writing-for-agents

Front-load the leading trigger word, state what the material is, and list one trigger per distinct branch the document handles. Cut synonyms and identity the body already carries, since every word of an always-loaded description costs context on every turn.

Treat each line as a context pointer naming out-of-context material and the condition for reaching it. Keep steps inline, push branch-specific reference behind pointers, and prune lines that restate what the environment already shows.

Choose model-invoked when the agent or another skill must reach it autonomously, paying permanent context load for the description. Set disable-model-invocation: true for skills only fired by hand, trading discoverability for zero context load.

Premature completion happens when a step's completion criterion is vague and later steps remain visible, pulling attention toward being done. Sharpen the criterion's bound first; only split the sequence across a real context boundary if the rush persists.

Split by sequence when visible post-completion steps tempt the agent to rush the current one, or by invocation when a distinct trigger word deserves independent reach. Each split spends context or cognitive load, so the cut must earn it.