MODZ

19 mods for Claude Code. One panel.

See every move Claude makes: agents, files, plans, tests and risky commands, live, beside the chat.

_{Watch the whole film · 54 seconds, with sound · download it}

Install

In Claude Code, add the marketplace, then install the panel and the mods you want:

/plugin marketplace add partalemre/MODZ
/plugin install modz@modz
/plugin install savvy-progress@modz

Or install all of them at once from a terminal:

claude plugin marketplace add partalemre/MODZ
for m in modz savvy-progress savvy-flow filetree plan-steps handoff transceiver blast-radius secret-guard checkpoint \
         test-badge replay-theater session-budget cache-tax hud-pro effort-router done-ping reflect-mod skins terminal-browser; do
  claude plugin install "$m@modz"
done

Or ask your coding agent: Install MODZ for me. Follow https://partalemre.github.io/MODZ/llms.txt and ask me which mods I want.

Start a new session. The MODZ panel opens on the right with a line from every mod you installed.

• Where it draws. The Claude desktop app's Code tab draws everything. The terminal draws the same panels as text.
• Requirements. Claude Code 2.1.295 or later, tested on 2.1.295 and 2.1.296. terminal-browser draws in the terminal only. Outside the desktop app, done-ping falls back to a macOS banner.
• Updates. Automatic and free. The first session after you install turns on Claude Code's own auto-update for the modz marketplace, so every mod you installed, transceiver from its own repo included, updates itself in the background a few minutes into a session and loads fresh at the next start. No model call, no command to run. If you already chose under /plugin > Marketplaces, that choice stays; to turn it off, choose Disable auto-update there.
• Language. The interface is in English. Set MODZ_LANG=tr for Turkish.
• Transceiver needs a claude.ai login. If Claude Code runs inside another app (an IDE, claude -p, an SDK host), also set CLAUDE_CODE_ARTIFACT=1.

Both variables go in the env block of ~/.claude/settings.json:

{ "env": { "MODZ_LANG": "tr", "CLAUDE_CODE_ARTIFACT": "1" } }

One panel. Every mod, live.

The panel opens beside the chat when a session starts. Each mod reports one honest line: what it measured, never a guess. One click opens that mod's own pane next to the panel.

Every agent gets a face.

Each subagent gets a crab, a name and a card with its task, step, progress, cost and time, updated as it works. A band above the prompt shows the whole crew while a turn runs.

Risky commands wait for your OK.

rm -rf, git reset --hard, git clean, force pushes and migrations stop before they run. The card shows exactly what each would delete or throw away, and the safe choice has the focus.

Your team, inside Claude. ^beta

Transceiver lets you chat with teammates without leaving Claude Code. Your agent can ask theirs for status, diff, tests, log or a file. The answer is computed on their machine and costs zero tokens. Only collaborators on the repo can ask, and secrets never leave your machine. Rooms are private claude.ai artifacts; there is no extra server. Transceiver is in beta: it works end to end, and its rough edges are being found. How it works

The mods

Most mods use no model tokens. These can, each with the cheapest model and never more than one call per event:

• reflect-mod: one small check per correction it finds, and one per /mine.
• cache-tax: warming sends a request, only when you turn it on.
• handoff: one summary per handoff note (summarize, on by default; the note is written without it too).
• effort-router: one check per prompt, only in auto mode.
• blast-radius: one check per command it judges, only when .claude/house-rules.md exists.
• replay-theater: one call when a step has no description of its own.
• Transceiver's free-form ask: a small model on the other side, at most ten a day per room.

plan-steps (while it brings a plan back) and secret-guard (while placeholders exist) add a short section to the system prompt.

What each mod can touch

Read from the code, not from a promise. "Store" is the mod's own file in ~/.claude; every mod also keeps session state in memory.

Jev is TypeSafe's classifier: set MODZ_JUDGE=jev and TYPESAFE_API_KEY to use it instead of your own model account. What it is sent is masked first.

Make it yours

• Arrange the panel. Press Customize at the bottom of the MODZ panel: move rows up, hide the ones you do not use, and bring them back from Hidden. Reset puts everything back. Your layout is kept across sessions.
• Get back from any pane. Every pane the panel opens has a ← MODZ button at the top. It brings the panel forward, and opens it again if it was closed. /modz does the same.
• Choose what loads. Each mod is its own plugin: install or uninstall any of them, and the panel follows. /plugin configure modz@modz turns off opening at session start.

Add your own mod to the panel

The panel lists any mod that reports to it, with no change to MODZ. Your mod talks to it through $.state:

1. Report a line. Write your own { plugin: '<your-mod>', key: 'hub' } value: { pane, detail, tone, at }, plus your row's title, hint, icon (a kit icon name), section (work, safety or meter) and an optional badge. tone is idle, live, ok, warn or bad; write it only when the text changes.
2. Open on request. Hook state.set on { plugin: 'modz', key: 'open' }; when its pane is yours, call $.ui.open({ id, title, focus: true }).
3. Offer the way back. While { plugin: 'modz', key: 'isOn' } is true, draw HomeButton from kit/home.tsx first in your pane; its press writes your own { plugin: '<your-mod>', key: 'home' } as { at }.

DESIGN.md §9 has the full contract, and mods/modz/types/index.d.ts the types. bin/modz new <name> starts a mod from the template.

For AI agents

If a person asks you to install MODZ, do exactly this. Every step is safe to repeat.

1. Check the version: claude --version must print 2.1.295 or later. If it is older, ask the person to update Claude Code first.
2. Add the marketplace: claude plugin marketplace add partalemre/MODZ.
3. Install the panel: claude plugin install modz@modz. Then install each mod the person wants, by the names in the table above, as claude plugin install <name>@modz. If they did not choose, install all twenty with the loop under Install.
4. Optional settings: put them in the env block of ~/.claude/settings.json, merging with what is already there and never overwriting it:
   • "MODZ_LANG": "tr" for a Turkish interface.
   • "CLAUDE_CODE_ARTIFACT": "1" if Transceiver is installed and this Claude Code is driven by an SDK host.
5. Verify: claude plugin list shows each <name>@modz as enabled.
6. Tell the person to start a new session, or run /reload-plugins in an open one.

To remove a mod: claude plugin uninstall <name>@modz. To remove everything: also run claude plugin marketplace remove modz. A machine-readable summary is at partalemre.github.io/MODZ/llms.txt.

Develop

mods/<name>/ holds each mod; kit/ is the shared drawing kit, copied into the mods by bin/kit-sync.

• Unit checks. bin/modz check [name] validates each mod and runs its tests, on both the terminal engine and the desktop app's engine.
• End to end. bin/modz desk starts a real session, speaks the desktop app's own protocol to it, and clicks through every panel. Run both before opening a pull request.
• Seeing it. An agent cannot see the desktop app, so bin/modz shot [--pane] [--title t] saves the Claude window as a PNG, even when another window covers it. Mods load when a session starts: to look at fresh code, open a new session in its own window and shoot it with --title.
• Releases. No mod sets version: Claude Code then versions each one by its commit, so every push to main reaches every user. Keep it that way; a pinned version stops updates until someone bumps it.
• Types. Claude Code writes its API types into .claude-plugin/types/ beside a mod it hot-reloads. Each mod's tsconfig.json extends them, so tsc -p mods/<name> type-checks that mod.
• Kit gallery. tools/kit-gallery draws every kit component; its tests check contrast in both themes.
• Design rules. DESIGN.md (in Turkish) covers colour, type, components and the rules every mod follows.

Credits

Several mods are adaptations of open-source projects, each under its original license, which lives in that mod's folder:

Everything else is MIT.

upstreams.json records the upstream date each adaptation is synced to; a weekly workflow (.github/workflows/upstream-watch.yml) keeps one open upstream issue per mod listing the newer upstream commits.
A second weekly workflow (.github/workflows/mod-radar.yml) keeps one radar issue with the most-starred new mods from the awesome-claude-code-mods catalogue that MODZ has not adapted. MODZ keeps one mod per job: an idea worth taking is folded into the mod that already does that job.