← All guides

Guide / SKILLS

Four Claude Skills to use first — plan, write, debug, ship

Stop opening Claude from zero every morning. Install these four free skill types once — brainstorm, humanizer, debug, verify — then run them at the start of real work. Links, install commands, and when to use each.

6 min read
  • claude-code
  • claude-skills
  • ai-agents
  • developer-tools
  • workflows

You commented SKILLS, so here is the full pack: what each skill is for, a real install path, and when to run it in a normal workday.

The reel’s point is simple. Blank Claude every morning is the expensive habit. Skills are install-once folders of instructions (SKILL.md) that make the agent show up already briefed — for planning, writing, debugging, and shipping.

What a “skill” actually is

A skill is not a new model and not a chat plugin with a subscription wall. It is a reusable package — usually a SKILL.md plus optional scripts — that Claude Code (and other agents) can load when you invoke it.

Official overview from Anthropic:

Two install locations you will see everywhere:

  • Personal (all projects): ~/.claude/skills/<name>/
  • Project (shared with the repo): .claude/skills/<name>/

If your setup uses the plugin marketplace instead of raw folders, that is fine — same idea, different install path.

The four — day-to-day map

# Skill type When you run it What it stops
4 Brainstorm (ask-first) Start of day / new task Building junk before you know the goal
3 Humanizer Emails, posts, client notes Robot voice and fake TED-talk tone
2 Debug (systematic) When something breaks The random-fix spiral
1 Verify before ship Before send / merge / post The “looks done” lie

Habit that actually pays: close every task with the verify skill. That is the #1 from the reel for a reason.


4 — Brainstorm (interviews you first)

Use this when the task is fuzzy and you would otherwise paste a wall of hopes into a blank chat.

Good real pack: Superpowers includes a full methodology stack (brainstorm → spec → build → ship), not just one file.

/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers@superpowers-marketplace

After install, use the brainstorm / planning skills that come with the pack before you ask it to “just build the app.”

What “good” looks like: it asks you questions first. If it immediately dumps a 40-file architecture, that is not this skill working — that is cold Claude again.


3 — Humanizer (robot out of the text)

Use this on anything a human will read: email, LinkedIn, landing copy, README intros.

Solid open skill:

Claude Code plugin path:

/plugin marketplace add blader/humanizer
/plugin install humanizer@humanizer

Manual (any agent that reads skill folders):

git clone https://github.com/blader/humanizer.git ~/.claude/skills/humanizer

(Or copy only SKILL.md into ~/.claude/skills/humanizer/ if you prefer a thin install.)

What “good” looks like: fewer “delve / landscape / robust” piles, fewer fake-inspirational closers. If it still sounds like a keynote, run it again with a short sample of your writing.


2 — Debug (stops the random-fix spiral)

Use this when something is broken and you are one more “try this” away from making it worse.

Good real pack: Superpowers (and similar agent methodology packs) include systematic debug / investigation skills — the point is process, not vibes.

Pattern you want, whatever the skill name in your install:

  1. Reproduce
  2. Locate
  3. Hypothesize
  4. Fix one thing
  5. Prove it

What “good” looks like: it refuses to shotgun five unrelated changes. If your agent rewrites half the repo on the first broken test, you do not have a debug skill loaded — you have panic mode.


1 — Verify before you ship (use this first on the finish line)

Use this before you hit send, merge, or post.

This is the reel’s #1 on purpose. The expensive mistake is not “I need better models.” It is “I thought it was done.”

Same Superpowers / methodology world:

What “good” looks like: a checklist against the claim (“does the guide link work,” “did tests run,” “did we invent a number”). If the agent says “looks good” with no checks, that is the lie the reel is roasting.


Install once — then use first

Suggested daily loop:

  1. Morning / new work → brainstorm skill (questions before code or copy)
  2. Anything written for humans → humanizer
  3. When it breaks → debug skill
  4. Before you ship → verify skill

That is the whole product of the video: not “380 free tools,” four jobs with names.

Bigger libraries (if you want more later)

These are optional. Do not install hundreds on day one.

Curate. A skill you never invoke is just disk noise.


Related guides on this site


Short version

Need Skill type Start here
Clear plan Brainstorm Superpowers
Natural writing Humanizer blader/humanizer
Broken code / agent mess Debug Superpowers
Done-for-real Verify Superpowers + anthropics/skills

Install once. Run them before the blank chat. Comment SKILLS anytime you want this page again.

Keep reading

More guides.

Next

AI Lab & shipped work