Each skill is a focused, progressively disclosed procedure
A skill adds value only when it triggers for the right task, supplies knowledge the agent lacks, and leads the work through a repeatable procedure. Generic explanation, broad menus of equal choices, and unnecessary reference loads consume context without improving the result.
Observable check
the description states both the task outcome and when to use the skill; the body keeps one coherent workflow, gives a default when several approaches exist, preserves non-obvious gotchas, and includes a verification loop; conditional detail lives in directly linked files loaded only at the branch that needs it; and the instructions name capabilities rather than harness-specific tools.