Every time I set up Claude fresh — new computer, new workspace, new client project — there are ten skills I install before I do anything else. Not because they look impressive in a demo. Because they solve real friction I hit every week. Here they are, in the order I actually think about them.
1. deep-research
This skill fans out across multiple web searches, pulls primary sources, and cross-checks claims before writing anything. It kills the hallucination problem for research-heavy work. I use it any time I need to build a dossier on a prospect, verify a claim before publishing, or pressure-test an idea against real data. One prompt kicks off a full research loop instead of me manually running six searches and stitching the results together.
/deep-research What are the three largest pain points B2B coaches cite when trying to scale past $30k/month, with sources?2. code-review
You don't have to be a developer to use this. If you're running any automation, prompt chains, or custom tools — and someone else (or Claude itself) wrote the logic — this skill reads the diff and flags real bugs, not style nitpicks. It saves me from shipping broken automations to clients. Set effort level to high when anything touches money or live data.
/code-review --effort high3. update-config
This is how you make Claude actually remember your preferences without saying them every session. 'From now on, always format output as plain text, not markdown' — that's a hook, not a memory. Without update-config, that instruction evaporates when the session ends. With it, it gets written into settings.json so the behavior sticks. Install this first. Configure it before anything else. Every other skill works better once your environment is properly set.
4. spec-miner
This skill reads a codebase or document set and pulls out the implicit rules — the patterns, constraints, and conventions that aren't written down anywhere but are obviously true once you see them. I use it when I inherit someone else's project or when a client hands me a folder of assets and says 'build something like this.' It surfaces the spec that nobody wrote.
5. secure-code-guardian
If you're building anything that touches user data, API keys, or payment flows, this is non-negotiable. It audits code for security vulnerabilities — exposed credentials, insecure endpoints, injection risks. I run it on every piece of automation before it goes live with client data. One caught bug is worth more than the five minutes it takes.
/secure-code-guardian6. fullstack-dev-skills:project:execution:execute-ticket
This is the skill that turns a written ticket into working code without you babysitting every step. You give it a clearly scoped task — 'add an email capture form to the landing page that posts to this endpoint' — and it executes the full implementation. The key word is 'clearly scoped.' Vague tickets get vague output. This skill rewards you for writing tight requirements.
7. prompt-engineer
I use this to turn rough instructions into production-quality prompts. You describe what you want the AI to do in plain language, and this skill structures it properly — role, context, task, output format, constraints. The output is copy-pasteable. I use it every time I'm building a new automated workflow or writing a system prompt for a client's AI tool.
/prompt-engineer I need a prompt that takes a sales call transcript and outputs: 1) the three main objections raised, 2) the prospect's stated budget, 3) a recommended next-step message. Output should be plain text, under 200 words.8. debugging-wizard
Things break. Automations fail silently. APIs return unexpected responses. This skill diagnoses before it fixes — you get a literal evidence report (file, line, quoted output) before any solution is proposed. That discipline matters. It stops Claude from pattern-matching to a fix that sounds right but misses the actual problem. I use it whenever I hit an error I can't immediately explain.
9. schedule
This skill creates and manages cloud agents that run on a cron schedule. Recurring reports, automated checks, timed follow-up sequences — anything you want to happen on a clock without you manually triggering it. I use it to set up weekly performance summaries and to automate the parts of client delivery that are predictable and repeatable. The one thing to use it for: any task you're currently doing manually on a schedule.
/schedule every Monday at 8am: pull last week's funnel metrics from [source] and write a plain-text summary to [destination]10. verify
This is the skill that closes the loop. Once you've built something, verify actually runs the app and checks the behavior against what you said it should do. It's not just reading code and guessing — it observes real output. I run it before anything ships to a client. If verify passes, I'm confident. If it flags something, I fix it before it becomes a client support issue.
The order matters. Install update-config first to lock your environment. Then install the research and security skills before you write a line of code or send anything live. Build the safety layer before you build the speed layer.
How to install a skill
Skills in Claude Code live in your settings or are loaded through your Claude configuration. Most are available through the skill system built into Claude Code. To invoke any skill, type its name with a forward slash. If the skill isn't available, check your Claude Code version and make sure you're running the latest release — most of these are included by default or available in the skill library.
Start with two or three, not all ten at once. Get comfortable with deep-research, update-config, and verify first. Once those feel automatic, add the rest. You'll get more out of each skill when you're not context-switching between ten new tools at the same time.