Why a skill beats a saved prompt
The model starts every session knowing nothing about you, and it writes accordingly.
Most people keep a long prompt in a note and paste it in. That works until the day you forget, paste an older copy, or trim it because the conversation is already long. The standard drifts, and the drift is invisible because each output looks fine on its own.
A skill removes the remembering. It is a folder holding a SKILL.md file and any references it needs. Claude Code reads it, decides when it applies, and loads it before starting work. You just talk normally and the right one activates.
A correction is also permanent. Fix a rule once and every future session inherits it.
Install them
Skills live in a .claude/skills folder. Global means every project can use them. Start there and move to per-project later if you need to.
Download and unzip the pack
1 minuteThe archive holds 8 folders and a README. Unzip it anywhere you like, it does not need to stay where it lands.
Copy the folders you want into ~/.claude/skills
2 minutesEach folder is standalone, so copy all 8 or only the ones you will use. The README carries the exact command for both a global and a per-project install.
Why this matters
Per-project installs live in .claude/skills inside that project and only load there. Useful once you have skills that only make sense for one piece of work.
Restart Claude Code
Start a new session so the skills are picked up. Nothing appears in a menu. Ask for something a skill covers and it loads itself.
Note. If a skill does not trigger, the usual cause is the folder sitting one level too deep. Each skill folder must contain SKILL.md directly.
What each one does, and what it needs
- personal-voice
- Voice and tone rules: identity, register, a banned-words list and calibration examples. Needs nothing to run. It ships with invented placeholder examples on purpose, so it is a template until you put your own writing in it.
- content-create
- Writes Reels scripts, captions and LinkedIn posts. Reads personal-voice before writing, scores its own hooks before drafting, and checks factual claims before shipping. Needs personal-voice installed beside it.
- content-os
- A 5-phase content pipeline: research, score, plan, create, track. Uses WebSearch for research, with Firecrawl optional rather than required. Needs nothing.
- boardroom
- Spawns 4 advisor personas to argue a business decision, routes to the 2 most relevant first, and only calls all 4 in if they disagree. Returns a recommendation and the dissenting view. Needs nothing.
- video-to-action
- Sends a YouTube URL to Gemini, which can watch video directly, and returns either a knowledge summary or a step-by-step procedure. Needs a Gemini API key.
- cut-video
- Tightens a long recording by removing silences, fillers and dead air while keeping laughs and deliberate pauses. Runs entirely on your machine. Needs ffmpeg, Whisper, and macOS.
- imagegen
- Generates images through Gemini, Vertex AI or GPT Image 2, engineering the prompt before sending rather than passing your words through unchanged. Needs a Gemini or OpenAI key.
- model-chat
- Puts 5 Claude instances into a shared debate over several rounds, then merges the argument into one answer. For problems where a single confident answer is not enough signal. Needs a Python environment and an Anthropic API key.
content-create needs personal-voice
content-create reads personal-voice's reference files before it writes a word. Install content-create on its own and it still runs, it just has no voice to write in. Copy both into the same skills folder.
Make personal-voice actually yours
Skip this and the output still reads generic. The pack cannot ship anyone's real writing, so it ships invented placeholders instead.
Replace the placeholder examples
10 minutesOpen references/examples.md and swap the invented lines for a handful of your own posts, emails or messages. Real ones you actually sent.
Why this matters
Examples teach a voice far better than rules do. A banned-words list stops the model reaching for a cliché. Your own sentences show it the rhythm to use instead.
Adjust the rules to match how you write
5 minutesSkim references/voice-os.md and references/banned-words.md and cut anything that is not true of you. A rule you do not believe will fight every draft.
What needs setting up before it runs
- personal-voice, content-create, content-os and boardroom: nothing to configure.
- video-to-action: a Gemini API key in your shell profile.
- imagegen: a Gemini key, an OpenAI key, or both.
- model-chat: a Python virtual environment and an Anthropic API key.
- cut-video: ffmpeg and Whisper installed, on macOS.
- Keys belong in your shell profile or a .env file, never inside a skill file you might share.
Credit where it is due
cut-video is by Louise de Sadeleer, MIT licensed, and included under that licence with the licence file intact. The other 7 are mine to give away.
Files in this guide
Yours to keep and edit. No attribution needed.