writing-for-agents

Write the description as a context pointer: front-load the leading word, state what the material is, and list one trigger per distinct branch the document handles. Cut synonyms that rename a single branch and identity the body already carries, since every word of an always-loaded pointer costs context on every turn.

A model-invoked skill keeps a description so the agent and other skills can fire it autonomously, paying permanent context load. A user-invoked skill sets disable-model-invocation: true, so only a human typing its name can invoke it, costing zero context load but requiring the human to remember it exists.

Split by sequence when visible later steps tempt the agent to rush the current one, and split by invocation when a distinct leading word should trigger a skill on its own or another skill must reach it. Splitting spends context or cognitive load, so the cut must earn it.

Progressive disclosure moves reference material out of the main file and behind a context pointer, loaded only when the pointer fires. Inline what every branch needs and disclose what only some branches reach, keeping the top of the document legible and protecting the information hierarchy.

Negation drags the forbidden behavior into context and makes it more available, since the strongly-activated concept overruns the weak negation modifier. State the positive target behavior instead, reserving prohibitions for hard guardrails paired with a positive instruction.