Reliability concerns for LLM agents are typically bundled into one heavy framework that asks you to adopt prompting, tool routing, and runtime governance as a single dependency. Production teams want them à la carte. They want small primitives they can drop in around existing tool calls without buying into a new programming model. That observation is the design centre of agent-stack : six small, single-concern reliability libraries published independently to npm , PyPI , and the Model Context Protocol registry. Each library is zero-dependency, under 500 lines of code, and addresses one specific failure mode that production agent teams have to handle. This post is a tour of the six primitives, the cross-cutting invariants they enforce, and the trade-offs of "composable by inclusion" instead of "composable by framework." The six primitives Library Concern Failure mode it addresses AgentFit Context-window fitting Token-aware truncation. Pluggable tokenizers for OpenAI / Anthropic / open models.…