Semantic Anchors are words — established terms like Clean Architecture, Cockburn Use Cases or MECE — that activate rich, well-defined concepts inside any modern LLM.
You don't install them. You don't import them. You type them.
Semantic Anchors is a curated catalog of well-defined terms — methodologies, frameworks, and principles like MECE, arc42, and BLUF — that act as shared vocabulary for communicating with large language models. Naming an anchor activates a dense, pre-computed cluster of knowledge inside the model, so a single word steers the output precisely instead of a long prompt.
Featured in: heise online (DE) — heise online — Semantic Anchors shorten the context for agentic coding (German). As seen on: HMZE (DE) — HMZE Podcast — Beyond Vibe Coding (German), rabauer.dev (EN) — rabauer.dev — Live coding session: From Vibe Coding to a Real Java App (English), Beyond Code (DE) — Beyond Code Podcast, Episode 16 with Felix Becker — Semantic Anchors, Loop Engineering and architecture (German). Adapted in: Claude Code Ultimate Guide (EN) — Claude Code Ultimate Guide by Florian Bruniaux — adapts this catalog for Claude Code workflows (English, CC-BY-SA-4.0), opencode-semantic-anchors (EN, archived) — opencode-semantic-anchors by Jens Grote — an opencode plugin that enforced anchors and contracts; deprecated in 2026 in favour of OpenCode agent configuration (English), speq-skill (EN) — speq-skill by marconae — carries a Semantic Anchors documentation page linking this catalog (English), wandelbots-nova (EN) — wandelbots-nova by Wandelbots — ADR convention informed by MADR and this catalog (English). Listed on: Good AI List (EN) — Good AI List by Chip Huyen — a daily-updated directory of open-source AI repositories with 500+ GitHub stars, listing this catalog since October 2026 (English).
Anchors for how a text or talk is built and pitched: structure (Pyramid Principle, BLUF, Inverted Pyramid), style (Strunk & White, Gutes Deutsch, Plain Language) and the audience (Curse of Knowledge, Four-Sides Model). Combine one structure anchor with one style anchor; two structures in one prompt pull against each other.
Story structures from screenwriting and narrative theory. Each one fixes where the turning points sit, so naming one shapes the plot of a story, talk or case study. Pick one; mixing structures blurs the arc.
Principles and patterns for designing code: SOLID, the GoF patterns, coupling and cohesion. Name the pattern and the model applies its known structure and roles; name the principle and it reviews code against it. For visual and interface design, see UX & Visual Design.
Conventions and practices for the daily work around the code: commits, branches, versions and changelogs, the steps of a change, and continuous improvement. The conventions are quick to name and easy to check, which makes them a good fit for a CLAUDE.md or AGENTS.md.
Anchors that change how the model talks with you rather than what it produces: asking instead of answering, uncovering the real problem behind a question. Useful in a system prompt or at the start of a session.
Anchors for structuring and maintaining documentation: what kind of document to write (Diátaxis), how to build and review it (Docs-as-Code) and how to word it (Plain Language, Simplified Technical English).
Methods for capturing, organising and passing on knowledge, from personal task systems to an LLM-maintained wiki. Name the method and the model sorts notes, tasks or learning material by its rules.
Anchors about semantic anchors themselves: the theory behind them and the spectrum from loose term to precise method. Read these to understand why naming works and where it stops working.
Thinking tools for analysing a problem and its causes before solving it. Some bring a fixed procedure (Five Whys, A3, DMAIC) that the model works through step by step; others are heuristics (Occam's Razor, First Principles) that change how it weighs an answer.
Formats and techniques for eliciting and writing down requirements: user stories, use cases, EARS and the workshops that find them. A named format gives the model a template and a checklist at once.
Styles, patterns and documentation methods for software architecture. Name a method (arc42, C4, ADR) for the structure of the documents, and a style or pattern for the structure of the system.
Methods for monitoring processes with data: control charts, the rules for reading them, and improvement cycles built on them. Useful for metrics dashboards and incident analysis.
Frameworks for setting goals, prioritising and making decisions under uncertainty. Many of them classify the situation first (Cynefin, Wardley Mapping) and then suggest a fitting response.
Testing styles, test doubles, reviews and security and safety standards. Naming the style (Chicago or London School TDD) decides how tests are written; naming the standard (OWASP, WCAG) decides what a review checks.
Visual styles, design systems, layout and typography rules, usability heuristics and accessibility. Combine one style or design system with rules for the details and with WCAG 2.2 AA. Name controls by their common name, which The Component Gallery (opens in a new tab) lists with its aliases, and fix how they behave with the WAI-ARIA Authoring Practices Guide (APG) (opens in a new tab). Not every design system is an anchor: of the many design systems in the Component Gallery (opens in a new tab) only a few are in this catalog. For one that is not, give the LLM a link to its documentation, and if the system offers them, its design tokens or MCP server as well. All examples side by side: Design Styles Compared.