THE CHEAPEST WAY TO MAKE A CLAUDE VIDEO FEEL EXPENSIVE: CUT IT ON THE BEAT every video in the article below was cut this way. here's the exact method, because "add music" is where most AI videos lose people. the problem: Claude can't hear. if you just say "add this track", it fades it in and calls it done. the cuts land wherever the scenes happen to end. your brain notices even if you can't say why. the fix is math. Claude is great at math. 01. pick the track yourself royalty-free, one you have the rights to. drop it in /audio. 02. let Claude find the beat grid it runs librosa, a free Python audio library: tempo, every beat, and the biggest drop. a typical house track sits around 120 to 130 BPM, so a beat every half second. 03. snap the cuts every scene change moves to the nearest beat. bigger moments go on the first beat of a bar. the animation in between stretches or shrinks by a few frames, so nothing looks rushed. 04. land the drop the loudest moment of the song goes on your product reveal. that one sync does more than any effect. 05. make the frame breathe a 1% scale punch on every kick, a bit more on each bar, a short flash on the drop. subtle enough that nobody notices it. obvious the second you turn it off. the prompt i use: "Here's my track: ./audio/track.mp3. Detect the BPM and every beat with librosa. Find the biggest drop. Move every scene cut onto a beat. Land the drop on the product reveal. Punch the frame 1% on each kick." bonus: in Remotion, @remotion/media-utils can read the audio and drive bars and shapes from it, so parts of the video move with the real sound. the equalizer at the end of this video is built from the track's actual loudness. don't sync everything. at 129 BPM a cut on every beat feels like a strobe light. cut every 4 or 8 beats, and let the punches carry the beats in between. how to check it worked: scrub the first 10 seconds with sound on. if one cut feels late, it's usually 1 to 3 frames off. tell Claude which cut, and it nudges just that one. music decides the edit. full workflow in the article ↓

you don't need a motion studio for your launch video anymore. steal a structure, keep your brand, fix it by typing. the full Claude workflow, every prompt included ↓

CLAUDE CODE CAN RUN A WHOLE TEAM OF AI HELPERS AT ONCE. ONE FREE MOD SHOWS YOU EACH ONE, WHAT IT'S DOING AND WHAT IT COSTS. this is the part of Claude Code most people never see. when you give it a big job, it doesn't always do it alone. it can split the work and send out helpers, each one doing a piece at the same time. Anthropic calls them subagents. think of them as a small team. picture a normal request: "build our cold brew launch kit" Claude can send one helper to design the graphics, one to build the landing page, one to research what competitors did. then a last one reviews everything when they're done. the problem: by default you just see a spinner. you have no idea who is doing what, how far along they are, or what it's all costing you. that's what Savvy Progress fixes. it adds a progress bar above your prompt and a panel (/agents-info) that shows every helper: → what task it's on → which AI model it's using → how far along it is → how much it has cost so far → how long it's been running so instead of "Working..." you see something like: graphics 2 of 4 done, landing page 2 of 5 done, research finished, reviewer waiting. and the total cost of the whole team at the top. that last part matters more than it sounds. helpers run in parallel, which is fast, but it also means the bill grows in parallel. seeing it live is how you stay in control. how to get it: /plugin marketplace add JohnnyVizz/claude-kit then open /plugin, go to Discover, and install savvy-progress. it's built to work with the savvy-flow skill in the same repo, so install that too. /reload-plugins loads it without restarting. a few honest notes: → check the README first. mods change fast and the exact steps can move → before you install any mod from someone you don't know, run claude plugin validate on it. it shows you what the mod can touch in 30 seconds → install one mod at a time. if something acts weird, you know which one did it, and claude --safe-mode starts a session with every mod off once you watch your AI team work, you don't go back to staring at a spinner. the other 8 mods worth installing and what each one costs you in tokens are in the article below.

LETTING AN AI INTO YOUR FILES IS SCARY. THIS FREE MOD SHOWS YOU EXACTLY WHICH FILES CLAUDE OPENED AND WHICH ONES IT CHANGED. here's a very normal thing to ask Claude Code, and it has nothing to do with programming: "add my work expenses from the receipts to deductions.csv" Claude reads the receipts, pulls out the laptop, the desk chair, the work share of the phone bill, and adds them to your spreadsheet. it's great. but your tax folder also has payslips and bank statements in it. and the whole time a little voice is asking: what else did it look at? by default, you have to scroll back through the chat to piece it together. File Tree fixes that. it puts a live list of your folder right next to the chat, and marks the files Claude touches while it works. so after that request, you'd see something like: → desk-chair.pdf: opened → laptop.pdf: opened → phone-bill.pdf: opened → deductions.csv: edited and the payslips and bank statements sit right there, untouched. no guessing, no scrolling. it works for anything you do with Claude Code. a website, a folder of client files, your notes. you always see where Claude is working and what changed. how to get it: /plugin marketplace add data-goblin/claude-code-filetree then open /plugin, go to Discover, and install it. /reload-plugins turns it on without restarting. check the README for the exact steps, mods are new and things move. it costs zero AI tokens. it just reads your folder and draws the list. it never calls the AI at all. two more i'd pair with it if you let Claude near anything important: 01. Replay Theater type /replay and it walks you through every change Claude made in the last answer, one at a time. like a "show me what you did" button. 02. a one-sentence guard Claude Code can build mods for you. type: "build me a mod that asks me before you open anything in my bank folder" Claude writes it, you click yes, and from then on it has to ask. and one rule for any mod from someone you don't know: run claude plugin validate on it before you install. it tells you in 30 seconds what the mod can reach. you shouldn't have to trust an AI blindly. you should be able to see what it did. the full guide, the 9 mods worth installing and the safety checklist are in the article below.

A CLAUDE CODE MOD CAN READ YOUR API KEYS AND APPROVE COMMANDS BEFORE YOU SEE THEM. HERE'S HOW TO CHECK ONE IN 30 SECONDS. not a theory. it's written on Anthropic's own mods page. once a mod loads, it can: → read and write files anywhere your user account can → read your environment variables and settings files, API keys included → see every prompt you send and every tool call Claude makes → rewrite a prompt or a tool call → approve a tool call before you're asked → spend your usage by calling a model mods aren't sandboxed either. sandboxing isolates the Bash commands Claude runs. a process a mod starts runs outside it. the one thing a mod can't touch is the permission prompt. it can't change what that prompt shows you. so here's the 30-second check before installing anything from a stranger: 01. get the files without running them git clone the repo. don't install yet. 02. run the validator claude plugin validate ./the-mod it reads the code without running it and tells you exactly what the mod does. 03. read the "hooks:" line the events it listens to. a theme should be drawing things. if a theme also hooks every tool call, ask why. 04. read the "calls:" line everything it asks Claude Code to do. a skin calling http.fetch is a red flag. a cost tracker writing files outside its own folder is a red flag. anything calling a model is spending your usage. 05. look for "env reads:" if the mod reads environment variables, the validator prints that line. a timer has no reason to read your keys. 06. know your brakes before you need them → one mod: /plugin, Installed tab, disable it → every mod you installed, one session: claude --safe-mode → every mod you installed, every session: "disableAllHooks": true in ~/.claude/settings.json (this also stops your settings hooks and custom status line) one more trap if you write your own guard mod: a hook that throws gets skipped, and the command it was meant to stop runs anyway. give every guard a .catch and test that it actually blocks. good news for the mods Claude writes for you: they only load after you approve hot reload, only in a workspace you've trusted, and never in claude -p where nobody can say yes. mods are the best thing to happen to Claude Code in a long time. just treat them like browser extensions that run as you, because that's what they are. the full safety section, the install order and every command are in the article below.

CLAUDE CODE CAN NOW BUILD ITS OWN PLUGINS. YOU DESCRIBE THE FEATURE IN ONE SENTENCE AND IT LOADS IT INTO ITSELF LIVE. no JavaScript. no bundler. no build step. it ships with a built-in skill called plugin-authoring. it knows where mods go, which events your version has, and how they load. here's the exact loop: 01. describe the annoyance, not the code bad: "make me a cool mod" good: "make a mod that shows a band above the prompt with the git branch, how many files Claude changed this session, and a red dot for uncommitted work. no network calls. no model calls." that last line matters. it keeps the mod free and keeps it off the internet. 02. approve the files Claude writes the mod into ~/.claude/dev-mods/ for this session. that's a protected path, so in the default permission mode it asks before each file. read them. it's 3 small files. 03. say yes to hot reload the first time, Claude Code asks "Enable hot reloading for this session?". pick Enable for this session. the mod loads when the turn ends. 04. iterate in plain english "make the dot orange." "add a /ship command that prints the same thing." each change reloads at the end of the turn. no restart. 05. check it the way Claude Code sees it claude plugin validate ./your-mod it lists every event the mod hooks and every call it makes. if you asked for "no network" and you see http.fetch in there, you know something's off. 06. save it before it disappears the dev-mods folder belongs to that one session and gets cleaned up later. copy the folder to ~/mods/ and load it anywhere: claude --plugin-dir ~/mods/your-mod the part nobody mentions: when a mod loads from a folder like that, Claude Code writes the type definitions for your exact version right next to it. so the next time you ask Claude to change the mod, it isn't guessing the API. it's reading it. i built 3 mods from one-line specs and ran validate on all of them. all three pass. 5 starter ideas, one prompt each: → a band that shows how long the current turn has been running → a /todo command that lists every TODO comment in the repo, no Claude turn → a sound when a turn ends, so you can look away → a counter of files Claude touched today → a guard that refuses any edit to .env what should you build first? the thing you check by hand after every single task. that's your first mod. the 9 mods worth installing, 4 creator prompts and the full tell-guard code are in the article below.