. Hooks watch each action Claude takes in a session (files read, edits made, commands run). A second “observer” model turns that activity into short notes and writes a summary when the session ends. The next session starts with a compact index of those notes and only pulls the full detail when it’s relevant, so you get continuity without stuffing the whole history into every prompt.
The notes live in a local SQLite database on your machine, with keyword and semantic search on top and a small web viewer so you can see what it has kept. It installs with npx claude-mem install and also works with Codex, Gemini, Copilot and OpenCode.
A few things worth knowing before you install it:
Anonymous usage analytics are on by default. Turn them off with
npx claude-mem telemetry disable.The observer is a model call. Your session activity goes to whichever provider you choose to run it, so check that provider’s terms if you work on client code.
You can keep things out. Anything wrapped in
<private>tags isn’t stored, and cloud sync is opt-in.The README also promotes a third-party crypto token. It isn’t tied to how the software works, but it’s the kind of thing your security reviewer will ask about.
Best for: long-running projects and large codebases, and anyone who is tired of pasting the same context into every session.
2. mattpocock/skills: make Claude Code ask before it builds
Getting AI to write code quickly is the easy part. Getting it to write the right code is harder, and a lot of bad AI output comes from a vague plan rather than a weak model.
Matt Pocock, the TypeScript educator behind Total TypeScript and AI Hero, published the skills from his own .agents folder for anyone to use. There are 27 of them. The one everyone talks about is grill-me, which interviews you about a plan until every design question has an answer. Others turn a conversation into a spec (to-spec), break a plan into tickets (to-tickets), run test-driven development (tdd), work through hard bugs step by step (diagnosing-bugs) and review a diff before you merge it (code-review). Some you call yourself; others Claude picks up when the task fits.
It is the most-starred of the three by a wide margin, and part of the appeal is a little counterintuitive: it adds friction on purpose. Claude slows down, asks the questions a senior engineer would ask, and only then writes code.
It installs from Anthropic’s official plugin marketplace with claude plugin install mattpocock-skills@claude-plugins-official. Run /setup-matt-pocock-skills once in each repository and it asks about your issue tracker, your triage labels and where to save docs.
Best for: teams who want AI-assisted work to follow the same planning, testing and review steps they expect from their people.
3. Impeccable: so your AI-built site doesn’t look AI-built
You can usually spot an AI-generated website in a second. Impeccable’s creator, Paul Bakaus, lists the tells: Inter for everything, purple-to-blue gradients, cards inside cards, grey text on coloured backgrounds, and a rounded icon tile above every heading.
Impeccable started from Anthropic’s own frontend-design skill and grew into a design vocabulary for coding agents. It has 24 commands. Some steer the look (bolder, quieter, typeset, layout), some review it (critique gives a scored UX review, audit checks accessibility, performance and responsiveness) and some finish the job (polish, harden, adapt). It also ships a detector with 62 rules that flags those tells, plus problems like cramped padding and skipped headings, without needing a model or an API key.
It installs with npx impeccable install and works with Claude Code, Cursor, Codex CLI, GitHub Copilot and 15 other tools.
Best for: frontend developers, designers and anyone shipping marketing sites or internal tools built with AI.
Do you need all three?
No. They do different jobs and don’t get in each other’s way. If you only try one, pick by the problem you have:
You keep repeating yourself to Claude: claude-mem.
The code comes fast but it’s not what you meant: mattpocock/skills.
It works, but it looks like a template: Impeccable.

Data source: each repository’s GitHub page, checked 18 November 2026.
Are GitHub stars a good reason to install something?
This is the fair objection, and the answer is no. Stars measure attention, not quality. A repository can go viral on a good demo and then go quiet.
Better signals are easy to check on any GitHub page: how recently it was updated, how many open issues there are and whether the maintainer answers them, and what licence it uses. On activity alone, all three look healthy: claude-mem shows 4,604 commits, Impeccable 3,297 and mattpocock/skills 551 (its skills are short text files, so fewer commits is normal).
Claude Code also gives you a built-in check. When you open a plugin’s details, it shows what the plugin will install and, for plugins in Anthropic’s official marketplace, an estimate of how much it adds to every message. The Claude Code plugin documentation puts it plainly: “A plugin can run hooks and MCP servers, so read the pane before you install.” Plugins from third-party marketplaces also don’t auto-update by default, so somebody needs to own keeping them current.
What should a business check before staff install these?
If your developers use Claude Code at work, these plugins run on their machines with access to your code. That makes them a business decision, not just a personal preference. Three questions cover most of it:
What does it capture, and where does it go? claude-mem is the one to read closely, because its observer sends session activity to a model provider. Our guide to data privacy in the age of AI walks through the questions to ask, and what sovereign AI actually takes in Australia covers where the data ends up.
Who approved it? Claude Code lets a plugin be installed for one person or turned on for everyone in a repository. Decide which tools are allowed and write it down. A short section in your responsible AI policy is enough.
Does the team know how to use it well? A skill like
grill-meonly helps if people let it do its job. Hands-on AI training for your team closes that gap faster than a wiki page.
The good news is that all three projects are open source, so your team can read every line before anything gets installed.
What this says about where AI coding is heading
Put the three side by side and a pattern shows up. The model is no longer the bottleneck. The work now is everything around it: what the agent remembers, the process it follows, and the standard of taste it’s held to. That shift is what agentic AI means for business more broadly, and it isn’t limited to developers. Claude Cowork brings the same ideas to everyday office work, as our guide to getting started with Claude Cowork shows.
For most businesses the useful question isn’t “which repo is trending this week” but “which problem are we trying to fix?” That’s the same filter we use for choosing the right AI tool for your business, and it often settles whether to build custom AI or use off-the-shelf tools. When the answer is something built around your own systems, that’s where custom AI development comes in.
Frequently asked questions
What is a Claude Code plugin?
A package that adds skills, agents, hooks or MCP servers to Claude Code. You install it from a marketplace with the /plugin command, either for yourself or for everyone working in a repository.
Are these three repositories free?
Yes. claude-mem and Impeccable use the Apache-2.0 licence and mattpocock/skills uses MIT, so you can install, read and change them freely. Claude Code itself still needs a Claude plan or API access, and claude-mem’s observer runs on whichever model provider you choose.
Do they only work with Claude Code?
No. mattpocock/skills also supports Codex, GitHub Copilot and Gemini CLI. Impeccable lists 19 tools, including Cursor. claude-mem works with Codex, Gemini, Copilot and OpenCode as well.
Is claude-mem safe to use on client code?
It stores its memory locally, lets you exclude anything in <private> tags and keeps cloud sync off unless you turn it on. The part to check is the observer model, because your session activity goes to that provider. Run it past whoever owns your data policy first.
Which one should I install first?
The one that matches your biggest frustration. Most teams feel the planning problem first, which is why mattpocock/skills is a sensible starting point.
Where to start
If your team is already using Claude Code, or thinking about it, the hard part isn’t finding tools. It’s choosing the few that fit your work and rolling them out safely. Book a free 30-minute AI strategy call and we’ll help you work out where AI coding tools will make the biggest difference for your team.
Sources: thedotmack/claude-mem, mattpocock/skills and pbakaus/impeccable on GitHub; Claude Code docs, Install and manage plugins.






