Lexicon

Agent A system where an LLM drives its own loop Back pressure A feedback loop where an agent verifies output before continuing Compaction Summarizing older context to make room in the context window Context engineering Designing what an agent sees and in what order Context poisoning Quality loss caused by stale or contradictory context Context window Everything a model can see in one call Dumb zone The high-fill portion of context where model reliability drops Hallucination Confident output that isn't true Harness The software wrapper that turns a model into a working agent LLM Large Language Model — a neural network trained to predict text RAG Retrieval-Augmented Generation Ralph Loop A bash loop that keeps spawning fresh AI agents until the work is done Skills A reusable capability an agent can invoke beyond raw tool use Token The smallest unit of text a language model processes Tool use Letting a model call functions