Glossary
Key terms used throughout guide-LLiMes.
ADR (Architecture Decision Record) — A short document recording a significant technical decision: what was decided, why, and what alternatives were rejected. See ADRs.
Context window — The total amount of text an LLM can process at once, including system prompt, conversation, code, and your guidelines. Measured in tokens.
Guardrail — A constraint on LLM behavior: files it can’t modify, steps it must run, decisions that require human approval. See LLM Guardrails.
Guideline — An explicit instruction that gives the LLM context about your project. Dense, scannable, actionable. Not documentation for humans — instructions for the model.
Must-have — A guideline that prevents fundamental breakage. Without it, the LLM will produce code that doesn’t fit the project at all. See Must-Have Guidelines.
Nice-to-have — A guideline that matters at scale or in regulated domains. Adds rigor but isn’t needed until the project matures. See Nice-to-Have Guidelines.
Should-have — A guideline that reduces variance in LLM output. The model can work without it, but quality and consistency improve with it. See Should-Have Guidelines.
System prompt — Hidden instructions prepended to every LLM conversation. Your guidelines become part of this prompt (or are loaded alongside it) by the tool you use.
Token — The unit of text that LLMs process. Roughly 1 token = 0.75 words in English. Your guidelines consume tokens from the context window.
Token budget — The practical limit on how many instructions your guidelines can contain before the model starts ignoring or conflating rules. See Token Budgets.
Three-tier model — guide-LLiMes’ framework for prioritizing guidelines: must-have (prevent breakage), should-have (reduce variance), nice-to-have (scale and compliance). See Three-Tier Model.
Ubiquitous language — A shared vocabulary where every concept has exactly one name, used consistently in code, documentation, and conversation. From Domain-Driven Design. See Domain Glossary.