A living library of agent skills, forged and tested by the Agentic AI Bootcamp cohort. Jerusalem High-Tech Foundry x COMCEC.
Overview
The Skill Foundry A shared, living library of agent skills — built, tested and used by this cohort Browse the live catalog → Agentic AI Bootcamp · Jerusalem High-Tech Foundry (JHF) · In partnership with COMCEC Every skill in here was Nothing lands on a promise. If a skill is in skills/, it worked at least once, on a real machine, and you can see the proof. Keep contributing after Week 10. Found a skill that saved you an hour at work? Forge it here. Restart your agent. Then — don't name the skill: "Write me a viral LinkedIn post about finishing the bootcamp." If the skill was written well, it fires on its own. That is the whole test. Where does ~/.agents/skills/ live on my machine? Check with → tab. Your skill should be listed there with its description. If it isn't, the frontmatter is malformed — VS Code skips bad files silently.
README
The Skill Foundry
A foundry is where raw material gets forged into something useful. That is what this repo is for.
🔥 jhf-skills.azurewebsites.net — the live catalog
Every skill in here was built by someone in this cohort, tested on a real task, and merged only after the evidence was reviewed. Nothing lands on a promise. If a skill is in skills/, it worked at least once, on a real machine, and you can see the proof.
This repo outlives the course. Keep contributing after Week 10. Found a skill that saved you an hour at work? Forge it here.
🔥 Use a skill (30 seconds)
git clone https://github.com/Alaaldin97/Agentic-AI-Skill-Foundry.git
cp -r Agentic-AI-Skill-Foundry/skills/ ~/.agents/skills/
Restart your agent. Then describe the symptom — don’t name the skill:
“Write me a viral LinkedIn post about finishing the bootcamp.”
If the skill was written well, it fires on its own. That is the whole test.
📚 The catalog
→ jhf-skills.azurewebsites.net — the live site. Search by what you need done, copy an install command, and see every skill currently in review so you can go and review one.
| 🌐 Live catalog | jhf-skills.azurewebsites.net |
| 📄 Markdown index | INDEX.md — auto-generated on every merge |
| 🗂️ Browse by folder | skills/ |
| 🚀 Upload yours | UPLOAD-A-SKILL.md — step by step |
| ✍️ What makes a good one | CONTRIBUTING.md |
| 🙋 Request one | Open a skill request |
| 👀 Review one | Open pull requests |
The site rebuilds itself. Merge a skill and it appears — no deploy step, nothing to remember.
✍️ Contribute a skill
You submit a Pull Request, not an issue. Here’s why that matters:
| An issue | A pull request |
|---|---|
| A conversation that gets closed | Files that get merged |
| Your work dies in a thread | Your work is git clone-able forever |
| Nobody can run it | Anyone can install it in 30 seconds |
| No review | You get reviewed — and you review others |
| Ends with the course | Grows after it |
Four steps:
# 1. Fork on github.com, then clone YOUR fork
git clone https://github.com/YOUR-USERNAME/Agentic-AI-Skill-Foundry.git
cd Agentic-AI-Skill-Foundry && git checkout -b skill/my-skill-name
# 2. Copy the template and fill it in
cp -r skills/_TEMPLATE skills/my-skill-name
# 3. Check it before you push — same script CI runs
python scripts/validate_skill.py skills/my-skill-name
# 4. Push and open a PR
git push origin skill/my-skill-name
→ Full step-by-step walkthrough: UPLOAD-A-SKILL.md — includes a no-terminal option and troubleshooting → What makes a good skill: CONTRIBUTING.md
🚦 The merge bar
Three files, all required. CI blocks the merge if any is missing or malformed.
| File | What it must contain |
|---|---|
SKILL.md |
Frontmatter with name + description. The description must describe the symptom a user types, not the feature you built. |
EVIDENCE.md |
A real transcript — the prompt you typed, the output you got, on your machine. Plus one case where the skill correctly stayed dormant. |
meta.yml |
Author, date, category, and which client you tested on. |
Why the evidence file is not optional
Anyone can write a convincing
SKILL.md. The question is whether it fires when it should and stays quiet when it shouldn’t. Both directions are graded. A skill that triggers on everything is worse than no skill at all — it burns context on every single turn.
🧭 What makes a good skill here
Skills come in five shapes. Pick the smallest one that solves your problem:
| Shape | Contains | Use when |
|---|---|---|
| Instructions only | Just SKILL.md |
The model already knows how — it just needs to be told when and in what order |
| + a method | SKILL.md with a forced process |
The model rushes to output and skips the thinking |
| + scripts | scripts/ |
Something must happen the same way every time |
| + knowledge | resources/ |
Facts the model does not have and cannot guess |
| + a runtime | A whole framework | The output needs a real toolchain to exist |
A rule in a prompt is advisory. A gate in code is enforced. If it must happen, put it in a script — not in a sentence.
📁 Repo layout
skills/
_TEMPLATE/ ← copy this to start
SKILL.md
EVIDENCE.md
meta.yml
/
SKILL.md ← required
EVIDENCE.md ← required
meta.yml ← required
scripts/ ← optional
resources/ ← optional
scripts/
validate_skill.py ← run this before you push
build_index.py ← regenerates INDEX.md, the gallery, and the live feed
web/ ← the live site, deployed to jhf-skills.azurewebsites.net
index.html
skills.json ← generated; also read by the find-skills skill
docs/
gallery.html ← offline copy of the catalog
🔄 How the site stays current
you merge a PR
→ GitHub Actions runs build_index.py
→ regenerates INDEX.md, gallery.html and web/skills.json from skills/
→ zips web/ and deploys to Azure App Service
→ verifies the live URL returns 200, or fails the run
Open pull requests are fetched live in the browser from the GitHub API, so a skill shows up under In review the moment it is submitted — no deploy, no wait.
Recommended Tools
Try a different keyword or remove a filter.
Install
npx skillfish add alaaldin97/agentic-ai-skill-foundry