Not a list of fifty. Four plugins that change how Claude Code actually behaves: less token burn, memory between sessions, and a review pass that catches what you would have shipped. Every number below checked on GitHub and Anthropic's own pages, July 2026.
89,705 stars, MIT. Its own README benchmark reads “~54% less code (up to 94%) · ~20% cheaper · ~27% faster · 100% safe”, measured across 12 feature tasks on a real repo. Note the honest version: it is ~54% less code and ~20% cheaper, not 50% off your bill. /plugin marketplace add DietrichGebert/ponytail /plugin install ponytail@ponytail
02
02Memory between sessions
claude-mem
88,619 stars, Apache-2.0, shipped 3 days ago. Captures what the agent did, compresses it, and injects the relevant part into your next session, so you stop re-explaining your own project every morning. npx claude-mem install
03
03Anthropic's own
Code Review
“AI code review with specialized agents and confidence-based filtering for pull requests.” It launches five agents in parallel — CLAUDE.md compliance, bug detection, git history, prior PR comments, code comments — and only posts findings above a confidence of 80/100. 430,450 installs. /code-review on any PR branch.
04
04Where they come from
The official marketplace
anthropics/claude-plugins-official (32,692 stars) is Anthropic's curated directory — Code Review lives inside it. Add it once and every plugin is one command away. /plugin marketplace add anthropics/claude-plugins-official /plugin install <name>@claude-plugins-official
The order that actually helps
1. claude-mem first — it pays off from session two onward. 2. Ponytail next — less code written is less code to maintain. 3. Code Review last — run it before you ship, not while you build. # install one, use it a week, then add the next. All four at once and you # will not know which one changed the behaviour.
Want this running for you?
Build it step by step inside AI Creative OS, with me in the thread.
