A humble set of bite-sized tips for product and design people who want to vibe build and ship their passion projects.
Set a condition with Claude Code's /goal command and Claude keeps taking turns on its own until it's met, with an evaluator model checking after each turn. Plus why pairing a goal with ultracode is the fastest way to torch your token budget.
Claude writes a small program that runs dozens of subagents in the background, has them cross-check each other, and hands back one answer, plus how to tell when that's worth the tokens and when it's overkill.
Rules you added to CLAUDE.md months ago were written for an older model, and some of them now slow the current one down. Which rules to keep, which to delete, and a prompt that has Claude audit the file for you.
The permission mode that runs Claude without the constant approve prompts, with a background safety check that blocks the genuinely dangerous stuff like force-pushing, prod deploys, and leaking your secrets.
Model providers block actual harm, but they don't know what your product is for. Keeping your agent on topic falls to you, and for most vibers a tighter system prompt is the whole fix.
How to write a system prompt that holds up to real users: the 5 parts to include, a template for a customer support bot, and a stress test before you ship.
If you installed Claude Code via npm and it stopped working, here's what changed and how to switch.
How you interact with your AI doesn't just determine what it builds, it determines how well it builds it. Here's what promotes calm vs. what triggers desperation, based on Anthropic's research.
Claude Code's permission system, the flag that skips it entirely, and the middle ground that actually makes sense for most people.
A Claude skill that reads your RFCs on session start, shows a status table, and logs progress when you wrap up. For when your vibe project outgrows manual tracking.
Compaction saves space but loses details. Here's what survives, what doesn't, and why /clear is usually better.
Prompt engineering is about what you say. Context engineering is about everything the AI already knows.
Claude Code's new side-channel lets you ask questions without interrupting whatever Claude is building.
Subagents report back to Claude. Agent teams coordinate with shared task lists and peer-to-peer messaging. How to use Claude Code's experimental multi-agent feature for complex builds.
Cowork brings Claude Code's agentic power - local file access, multi-step execution, scheduled automation - to people who'd never open a terminal.
Skills teach Claude repeatable workflows, load automatically when relevant, and only cost tokens when they're actually needed. Here's everything you need to know.
Claude Code isn't a single agent. It's an orchestrator managing specialist subagents behind the scenes. Here's what they are and how to work with them.
As of January 2026, Claude Code merged slash commands into skills. Here's what changed and how to migrate.
Running multiple Claude sessions at once - one writes, one builds, one reviews. How parallel sessions became part of my daily workflow.
LLMs are stateless - they don't remember between sessions. Once you stop fighting this and start working with it, everything gets easier.
Getting Claude to set up scheduled publishing so your static blog can queue posts like WordPress does.
How to connect GA4 to Claude Code so you can query your traffic data without touching the dashboard.
How I run multiple vibe projects in parallel without sacrificing quality or consistency.
There's a whole lineup of Claude products now, and they don't all do the same stuff. Here's what actually connects to what.
I asked Claude to explain why it created 25 hardcoded CSS values despite clear instructions not to. Here's what it told me.
The quick sanity check before you send your URL to anyone. Favicon, social previews, mobile - the stuff that makes you look amateur when you skip them.
How to actually check if your static site is secure, and why it probably already is.
I'd been manually asking Claude to install MCPs and write skills. Then a friend mentioned /plugin and I realized there's a whole ecosystem I missed.
PRDs tell Claude what to build. RFCs track where you left off so the next session can pick up without re-explaining everything.
How I built a skill that manages my entire content pipeline - auto-updating the backlog, spotting scheduling gaps, and suggesting what to write next.