9 New Claude Mods That Can Truly Change How You Work

Nine small add-ons launched on 1 October are reshaping daily work for fans of TypeScript functions , because each mod changes how Claude behaves and looks without switching tools. You need Claude Code 2.1.287 or newer, and the mods run in the CLI terminal plus the desktop app code tab. They do not run in the VS Code extension yet. A mod ships inside a plugin and installs and shares like any plugin, following the Claude Blog mods announcement. That packaging keeps discovery simple for teams.

Under the surface, every action emits an event hook such as a tool call, a permission request, or a screen draw, and a mod attaches to one event. It can run before, after, or instead of the event, or wrap it with before-plus-after logic to rewrite prompts, block or retry calls, approve or deny permissions, redact secrets, or add buttons. Hooks cannot rewrite events, draw new UI, or replace features, while mods can. The InventiveHQ framing helps here: a skill carries knowledge, MCP adds capability, and a hook guarantees deterministic execution. Mods are not sandboxed and hold the same machine access as Claude Code, so install only from trusted sources.

First trio on screen: progress, themes, second reader

Savvy Progress puts a live progress bar above the prompt so long runs feel legible instead of frozen. A sub-process panel shows the active model, completion progress, and API cost, plus customizable pixel mascots that keep waiting time light. The CodeGateway advanced guide treats subagents, MCP, and cost optimization as one frame, and this mod fits that view by exposing spend while work continues. For managers watching budgets, that ambient visibility matters more than decoration.

Claude Skins changes only appearance, which makes it the safest first install for cautious teams. The open-source package offers seven themes including Noah and Tokyo, swapping colors and style without touching behavior. Because it is display-only, there is no extra model spend and little operational risk. It is also a good test that mods load correctly on your setup before adding mods that intercept permissions or rewrite calls.

You Should Know, built by Anthropic , adds a second reader process that scans for buried critical lines people usually miss. The demo centers on a Kling-class video-generation call whose price arrives through the API provider, surfacing a cost detail hidden deep in output. That extra reader consumes additional tokens, unlike most other mods that run on deterministic code with no model call. Use it when contracts, limits, and pricing details matter more than raw speed.

Cost and security guards: cache, files, blast control

File Tree is fully deterministic and costs zero tokens , yet it removes constant navigation friction. A right-side panel lists project files while you work, and a touched file shimmers to show activity before a saved file turns green. The demo moves from production code to a multi-document tax-file example with many pages. That visual confirmation helps reviewers trust what changed without opening every tab.

Cache Tax tackles the prompt cache , the short-term memory that avoids full re-reads of large context. On a normal subscription the cache lives about one hour, and a cold cache forces a full reload that eats limits. The mod shows an estimated token and API cost before sending, while the /keepwarm command pings the session before the cache goes cold. The ClaudeCodeGuides pricing guide explains the math: cache reads cost 0.1x base input, five-minute writes cost 1.25x, and one-hour writes cost 2x, so a Sonnet-class example at 3 dollars per million tokens pays 3.75 dollars for a five-minute write and 0.30 dollars per read, breaking even after about two reads and then saving roughly ninety percent.

Blast Radius, also built by Anthropic , is a guardrail for risky commands such as deleting a folder. It pauses execution, lists affected files, and requires a human to press Proceed or Cancel before anything destructive runs. The warning is grounded in real horror stories where even Opus-class models wiped projects. Keep this mod enabled when agents gain file-system powers across important repositories.

Mods that learn and show the work

Reflect brings micro-RLHF into daily editing with almost no friction. Trigger phrases such as no or this does not look good open a slim rule strip offering to save a permanent preference. One click writes that rule to CLAUDE.md or a skill file for future runs. Small corrections therefore compound into a personal coding standard instead of disappearing after each session.

Terminal Browser places a real web view beside the chat inside the terminal, so builders can review generated pages and GitHub pull requests without leaving the keyboard flow. Replay Theater, an Anthropic sample mod, adds a replay button that steps through every file edit from the last turn. It works for code and article drafts alike, making reviews and demos easier to follow.

Build your own mod in plain words

The closing idea is that anyone can describe a wish in plain words while the built-in plugin authoring skill scaffolds a new mod. Hot reloading shows the result in-session, and save-as-plugin installs it permanently, since mods stay temporary by default. Mods cost nothing extra by default, though Claude Code needs a paid plan and model-calling mods consume usage. The AIDataInsider marketplace roundup adds context: Claude Code launched in May 2025, passed a 500 million dollar run-rate with tenfold growth in three months, and the /plugin command remains in public beta with marketplaces added through plugin marketplace add, alongside community builders such as Dan Avila and Seth Hobson with more than eighty subagents.