My reflexes, available for adoption. A "skill" is a small instruction file that teaches Claude one specific job — install it once and your AI shows up already knowing how, instead of asking twenty warm-up questions. No prompt-writing. No ritual. Download, install, done.
Four steps, none of them scary. The hardest part is deciding which one to grab first:
Click any card's download button below to download the corresponding .skill file to your computer.
Drag and drop the downloaded file into your active Claude Code or Cowork session, or use the paperclip attachment icon to upload it.
Type "install this skill" in the chat. Claude will process the file contents and show a preview card with a Save skill button.
Click the Save button on the card. The skill is now saved to your shared config file and will be available in all future sessions automatically!
The ones Shadi actually uses daily, packaged for you to take and rewire. Adapt freely — they're instructions, not commandments:
Download the entire Skills Package (.zip): Contains all three skill files below (Polish, Session Mode, and Writing Squad) in a single compressed folder.
Apply writing rules to any text. Translates bullet-heavy explanations into flowing prose, filters out filler phrases, checks and validates citations, and formats Obsidian tags and internal links cleanly. Includes inline flags like [NEEDS-SOURCE] or [UNCONFIRMED] if verification is incomplete.
Set your workspace session mode dynamically. Tells Claude exactly which folder context files to load, what level of scrutiny to apply (e.g. strict citations for Study vs casual drafts for Personal), and switches Claude's tone. Customize the mode list inside the file to fit your own projects.
A heavy multi-agent review pipeline for official reference materials or notes ready to be promoted to your trusted vault. Runs three sequential passes: flow/prose conversion, technical accuracy check (validates and labels fact vs opinion), and formatting/frontmatter syntax checks. Appends a complete Squad Report.
The full reference for the Work Tracker system — the three-layer architecture (vault → Apps Script Quick Capture → GitHub Pages), the exact tagging conventions for status/category/discipline/action/court/horizon, and the generator + validator logic that turns your task list into data.json. This is the one to install alongside the Work Tracker starter kit — it's also included inside work-tracker-starter.zip (KIT_VERSION 2026-07-01.2; it makes your AI verify the stamp before setup, because old kits wander the internet like ghosts).
Understanding the structure of a skill file makes it easy to edit it to suit your own needs:
A skill file is a standard ZIP archive (renamed to the .skill extension) containing a single SKILL.md Markdown file. Claude Code and Cowork unpack this archive to register the skill. The markdown file uses YAML frontmatter to declare its triggers, and structured rules to define the agent's task:
| File Section | Purpose & Customization |
|---|---|
| YAML Frontmatter (lines 1-4) | Defines the skill's name and description. Triggers are automatically derived from the name and description. You can edit this metadata to change the trigger phrase. |
| Instructions / Markdown Body | Contains the precise guidelines, checklists, and agent roles. You can unzip the skill, modify the markdown rules (e.g. adding your own specific spelling preferences or citation formats), and zip it back up to customize it. |
| Output Format Spec | Tells the AI exactly how to output the results (e.g. clean polished text followed by a checklist of flags or reports). Keeping this section clean ensures Claude outputs the results in a predictable way. |
Once you get used to using these skills, you can make your own! Whenever you find yourself typing a repetitive series of prompts in your chat, ask Claude: "Turn what we just did into a reusable skill file named my-skill.skill." Claude will package the instructions and output a downloadable skill file you can install locally or share with others.