Claude Code mods went live on October 1, 2026. Two weeks in, and the community catalogue awesome-claude-code-mods says it’s already scanned 4,314 of them (scan dated October 16). Most are weekend toys. Pixel pets, arcade games, themes. Fun! But the ones that fix stuff I actually run into every day? A handful, maybe.
In my earlier post on Claude Mods I went into how mods work and the handoff panel I want to build myself. This one’s just the shopping list. I went through the catalogues, opened every repo below and read the README. Star counts are whatever GitHub showed on October 17, 2026, and honestly don’t read much into the small numbers, the whole category is two weeks old.
Update (Oct 18, 2026): Someone asked Boris Cherny, the guy who runs Claude Code, when we’d get a Chrome Dino game starring Claude, and his entire reply was “ask claude to make a mod for it.” Which, honestly, is kind of the whole point of this list.
How I picked
Same test as last time. With it installed, do I need fewer follow-up prompts? Less log digging? One less round of rework? If a mod just gives me more stuff to look at, nope, it’s out.
I also skipped anything without a public repo, plus anything whose README couldn’t say plainly what it does. A few mods that looked really promising got shown off on Reddit with no code attached… cool, but there’s nothing to install yet.
Before you install anything
Please don’t skip this bit. Anthropic’s own Mods overview is pretty blunt about it: mods aren’t sandboxed. Turn on Claude Code’s sandbox and a process a mod starts still runs outside it. It runs with your permissions and it sees your prompts and tool calls.
So here’s what I do with every mod:
- Clone the repo, run
claude plugin validate ./the-mod. It doesn’t run the mod. It just lists which events it hooks and what it asks Claude Code to do, like reading files or making network requests. - Try it for one session with
claude --plugin-dir ./the-modbefore you install it for real. - Know where the off switches are.
--safe-modegets you one clean session, or set"disableAllHooks": truein~/.claude/settings.json, though heads up, that one also kills your settings hooks and your custom status line. - You need Claude Code v2.1.287 or later in the terminal. Old README telling you to set
CLAUDE_CODE_ENABLE_FUNCTION_HOOKS? Ignore it, current versions don’t care.
Safety mods
blast-radius (Anthropic sample)
When Claude’s about to run something risky, think rm -rf, git reset --hard, a force push, a migration, it holds the command and shows you what you’d lose. Proceed or Cancel, your call.
If you let Claude run shell commands without approving each one (and I suspect a lot of us do) and you’ve got uncommitted work lying around, install this first. Seriously.
Install: it lives in anthropics/claude-code-playground (243 stars) under claude-code/mods/blast-radius. Clone the repo and load the folder with --plugin-dir, or add the mods folder as a local marketplace like the README says.
GitHubanthropics/claude-code-playground
The catch: it only knows the commands it knows. Anything else just runs. And the playground README is upfront that entries are provided as-is, no support. It’s a sample, so, fair enough.
transcript-redactor
I like the idea behind this one a lot. Any key that lands in the conversation, whether you pasted it, a tool printed it, or it was just sitting in a config file, gets swapped for a placeholder. Claude works with the placeholder and the real value goes back in when the command actually runs on your machine.
Worth it if your sessions touch .env files, production configs or API responses with tokens in them. A key in the transcript gets re-sent to the model on every single turn. Catch it early.
Repo is ray-amjad/claude-code-mods (70 stars), folder plugins/transcript-redactor. Some older guides send you to awesome-claude-code-function-hooks; don’t bother hunting, it redirects here now.
Gotcha: it’s pattern matching, so some secrets will slip through. Think seatbelt. Not a license to start pasting production keys into chat.
Cost and context mods
token-weather (Anthropic sample)
A one-line weather forecast for your context window. Clear, cloudy, showers, storm, then “compact soon”. Kinda silly, kinda great. If you’ve ever wondered why Claude suddenly forgot something halfway through a task, you get why I want it. Same playground repo, folder token-weather.
Small gotcha. It measures against the full window, but Claude Code’s own warning counts toward the compaction point, so the two numbers won’t match. That confused me for a minute.
cache-tax
What it does: two things, really. It keeps the prompt cache warm while you step away (/keepwarm 90m), and if you come back to a cold cache in a long session it stops your message once and shows you the estimated rewrite cost. The README example is kind of painful: a two-line message in a 330k-token session, estimated at $6.61.
Long sessions, or you’re on an API key paying per token? Take a look.
Install: karanb192/cache-tax (200 stars). Then claude plugin marketplace add karanb192/claude-code-mods and claude plugin install cache-tax@claude-code-mods.
Thing is, the warming pings cost tokens too. You also need a one-hour cache and Claude Code left running. Set a window you’ll actually come back within, otherwise you’re just paying to babysit a cache you walked away from.
Review and workflow mods
replay-theater (Anthropic sample)
Adds /replay, which steps through the edits Claude made last turn, one diff at a time. Who’s it for? Me, basically. Figuring out what an agent actually changed is the part of the job I most want to go faster.
Playground repo again, folder replay-theater. It only covers the last turn though, so for a whole session’s worth of changes the built-in /diff is still nicer.
cc-pr-tracker
What it does: paste a pull request URL and you get one line above the prompt with merge state, review decision and required checks, refreshed every minute by default. If a check fails, Claude gets a note with the log link.
For anyone who opens PRs from Claude Code and then just… sits there waiting on CI.
Install: sezaakgun/cc-pr-tracker (5 stars): claude plugin marketplace add sezaakgun/cc-pr-tracker, then claude plugin install cc-pr-tracker@cc-pr-tracker.
Gotcha: GitHub only, and you need the gh CLI signed in. It’s very new and very small, so read the code first.
baton
This one’s interesting. It passes a task from one Claude Code session to the session that owns another repo. /pass infra ... queues the job over there, and that session makes the change, runs checks and opens a pull request. If you keep one session per repo (API, frontend, infra) you’ll probably like it. Grab it from ashishsk93/baton-mods (5 stars).
I’d go slow with it though. It acts on your behalf in other sessions and goes all the way to a pushed branch. Turn on task confirmation and lock down which repos can send work before you trust it with anything.
cc-multi-cli-plugin
Brings other providers’ models into a Claude Code session, ChatGPT models via a Codex login, Cursor, Grok, so Claude can hand work off to them. Already paying for a second provider and want a second opinion without switching tools? This is for you.
Install: greenpolo/cc-multi-cli-plugin (193 stars). One plugin per provider.
It’s a lot heavier than a normal mod, with its own launch command and a recent Node version. And remember, whatever you hand off counts against your usage with that provider.
Where to find more
- karanb192/awesome-claude-code-mods (624 stars) is the big catalogue. The useful trick is it shows what the validator says each mod can read, write, run or send over the network. Filter on that before you fall in love with a screenshot.
- anthropics/claude-plugins-official (41,096 stars), Anthropic’s curated plugin directory. Broader than mods, it’s got skills, MCP servers, commands, and its README still tells you to trust a plugin before installing it.
- ccmods.app is smaller. It lists 96 mods validated against a specific Claude Code version, plus what each one can access.
My short list
Setting up a fresh machine today? I’d put blast-radius and transcript-redactor on first, then token-weather, replay-theater for review, and cache-tax only on long API-billed sessions. The rest depends on how you work.
The bigger point from my last post still stands. The best mod is often the one that kills the instruction you keep typing over and over. Try a few, read the code of the one you like most, then ask Claude to write a version that fits your own habits. Oh, and figure out how you’ll tell whether it actually helped. That’s harder than it looks, I wrote about why in my post on Karotte and agent evaluation, and about checking AI output against its own evidence in this piece on AI research reports.