Claude Code forgets your project between sessions and can burn through tokens fast. These four free, open-source add-ons tackle exactly that. Here is what each one really does, with the numbers from their own repos.
Runs a second agent beside Claude that takes notes while it works. Your next session starts with a short index of what matters and pulls in details only when needed. It works with Claude Code, Codex, Cursor, OpenCode and more. (Apache-2.0, 116k+ stars.)
npx claude-mem installSetup continues in your browser, where you pick your agents and options.
Sits between Claude Code and the model and compresses tool outputs, logs, files and history before they're sent. Honest numbers from the repo: about 20% fewer tokens for coding agents, and 60-95% fewer on JSON and logs. Plain prose barely shrinks. It runs locally. (Apache-2.0, 79k+ stars.)
uv tool install --python 3.13 "headroom-ai[all]"
headroom wrap claudeNo uv? pip install "headroom-ai[all]" installs the same tool.
From Anthropic's official plugin directory. It scans your codebase and recommends the top 1-2 MCP servers, skills, hooks, subagents and slash commands for your project. It's read-only: it never changes your files.
Install it from the official plugin page (Install in Claude Code button) or from the /plugin menu inside Claude Code, then ask:
recommend automations for this projectA meta-skill that watches your sessions, notices corrections, preferences and repeated workflows, and suggests skill updates and new skills, each with its reason. Nothing changes until you approve it. (CC BY 4.0, 3.9k stars.)
npx skills add rebelytics/one-skill-to-rule-them-all --skill task-observerAdd -g to install it for all projects. Then add the activation line from the repo's references/environments.md to your CLAUDE.md, because installing alone isn't enough.