LLM-Coding/Semantic-Anchors

What it solves

Large language models (LLMs) know a lot from their training data, but they need the right trigger words to access that knowledge. Explaining complex ideas from scratch wastes tokens and can be imprecise. Semantic Anchors solves this by giving you a set of well-known terms, methodologies, and frameworks — like "TDD, London School" — that instantly activate the relevant knowledge in the model.

How it works

The project maintains a catalog of 230+ semantic anchors, each documented in its own file. Anchors are organized into 12 categories (e.g., Design Principles, Testing & Quality) and can be filtered by professional role (like software-developer, qa-engineer). The catalog is available on an interactive website with search and visualizations. For local or smaller LLMs that don't know every anchor, the README provides guides to inject explicit definitions into agent instruction files for tools like Claude Code, Codex, or Cursor. The project also encourages contributions: you propose a term, they test it against quality criteria, and then GitHub Copilot helps create the anchor file.

Who it’s for

Anyone who works with LLMs — developers, architects, QA engineers, product owners, technical writers, and more. It's especially useful for teams using local or smaller models that need explicit definitions, and for anyone who wants more consistent and efficient prompts.

Highlights

• A catalog of 230+ semantic anchors across 12 MECE categories.
• Interactive website with search, role-based filtering, and visualizations.
• Guides for integrating anchors into Claude Code, Codex, Gemini CLI, Cursor, GitHub Copilot, and Windsurf.
• Simple contribution flow: propose an anchor, it gets tested, and Copilot helps author it.