AI Skills Overview

What are AI Skills?#

AI Skills are structured knowledge files designed to be loaded as context by AI agents (via MCP or similar tools). Each skill gives an agent deep, accurate knowledge of a specific part of the SkyCiv API — so it can build models, run analyses, generate drawings, and retrieve loads without hallucinating parameters or making up endpoints.

Rather than pointing an agent at general documentation, you load only the skills relevant to its role. A structural analysis agent loads the S3D skill; a drafting agent loads the CAD skill. The result is focused, reliable AI behaviour at a fraction of the context cost.

Open Source Skills Repository

All skills are open source and community-maintained at github.com/skyciv/skyciv-ai-skills

Skills Included#

Skill	What it lets your agent do
SkyCiv Core API	Foundation for every API skill — auth, session management, the request/response envelope, and shared call patterns. Start here.
S3D (Structural 3D)	Build, solve, repair, and query full 3D structural models — nodes, members, plates, sections, materials, supports, loads, load combinations, and results.
S3D Apps	Build custom, embeddable mini-apps that run client-side inside the S3D application — read/write the live model, react to the user's selection, and automate or generate model content.
CloudCAD	Generate 2D engineering drawings — floor plans, dimensions, gridlines, annotations, tables — with optional mapping into a structural 3D model.
Load Combinations	Define code-correct load cases and combinations on a structural model, with ready-made combination sets for the US, Europe, Canada, Australia, and India.
Load Generator	Look up wind, snow, and seismic loads/pressures for any location worldwide, across major design codes.
Run Quick Design	Call any of 150+ Quick Design calculators (steel, concrete, timber, aluminium, connections, foundations, loads) via one REST endpoint — includes a full catalogue plus schema and sample input/output for each calculator.
3D Renderer	Embed the client-side SkyCiv 3D Renderer to visualise a structural model and its analysis results interactively in the browser.
Schema Agent	Interpret an uploaded floor plan (DXF/DWG/PDF/image) into a precise structural schema that the S3D skill can build a model from.
QA Engineer (WIP)	Independent peer-review persona — checks a finished calculation or report for units, sanity, and completeness before it ships.

Recommended loading strategy#

All agents should load the Core skill — it covers auth, session management, and the API envelope.
Load only the domain-specific skills your agent needs. Narrower context = better focus.

S3D Agent       → Core + S3D
Draftsperson    → Core + CloudCAD
Loads Agent     → Core + Load Generator + Load Combinations
Full Stack      → Core + S3D + CloudCAD + Load Generator

Vibe Coding with AI Skills#

The fastest way to get an AI working with SkyCiv is to clone the skills repo directly into your project. Your LLM — whether that's Claude, GPT, Cursor, or anything else — will read off these skill files and have everything it needs to call our APIs correctly, first time.

git clone https://github.com/skyciv/skyciv-ai-skills.git

Then point your agent (or system prompt) at the relevant skill files:

You are a structural engineering assistant. Read the following skills before responding:
- skyciv-ai-skills/skyciv-api-v3/SKILLS.md
- skyciv-ai-skills/s3d-api/SKILLS.md

That's it. The LLM now understands the full S3D model schema, every API function, and how to correctly structure calls — no training, no fine-tuning, no guesswork.

Why this works#

Relying purely on an LLM to perform designs and calculations is not ideal - they are not deterministic and run the risk of hallucinations. Plugging into reliable FEA, design and calculation-based technology offers the perfect blend of man and machine. A well-built logic tool (LLM) built on reliable, verified calculations (SkyCiv API).

Each skill is written specifically to be consumed by an LLM, not a human. Parameters are precise, JSON examples are complete, and edge cases are documented inline. The result is:

No hallucinated endpoints — the agent calls real functions with real arguments
No missing required fields — schema coverage is comprehensive
No wasted tokens — load only the skills your agent needs for its role
Easy to keep current — pull the latest skills when the API updates

Works with any AI tool#

Tool	How to load skills
Claude Code / Claude	Paste skill content into your system prompt or use MCP file tools
Cursor / Windsurf	Add skill files to your `.cursorrules` or project context
Custom MCP server	Read skill files at startup and inject into `system` messages
OpenAI Assistants	Upload skill files as knowledge base documents
LangChain / CrewAI	Load skill text as agent tool descriptions or system context

How Skills Work with MCP#

Each skill file is a self-contained Markdown document. In an MCP-based setup, your server reads the relevant skill file(s) and injects them as system context before the agent processes a user request.

// Load skills relevant to this agent's role
const coreSkill = fs.readFileSync('skyciv-ai-skills/skyciv-api-v3/SKILLS.md', 'utf8');
const s3dSkill  = fs.readFileSync('skyciv-ai-skills/s3d-api/SKILLS.md', 'utf8');

const systemPrompt = [coreSkill, s3dSkill].join('\n\n---\n\n');

The agent then has full knowledge of the SkyCiv API for those domains without needing to retrieve documentation at runtime.

Contribute#

Skills are open source and community-maintained. If you find an error, a missing parameter, or want to add a new skill — open a PR.

github.com/skyciv/skyciv-ai-skills

Skills are versioned alongside the API. When an API change affects a skill, the skill file is updated in the same commit — so pulling the latest always keeps your agent current.