# Build your own Claude skill **The CV Method · The Playbook · v1 · Aug 2026** *by Corina · @the_cv_method · hello.thecvmethod.com · current as of Aug 2026 — the AI space moves fast, so double-check tool names* Teach Claude your exact workflow once — the steps, the rules, the voice — and reuse it forever with a single word. Here’s how I make mine, plus a fill-in-the-blank prompt that writes one in about five minutes. ## What you’re building A “skill” is just a **saved recipe**: a repeatable set of instructions plus a clear trigger word. Instead of re-explaining the same task every time — the report format, the way you like emails drafted, your brand rules, your review checklist — you write it down **once**, and Claude runs it your way, consistently. The best skills are narrow, specific, and honest about their guardrails. ## Rule zero A skill is a recipe with a trigger. Written once, reused forever. If you can’t name the trigger and the “done,” it’s a chat, not a skill — narrow it until you can. ## Step 1 · Spot the repeat *Start · the thing you keep re-typing* ### 01 · Find the task you explain to AI every week The report you always reformat, the way you like outreach drafted, your brand rules, your pre-ship checklist. That **repetition is the skill.** Pick the one task you’re most tired of re-explaining — that’s the one worth capturing first. If you’ve pasted the same instructions to Claude twice, it’s already a skill. Go make it one. ## Step 2 · Write the recipe Brief it like a new hire ### 02 · Inputs, steps in order, “done,” and the nevers Write the steps the way you’d brief a sharp new hire: the **inputs** it starts with, the **exact steps** in order, what **“done”** looks like, and the **guardrails** (what it must never do). Be specific — “use our navy + green palette,” not “make it on-brand.” Steal from your own past chats — the messages where you corrected Claude are your best rules, already written. ## Step 3 · Name it + give it a trigger So you can call it by name ### 03 · A short name and the phrases that launch it Give the skill a short name and the trigger phrases that should fire it — “draft an outreach email,” “run my review,” “make the monthly report.” The trigger is how you (or Claude) call it later **without re-explaining a thing.** ## Step 4 · Add examples + guardrails Examples are the cheat code ### 04 · One good example, one bad — and the hard nevers A single good example and one bad example teach more than a paragraph of rules — Claude matches patterns. Then spell out the hard **nevers**: never invent a number, never send without your approval, never use a client’s real name. Examples are the highest-leverage part of the whole skill. Spend your time here. ## Step 5 · Test it on a real task Run it, watch it drift, fix the recipe ### 05 · Every miss is a missing line Run the skill on a **real job** and watch where it drifts from what you wanted. Each miss is a line the recipe is missing — add it. Two or three passes and it’s reliable enough to trust. ## Step 6 · Save it + reuse Your process, captured once ### 06 · Drop it in a SKILL.md and call it forever Save the recipe as a **SKILL.md** in your skills folder, and it’s there for good — call it by name, tweak it as your workflow evolves. That’s the whole trick: your process, written down once, working for you every time after. ## The prompt · copy this *A · the SKILL.md skeleton (fill in the blanks)* ``` --- name: your-skill-name description: --- ## Trigger — say any of these "", "" ## When to use (and when NOT to) ## Steps 1. 2. 3. ## Guardrails (hard) - never - always ## Example Good: Bad: ``` The `name` is lowercase with hyphens and matches the skill's folder name (e.g. `your-skill-name/SKILL.md`). B · or have Claude write it for you (paste this) ``` Turn this into a reusable skill for me. Here's the task I keep re-explaining to you: Write it as a SKILL.md with: YAML frontmatter (name + one-line description), trigger phrases, when to use (and when not to), numbered steps, guardrails (the hard "nevers"), and one good + one bad example. Keep it tight and specific to how I actually work — no fluff. ``` ## Where to get skills **Official skills + the skill-creator: github.com/anthropics/skills** — clone it, or in Claude Code run **/plugin marketplace add anthropics/skills** and install a set (e.g. document-skills, example-skills). **Any skill:** install through the plugin marketplace in Claude Code (**/plugin**), or add it via the API. On Claude.ai, skills work on every plan (Free, Pro, Max, Team, Enterprise): turn on code execution, then upload the skill folder as a ZIP under **Customize → Skills**. In Claude Code, skills live in **.claude/skills/** in a project, or **~/.claude/skills/** for all your projects. **Build your own:** use the SKILL.md above — or the **skill-creator** from that repo, which scaffolds one for you. That's the whole point of this playbook. Docs: support.claude.com → “what are skills” · the open standard: agentskills.io ## Keep it honest & safe - **You approve** before anything sends or ships - **No invented** metrics - Real work stays **anonymized** - Keep **secrets out** of the skill file - Skills **propose** — you decide ## The starter checklist - The **one task** you re-explain most, picked - The **recipe**: inputs · steps · done · nevers - A **name** + trigger phrases - One **good** + one **bad** example - Tested on a **real task**, recipe patched - Saved as a **SKILL.md** you can reuse ### Want your workflow turned into skills for you? I turn the things you re-explain every week into a clean set of skills — your steps, your voice, your guardrails — so Claude runs your process the way you would. That’s the CV Method. Start at **hello.thecvmethod.com**.