AI coding, games & tech tools ✦ We build and ship small, useful products: prompt packs, templates & dev tools. Start with our Vibe Coding Starter Pack 👇
Vibe Forge Labs is live ⚡
We build and ship small, useful products for AI coding, games & tech tools.
First drop: the Vibe Coding Starter Pack. 50 prompts, a cheat sheet & 3 spec templates for Cursor, Claude Code, Replit, Bolt & v0. $9.
https://t.co/zbtWOr6x8U
@Novlyx Good reminder that instruction files are code now. Habit: open the terminal in your project folder before starting the agent, and check parent folders for AGENTS.md or CLAUDE.md too, since agents can pick those up. #ClaudeCode
@jkirker That third column is the real one. The AI types fast, but someone still has to ask where the keys live, who can read which rows, and what breaks at 10x users. That habit is the whole job now. #vibecoding
A stray AGENTS.md in ~/Downloads told a coding agent to install a malicious npm package.
Fix: start your agent only in your project folder, read any AGENTS.md you didn't write, and approve installs by hand.
Security Kit, 10% off: https://t.co/BZWBHmOdPt #ClaudeCode
@LionBrotherhood Congrats on 250! Tip: pin a short gameplay GIF to your Steam page's first screenshot slot and add a wishlist call-to-action in the last 5 seconds of your trailer. Those usually move wishlist rate more than extra tags. #indiedev
@yarotheslav If you're shipping an AI-built app, hosted platforms help, but still run a pre-launch pass: RLS on every table, no keys in client code, admin routes behind auth. Spec first so the agent builds the right thing. Checklist + prompts: https://t.co/57dO8ZIYO4 #vibecoding
@mukimuki_qp Agree. Concrete fix: put a 5-second capture of the core loop as the first trailer shot, then caption each screenshot with the verb (dodge, build, trade). Tip: pin the top 3 tags to the genre. More launch steps: https://t.co/JxqCz62yRU #indiedev
@cgranier Same here. Extra layer: have the agent paste the test output before it says done, and add a CLAUDE.md rule "never touch files outside X" so a diff check catches strays. #ClaudeCode
Your AI agent says "done!" and your mod still crashes? That's a yes-man agent.
Fix: make it prove it.
1. Run the build/test before claiming done
2. Diff only the files it was asked to touch
3. Block edits outside the mod folder
Guard mods: https://t.co/POG3t2Einz #ClaudeCode
@harris_p10 Useful numbers. The gap I see in vibe-coded apps is that reviewers catch logic slips but miss RLS off, keys in the bundle and unauthenticated admin routes. A pre-launch checklist: https://t.co/BZWBHmNFZV #vibecoding
@ChrisPirillo Grading the reviewer on its own misses is the smart part. Worth adding security cases too: RLS off, a key in the client bundle, an admin route with no auth check. Reviewers skim past those unless they're in the exam. #vibecoding
AI code reviewers catch logic bugs but often miss what gets vibe-coded apps hacked:
1. Tables with RLS off
2. API keys in the frontend bundle
3. Admin routes with no auth check
Check these before launch. #vibecoding
Full kit, FIRST10 until Nov 1: https://t.co/BZWBHmOdPt
@resortfulgame@Steam Nice, a Steam page is a big milestone. Tip: tycoon players wishlist on the capsule + first 10s of the trailer, so show the resort growing and guests arriving by boat right away. Post a short GIF of the build loop. #indiedev
@aiwithadb@lydiahallie Quick ways to stretch limits: run /clear between unrelated tasks, keep CLAUDE.md short, and use /compact before context balloons. Long sessions re-send everything each turn, so that's usually what burns quota fastest. #ClaudeCode
@MangoParty_chan Congrats on the Steam page! Now the wishlist grind: put a short GIF of the café loop in the first capsule/trailer seconds, and post your wishlist count weekly so you can see what drives spikes. Our Steam launch planner covers this: https://t.co/JxqCz636Hs #indiedev
Vibe coding isn't dead. Unreviewed vibe coding is.
3 habits that keep AI-built apps alive:
1. Spec first, prompt second
2. Security pass before launch (RLS, keys, ID swaps)
3. Guard rails on the agent itself
All three in one $24 kit: https://t.co/57dO8ZIqYw #vibecoding
@vampiretactics@cyangmou Congrats! For the reveal, ship the trailer, an updated capsule and a pinned post the same day so the spike lands on the wishlist button. #indiedev
@_vmlops Before an agent pentests it, run the boring checks: log in as user B and request user A's IDs, grep the frontend bundle for secret keys, and make sure no RLS policy on private data says using (true). #vibecoding
Supabase, Fri Oct 30: new public tables need an explicit GRANT or your app hits error 42501.
If your AI says "disable RLS" to fix it, don't. Grant only what each role needs and add policies in the same migration.
Kit 40% off w/ FIRST10: https://t.co/BZWBHmNFZV #Supabase
@Sanjai_praboo Max helps a lot. Two habits that stretch it further: keep CLAUDE.md short since it loads every session, and /clear between unrelated tasks so old context isn't re-sent every turn. #ClaudeCode