GuidesClaude Code

8 Claude Code mods and how to install them

By Stephen Thorn · 6 minute read ·

8 Claude Code mods: what each one does and how to install it

A mod is a plugin that changes how Claude Code looks and behaves. These are the install steps for the 3 mods in my posts (Blast Radius, Replay Theater and Reflect) and 5 more. Everything here comes from each mod's own README or Anthropic's mods docs: https://code.claude.com/docs/en/plugins/mods/overview

What you need

  • Claude Code 2.1.287 or later in the terminal (check with claude --version). In the Desktop app's Code tab, mods work from 2.1.286.

  • git, to clone Anthropic's sample mods.

  • A few minutes to read a mod before you install it (next section).

Before you install any mod

Anthropic's docs: "A mod is code that runs with your permissions. It can read and write your files, start processes, and make network requests." Mods aren't sandboxed. Install mods only from authors and marketplaces you trust.

  • Before you install one, clone it and run claude plugin validate ./some-mod. The hooks: and calls: lines list what it does.

  • /plugin shows which mods the session loaded, with the count and the names.

  • claude --safe-mode starts one session with every installed mod off. It turns off your other customizations too.

  • If you install a mod from your shell while a session is open, run /reload-plugins in that session.

Anthropic's sample mods (1 and 2)

Anthropic shares sample mods in https://github.com/anthropics/claude-code-playground (the claude-code/mods folder): token-weather, blast-radius and replay-theater. The repository shares them as they are, without support. Set them up once:

git clone https://github.com/anthropics/claude-code-playground.gitcd claude-code-playground/claude-code/modsclaude plugin marketplace add ./

The marketplace points at your clone, so a mod stops loading if you move or delete that folder.

1. Blast Radius: see what a risky command would change before it runs

When Claude calls Bash with one of these 6 kinds of command, Blast Radius holds the call and shows what it would touch:

  • rm -r, rm -f, rm -rf: The files it would delete, with the count and total size

  • git reset --hard: The files with uncommitted changes

  • git checkout -- ., git restore .: The files with unstaged changes

  • git clean: The untracked paths it would remove

  • git push --force: The commits on the remote branch the push would drop

  • Migrations (manage.py migrate, db:migrate, alembic upgrade, prisma migrate): The pending migrations

Press 1 for Proceed or 2 for Cancel. Cancel has the focus when the pane opens, so Enter refuses the command. Claude sees the refusal and the reason. If nobody answers within 10 minutes, the command is refused. The mod doesn't call a model.

In my test, in a throwaway folder, it held rm -rf build and showed 3 files, about 12 KB. I pressed Cancel, and ls build still showed all 3 files.

claude --plugin-dir ./blast-radius                                       # one session onlyclaude plugin install blast-radius@claude-code-playground-mods --scope user   # keep it

The side pane needs a terminal about 144 columns wide. In a narrower one, it draws in the band above the prompt.

Watch out: its README says "This is a safety net, not a permission system." It reads the command text, so bash -c "...", xargs rm, find -delete, $(...) and aliases get through, and it only watches the Bash tool. Keep your permission rules on for a hard block. In a narrow terminal, if another mod (such as Replay Theater) takes the band above the prompt, the Proceed and Cancel buttons may not show. Use a wider terminal when you rely on Blast Radius.

2. Replay Theater: step through every edit Claude made

After a turn that edits files, a hint shows above the prompt, such as â–¶ Replay: 3 edits (press r). Type /replay and a pane steps through each edit, one diff at a time: the file, the tool, the lines added and removed, red for what came out and green for what went in, up to 12 lines per step. Press n for next, p for previous, c to close.

In my test, in a throwaway folder, Claude renamed a function across 3 files, and the hint read â–¶ Replay: 3 edits (press r). The pane stepped through all 3 edits.

Its README: "Replay Theater only watches. It never blocks, changes or delays an edit." It doesn't call a model.

claude plugin install replay-theater@claude-code-playground-mods --scope user

Watch out: it records Claude's Edit, Write and MultiEdit tool calls. A change made through a shell command, like sed, isn't one of those, so it won't show. An edit you denied, or one that failed, still shows. The replay is kept in memory for the current session only.

3. Reflect: turn your corrections into CLAUDE.md rules

A community mod by Bayram Annakov (https://github.com/BayramAnnakov/claude-reflect, the mod folder), not Anthropic's. Needs Claude Code 2.1.286 or later.

Each prompt you type that's 500 characters or less (not a slash command or an "ok") gets one quick check with Haiku. When it finds a correction worth keeping, a band above the prompt shows the rule with these choices:

  • Save to ./CLAUDE.md (this repo) or Save to ~/.claude/CLAUDE.md (every project)

  • This repo instead / Global instead

  • Edit, to reword it first

  • Skip

It writes only the rules you approve, under a ## Learned rules heading, and it doesn't add a rule that's already in the file.

  • Start a message with remember: and then the rule to skip the check and go straight to the band.

  • /reflect-queue lists what it found. /reflect-pause stops and resumes checking.

The checks cost something: one small model call per eligible prompt, and its README says "it counts toward your Claude usage like any request."

claude plugin marketplace add bayramannakov/claude-reflectclaude plugin install reflect-mod@claude-reflect-marketplace

How it differs from auto memory: Claude Code's auto memory is on by default in local sessions, and it saves notes Claude writes itself "based on your corrections and preferences". CLAUDE.md is the file you write. Reflect puts the rules you approve in CLAUDE.md, where you can read, edit and share them.

Watch out: use this or the older Python version of claude-reflect, not both, or every correction gets captured twice. Only save rules you'd want every time. A short CLAUDE.md is easier to keep right.

5 more mods

4. Skins

Redraws the transcript: tool calls with icons and timings, edits as diff cards, and a band above the prompt with your context and plan limits. 7 skins, light and dark. /skin opens the settings, /skin <name> switches, /skin gallery shows every element, /skin off turns it off. Its README: "It starts no process, touches no file and makes no network call." Needs 2.1.287 or later. In a Claude Code session:

/plugin marketplace add hellosverre/claude-skins
/plugin install skins@hellosverre-mods

5. Savvy Progress

A progress bar above the prompt and a live panel of your subagents: model, effort, task progress, context, estimated cost and time. /agents-info toggles the panel. Its README calls the cost "a rough estimate", "not a bill". In a Claude Code session:

/plugin marketplace add johnnyvizz/claude-kit
/plugin install savvy-progress@claude-kit

6. Filetree

An IDE-style file tree that shows where Claude is working: files it reads and searches shimmer purple, writes orange, commits green. It shows in the right sidebar, which needs the fullscreen layout (/tui fullscreen) and a terminal at least 110 columns wide. Needs 2.1.287 or later.

claude plugin marketplace add data-goblin/claude-code-filetreeclaude plugin install filetree@claude-code-filetree

7. Cache Tax

Keeps Claude Code's prompt cache warm while you step away, and shows the estimated rewrite cost before a cold send. /keepwarm arms 6 hours, /keepwarm 90m a shorter window, /keepwarm off stops it. /cache-tax shows the current estimate. Its README: "Pings cost tokens; choose a window you expect to return within." It needs a one-hour cache: subscription usage has it by default, and API-key sessions need promptCacheTtl set to "1h".

claude plugin marketplace add karanb192/claude-code-modsclaude plugin install cache-tax@claude-code-mods

Then restart Claude Code or run /reload-plugins.

8. You Should Know (built into Claude Code)

A side agent that watches your back while Claude works on longer tasks. When it finds something worth knowing that you might miss, it shows a note above the prompt. It's off by default. If it's available for your organization, turn it on in a session:

/plugin enable cc-plugin-you-should-know@builtin

The mistake to avoid

Installing a mod you haven't read. A mod runs with your permissions and isn't sandboxed, so read its README, run claude plugin validate on it, and install only from authors you trust. And don't treat a safety mod like Blast Radius as a lock: keep your permission rules on.