AI Guides › Step-by-step guides

Five Claude Code Skills, Installed One at a Time

By Nigel Guy · 6 min read

Most people hear about a good Claude Code skill and install five in one afternoon. Then the first session goes oddly, and nobody can say which one caused it. Skills change how Claude behaves, so each one needs its own test before the next goes on.

The rule: install one skill, run one real task with it, keep it only if the result is visibly better, then add the next.

This guide covers five skills and plugins from the Claude Code ecosystem: Find Skills, Superpowers, Claude Mem, Task Observer and Impeccable. They are third-party projects, not Anthropic products. Commands and repository details below were checked against each project's own page on 2026-10-04, and they change often.

Before you start

Where skills live, so you can find and remove them: personal skills sit in ~/.claude/skills/<name>/SKILL.md, project skills in .claude/skills/<name>/SKILL.md, and plugin skills are namespaced as /plugin-name:skill-name. Run /skills inside Claude Code to see what is loaded (Claude Code skills docs).

Step 1 — Find Skills, so you search before you install

Find Skills is a skill from Vercel Labs that teaches Claude to look for an existing skill when you ask "is there a skill for X". It uses the skills command line tool and the skills.sh directory.

  1. In your terminal, run: npx skills add vercel-labs/skills --skill find-skills
  2. Add -g if you want it available in every project rather than just this one.
  3. Restart Claude Code, then ask: "Is there a skill for reviewing pull requests? Show me what you find and the install command, but do not install anything."

You can also search without Claude: npx skills find [query], then npx skills list to see what you have installed.

Note that the CLI sends anonymous usage data by default. Set DISABLE_TELEMETRY=1 or DO_NOT_TRACK=1 if you would rather it did not.

Step 2 — Superpowers, so Claude plans before it types

Superpowers is a plugin of linked skills: brainstorming, planning into small tasks, test-driven development, code review, git worktrees and a four-phase debugging process.

  1. Inside Claude Code, run: /plugin install superpowers@claude-plugins-official
  2. Or use the author's marketplace: /plugin marketplace add obra/superpowers-marketplace then /plugin install superpowers@superpowers-marketplace
  3. Start a new session and ask: "Do you have Superpowers? List its skills."
  4. Give it a small real feature and say: "Brainstorm this with me before writing any code. Ask questions one at a time."

Expect it to slow you down at the start. That is the point.

Step 3 — Claude Mem, so you stop re-explaining your project

Claude Mem records what Claude does in a session, summarises it, and feeds relevant pieces into later sessions. It stores data locally in ~/.claude-mem/, with an optional cloud backup, which you can leave off.

  1. Run: npx claude-mem install
  2. Alternatively, in Claude Code: /plugin marketplace add thedotmack/claude-mem then /plugin install claude-mem, then restart Claude Code.
  3. Do not use npm install -g claude-mem. According to the project, that only installs the library and does not set up the hooks.
  4. Work for a session, close it, open a new one and ask: "What did we decide about [TOPIC] last time?"

Its search follows three layers: a compact index first, then a timeline, then full detail only for the entries you need. Wrap anything sensitive in <private> tags so it is excluded from storage.

Step 4 — Task Observer, so repeated corrections become skills

Task Observer, from Eoghan Henn (rebelytics), watches for three things: corrections you keep making, manual work that could become a skill, and gaps in its own method. It logs observations and proposes changes. You approve before anything is edited.

  1. Download the one-skill-to-rule-them-all repository from github.com/rebelytics/one-skill-to-rule-them-all.
  2. Copy the task-observer folder to ~/.claude/skills/task-observer/ (all projects) or .claude/skills/task-observer/ (one project). Keep the references/ and scripts/ folders inside it.
  3. Install is not activation. The author says description matching alone under-triggers, so add the activation instruction from references/environments.md to your CLAUDE.md.
  4. After several real sessions, check that skill-observations/observation-log/ exists. At the end of a session ask: "Any observations logged?"

Step 5 — Impeccable, so interfaces stop looking generated

Impeccable, by Paul Bakaus, is a design skill with a set of commands and rules aimed at common generated-interface habits such as default fonts and purple gradients.

  1. Run: /plugin marketplace add pbakaus/impeccable
  2. Run /impeccable init so it learns your project's goals and constraints.
  3. Try /impeccable audit on an existing page first, then /impeccable craft for a new one.

The project's page gives command and rule counts that differ between sources, so check the repository for the current list rather than trusting a number.

Check it worked

Skill Proof it is working
Find Skills npx skills list shows it, and Claude answers with a command rather than installing
Superpowers Claude asks questions and proposes a plan before writing code
Claude Mem A fresh session recalls a decision from an earlier one
Task Observer An observation log folder appears after real sessions
Impeccable /impeccable audit returns specific findings on your own page

What to skip

Guardrails

Sources

All 751 AI guides · JulieMango plans from £17/mo