You Can Now Change How Claude Code Works: Here Are 5 Mods
The new extension system can monitor context, interrupt risky commands, change the interface, and even alter how Claude Code handles project instructions.If long drafts keep outgrowing your chat window, you are not alone.Mods are small add-ons for Claude Code that change what it does or shows, and…
Annons
Annons
The new extension system can monitor context, interrupt risky commands, change the interface, and even alter how Claude Code handles project instructions.If long drafts keep outgrowing your chat window, you are not alone.Mods are small add-ons for Claude Code that change what it does or shows, and they help when you want a warning, a guard, or a calmer review view without coding. They launched on October 1, 2026, and they run inside your session.You can pick the one that fits your work and try it today.What a mod actually isA mod is a small JavaScript or TypeScript function that lives inside a plugin and watches what Claude Code does as it happens.You need Claude Code v2.1.287 or later, in the CLI or the Code tab of the desktop app, because mods run inside the session. They are on by default, so there is nothing to switch on.A mod can step in at a few useful points:Rewrite a prompt before it reaches the model, or redact secrets from tool output before Claude reads it.Hold, rewrite or retry a tool call, and approve or deny a permission request.Draw a pane beside the chat or a band above the prompt, with buttons and fields you can use.Add a /command that runs your function at once, even while Claude is working.Because one hook can record and another can show, a mod can count turns in the background while it draws the count on screen. You can read more in the Mods overview docs.Faisal Ahammad also walks through installing plugins in the Claude Code CLI in the video below.How I picked these fiveEvery pick had to meet the same three tests, so you can compare them by job, not by rank.I kept mods that meet these rules:Show one everyday job, like watching context, checking risk or reviewing edits.Work without coding, because Claude can write and install the mod from a plain-word ask.Come from Anthropic examples or built-in sources, including Token Weather at about 80 lines and the built-in /diff pane and AGENTS.md support.That mix shows different needs without ranking them.Find your mod fastMatch the job to the mod:Token Weather: best for watching context fill up during long drafts.Blast Radius: best for pausing before a risky command to see what it would touch.Replay Theater: best for stepping through the last turn edit by edit.Diff: best for reading uncommitted changes beside the chat as Claude works.AGENTS.md Support: best for carrying shared house style and rules into each session.1. Token Weather: watch context fill upToken Weather is a live forecast of the context window for writers working through long drafts. It suits you if you paste chapters or briefs and want an early warning before the window fills.It draws one line above the prompt with five levels:☀ Clear for under 25%☁ Cloudy for 25–49%☂ Showers for 50–74%☇ Storm for 75–89%↯ Compact soon for 90% and upEach level shows the percent, the tokens used out of the window, and a sparkline of recent turns, so you can act before Compact soon appears.What you seeThe band sits above the prompt and refreshes when a turn ends. It pairs the icon and word with the share used, the token count, a sparkline of recent turns, and the gain from the last turn.How it helpsLong drafts fail late, when the window fills, and Claude starts to compact. The forecast lets you split, summarize, or start fresh while you still have room.You are editing a 12,000-word guide built from keyword research for your blog when the band moves from Clear at 18% to Showers at 67% after you add sources, then to Storm at 81% after a full rewrite.You see the jump, so you save the draft, ask Claude to summarize the sources, and continue in a fresh session with room to finish. The line gives you time to act instead of hitting the limit mid-edit.Keep in mindYou only get a fresh reading after a turn ends, so it cannot warn you mid-turn. Plan your big pastes before the band moves, not during.2. Blast Radius: preview risky commands firstBlast Radius is a guard for scary shell commands for anyone who lets Claude run file actions. It suits cautious runners who want to see what would change before anything runs. It pauses the call and asks you to choose Proceed or Cancel after you review the list.What you seeA review pane opens beside the transcript when a risky command appears. It lists the files and folders the command would touch, with buttons to proceed or stop.How it helpsHoping a delete is safe keeps you tense, while a preview lets you catch a wrong folder. You can stop a reset that would wipe drafts or allow a cleanup that only touches temp files.Keep in mindA preview cannot make a risky command safe on its own. You still need to read the list and cancel when the scope looks wrong.3. Replay Theater: replay last turn's editsReplay Theater is a calm review view for writers and marketers after a big rewrite. It suits you if you want to step through the last turn edit by edit instead of scanning a wall of text. It replays each change in order so you can spot tone shifts before you keep them, with the last turn always in focus.What you seeThe replay view steps through the edits from the latest turn in sequence. Each step shows what changed, so you can read at your own pace.How it helpsBig rewrites hide small voice changes that matter to readers. A jewelry seller reviewing a product page can pause on a rewritten guarantee, keep the clearer headline, and ask Claude to restore the original promise.Keep in mindStepping through edits does not undo them by itself. You still need to ask Claude for the fix or keep the version you prefer.4. Diff: review uncommitted changes calmlyDiff is a built-in mod that shows uncommitted changes beside the chat. It suits non-coders who want to read what Claude changed before accepting it. It refreshes as Claude edits and keeps the view file by file so you can scan calmly.What you seeChanged files line up in a pane beside the transcript. Each file opens into hunks that update while Claude works.How it helpsAccepting blind feels risky when copy is on the line. A marketer checking a landing page can read each hunk, catch a swapped headline, and ask for a fix before publishing.Keep in mindYou see only uncommitted changes here, not older history. You can turn the pane off in /plugin or replace it with your own version when you prefer a different view.5. AGENTS.md Support: carry project instructions alongAGENTS.md Support is a built-in mod that loads shared project notes so Claude follows house style. It suits teams and repeat writers who reuse voice and rules across sessions. It places the notes where Claude reads instructions, with project instructions doing the work.What you seeShared notes load as project instructions when a session starts. You notice steadier wording because Claude has the rules in front of it.How it helpsPasting style notes each time wastes time and invites drift. A newsletter team can keep tone, banned phrases, and structure in one file, so each draft starts on-brief.Keep in mindBy default, it loads where the project has no CLAUDE.md of its own. You can also run it beside CLAUDE.md, or pare back to managed or CLAUDE.md-only setups when your org requires it.Ask Claude to build it for youYou do not need to code to get a mod. Describe what you want in plain words, and Claude writes it and reloads it in your session.Ask in this order for the smoothest path:Say what to show and where, in one short wish.Allow hot reloading for the session when Claude asks, so you see each tweak at once.Ask for small changes until the view reads right.Say keep this mod and copy the folder out, so it survives cleanup.Get updates from the Claude directory or with /plugin, as shown in Getting started with mods.Paste a wish like this:Make me a mod called [name]: [what it should show or do], shown [where it should appear].It should update [when it updates].Keep the wording plain for a non-coder.To keep it, say “keep this mod in this session” and then “copy the folder out for later install”. If you run a regular send, a publishing cadence you control pairs well with a small mod that keeps drafts on-brief.Check a download before you install it:claude plugin validate ./some-mod# shows hooks: and calls: lines for reviewIn a session you can also run /plugin install token-chart@your-org to install, then /reload-plugins to load it without restarting.Stay safe with modsA mod runs with your permissions, not in a sandbox. It can read and write files, start programs, and reach the network, so install only from authors you trust.It can reach further than you expect:Read secrets in environment variables and settings files.See every prompt you send and every tool call Claude makes.Change a prompt or a tool call, approve a call before you are asked, and spend usage on your plan.Before you install, get the files and run claude plugin validate ./some-mod to list hooks and calls, as shown in Decide whether to trust a mod. Teams get an extra guard because sec-default loads first and keeps managed rules out of reach, and your session shows 1 mod active when one is loaded.Start with one small changeStart with one small change that saves you time, like a context forecast or a calmer review view. Ask Claude for that tweak in plain words and keep only what you use.Open Claude Code, run /plugin, and pick one mod from trusted sources to try.This story is published under the Generative AI publication. Connect with us on LinkedIn and follow Zeniteq to stay in the loop with the latest AI stories. Let’s shape the future of AI together!You Can Now Change How Claude Code Works: Here Are 5 Mods was originally published in Generative AI on Medium, where people are continuing the conversation by highlighting and responding to this story.