Every new Claude Code session opens the same way. It has no memory of what you told it an hour ago, so you re-explain your stack, your conventions, and the thing you were in the middle of building. The plugin ecosystem has quietly fixed most of that. These are the five I install on every project.
Start with what Claude Code already gives you, because it is more than most people realise. Your CLAUDE.md file is loaded in full at the start of every session, so anything you write there is always in front of the model. And auto memory, on by default, lets Claude write its own notes to a MEMORY.md file in your project and reload them next time. You manage both with the /memory command. The five plugins below build on that base. They carry your past sessions forward, review what gets written, and keep everything you know in one place.
First, What a "Plugin" Actually Is
One quick bit of plumbing. A Claude Code plugin is a folder with a manifest. It can bundle skills, slash commands, subagents, hooks, and MCP servers. You install plugins from a marketplace, which is just a git repo Claude Code knows how to read:
/plugin marketplace add <owner/repo> /plugin install <name>@<marketplace>
Every plugin below installs that way. Here is the lineup at a glance, then a closer look at each one and the command to add it.
| Plugin | What it does | Official? | Best for |
|---|---|---|---|
| Ponytail | Steers Claude to write leaner code | Third-party | Cutting token cost |
| Code Review | Runs four agents over every pull request | Yes | Catching bugs pre-merge |
| Claude-Mem | Records sessions and replays them into new ones | Third-party | Persistent memory |
| Obsidian Second Brain | Connects Claude to your knowledge vault | Third-party | Notes and research |
| Official Anthropic pack | Curated marketplace of Anthropic plugins | Yes | Trusted starting point |
One: Half the Code, Same Result. Ponytail.
Ponytail, by Dietrich Gebert, is a skill that steers the model to write minimal code: reuse before you write, the standard library before a new dependency, the smallest change that solves the problem. Its own benchmark reports the headline result, half the code, around 20% cheaper, zero corners cut. Less code is fewer tokens, which is a smaller bill and a codebase that stays easier to read six months from now.
/plugin marketplace add DietrichGebert/ponytail /plugin install ponytail@ponytail
Repo: github.com/DietrichGebert/ponytail. It is a young project, so read the skill before you route a large codebase through it. The idea underneath it, write less to spend less, is sound whatever the exact number turns out to be for your repo.
Two: Four Agents on Every Pull Request. Code Review.
Code Review, built by Anthropic's Boris Cherny and shipped in the official marketplace, adds a /code-review command that spins up four agents in parallel. They check your changes against your CLAUDE.md rules, hunt for bugs, and read your git history for context. It scores each issue by confidence, drops the weak ones, and with the --comment flag posts the survivors straight onto your pull request. Four AI reviewers catching bugs before your users do, on every diff.
/plugin marketplace add anthropics/claude-plugins-official /plugin install code-review@claude-plugins-official
Repo: github.com/anthropics/claude-code/plugins/code-review. It is official, it is free, and a four-way read of every change is one of the highest-leverage things you can automate.
Three: The One That Truly Remembers. Claude-Mem.
Claude-Mem, by Alex Newman, is the memory upgrade the whole list is built around. During a session it captures what happened, compresses it, and stores it as searchable memory: a local SQLite database plus vector embeddings. Next session, it pulls the relevant pieces back into context, so Claude picks up not just your rules but what you were doing and why you did it. It remembers your project, your files, and your decisions.
Install it the right way. The command is npx claude-mem install. That wires in the hooks that make the memory actually work, so the plugin records and replays your sessions from the first run. Repo: github.com/thedotmack/claude-mem.
If you install nothing else on this list, install this one. It is the plugin the "goldfish memory" complaint is actually asking for, and of the five it is the one that carries the most across sessions.
Four: Every Note in One Connected Graph. Obsidian Second Brain.
Obsidian Second Brain, by Eugeniu Ghelbur, connects Claude to your Obsidian vault and turns it into an AI-maintained knowledge base: notes that update themselves, semantic search across everything, scheduled clean-up agents, and a command that generates documentation from a codebase. Your files, decisions, and research land in one connected graph that Claude can read and extend, so the context of a project lives somewhere permanent instead of evaporating when the session ends.
/plugin marketplace add eugeniughelbur/obsidian-second-brain /plugin install obsidian-second-brain@obsidian-second-brain
Repo: github.com/eugeniughelbur/obsidian-second-brain. If you already live in Obsidian, this is the one that remembers your thinking across every project, not just the current repo.
Five: Cheat Codes From the Source. The Official Anthropic Pack.
The official pack from Anthropic is the first place you should shop. It is a curated marketplace maintained by the team that builds Claude Code, so everything in it is tested and trustworthy: code review, a code simplifier, CLAUDE.md management, session reports, and more. These are the cheat codes from the source, and two of them, claude-md-management and session-report, feed straight back into keeping Claude's memory sharp.
/plugin marketplace add anthropics/claude-plugins-official
Repo: github.com/anthropics/claude-plugins-official. Add it once and browse. It is the safest way to try new plugins, because Anthropic maintains every one inside it.
The Honest Part
Two things before you install all five at once. First, start with the memory Claude Code already has. A well-written CLAUDE.md does half the job, and every plugin here works better sitting on top of it. Second, three of the five are third-party (Ponytail, Claude-Mem, Obsidian), so give the repo a two-minute read before you wire it into anything sensitive. That is not a knock on any of them. It is just how you install tools that touch your codebase.
A plugin cannot remember what you never wrote down. Start with the CLAUDE.md, then stack the five on top.
If You Install Them, Install In This Order
- 01Claude-Mem, via
npx claude-mem install, for real session memory. This is the core fix. - 02The official Anthropic pack, added once, as your trusted shop for everything else.
- 03Code Review, for a four-agent pass on every pull request before it merges.
- 04Ponytail, for leaner code and a smaller bill on every task.
- 05Obsidian Second Brain, if your notes and research already live in Obsidian.
The memory of a goldfish is a fixable condition. Fix it with Claude-Mem and a well-written CLAUDE.md underneath, then stack the other four on top for cleaner code, safer merges, and a knowledge base that outlives the session.
Questions People Ask
Does Claude Code remember anything between sessions by default?
Yes, partly. It loads your CLAUDE.md file in full every session and, with auto memory on, reloads notes it wrote to its own MEMORY.md file. For the full back-and-forth of past sessions, a plugin like claude-mem adds that layer.
Which Claude Code plugin gives it persistent memory?
Claude-Mem, by Alex Newman. It records each session, compresses it into a local database with vector search, and replays the relevant parts next time. Install it with npx claude-mem install.
What is the official Anthropic plugin pack?
anthropics/claude-plugins-official is a curated marketplace of plugins maintained by Anthropic. You add it once and install individual plugins like code-review, code-simplifier, or session-report from it.
What does the Ponytail plugin do?
It steers Claude to write leaner code: reuse before writing, standard library before a dependency. Its benchmark reports about half the code and roughly 20% lower cost per task, which means fewer tokens on your bill.
How do I install a Claude Code plugin?
Add its marketplace, then install it: /plugin marketplace add owner/repo, then /plugin install name@marketplace. Plugins can bundle skills, commands, agents, hooks, and MCP servers.
Want an AI setup that remembers how your team actually works?
I help founders and brands build generative AI systems that ship: agentic workflows, coding-agent setups, and content engines that are production-grade down to the plumbing. If you want a second pair of eyes on yours, book a session.
Book a Growth Chat
Plugin behaviour, install commands, authorship, and repository links were checked against each project's GitHub repository and Anthropic's Claude Code documentation on plugins and memory, current as of July 2026. Benchmark figures, including Ponytail's half-the-code and 20%-cheaper numbers, are the projects' own and are described as such. Nothing in this post is sponsored, and none of these tools were provided to me.