🌐 llm-skills

@ CreativeAct Technologies, 2026

A model-agnostic, single-file framework for extending browser-based LLMs with specialized capabilities. Upload or paste one SKILL.md into any web-based AI interface to instantly activate structured workflows, output schemas, and fallback logic.

✨ Zero dependencies. Zero setup. Just Markdown.

Traditional agent skill systems rely on multi-file directories, progressive disclosure, and CLI-based execution. Browser LLMs operate differently:

• 📥 No file system access: Everything must be self-contained in one SKILL.md
• 🧠 Single-context loading: The entire prompt enters the context window at once
• 🔌 Model-agnostic: Works with Claude, ChatGPT, Gemini, Mistral, open-weight UIs, and local web frontends
• ⚡ Zero setup: Copy → Paste → Activate

This framework adapts the Agent Skills specification for direct browser upload, enforcing context efficiency and cross-model compatibility.

---
name: skill-identifier          # lowercase, hyphens only, ≤64 chars
description: Clear trigger + when-to-use guidance, ≤1024 chars
version: "1.0"                  # optional
tags: [domain, workflow]        # optional
context_priority: high          # optional: high|medium|low
---

1. Copy any skills/*/SKILL.md file
2. Paste into your browser LLM chat (Claude, ChatGPT, Gemini, etc.)
3. Prompt naturally: The LLM matches your request to the skill's description
4. Receive structured output per the skill's Output Format

💡 Tip: Keep skills ≤ 4,000 tokens. Put high-signal instructions first, reference data last

Use the browser-compatible validator:

# In terminal (Node.js)
node tooling/validate-skill.js skills/data-cleanup/SKILL.md

# Or paste into browser console (if running locally)

Or use the token counter to stay within context limits:

python tooling/count-tokens.py skills/data-cleanup/SKILL.md
# Output: 3,842 tokens ✓ (under 4,000 limit)

1. Fork the repo
2. Create skills/<your-skill>/SKILL.md following the template
3. Keep it ≤ 4,000 tokens and fully self-contained
4. Submit a PR with 2 test prompts that activate it

MIT — Use, modify, and distribute freely. Designed for open, interoperable AI workflows.