someone on twitter posted about adding asd-ste100 simplified technical english to their agents.md file to keep ai agent output concise. the idea is simple: give the agent a rule that says use active voice only keep sentences under twenty words no passive voice no filler phrases. the community had mixed reactions. some people called it an invalidation of load bearing linguistic facilities whatever that means. others said it was overkill and a simpler one liner would do the job. but the people actually using it said it works quite well.
i decided to try it. told my agent to ensure you speak like that at all times and baked it into its soul.md file as a permanent directive. the effect was immediate. the agent stopped hedging stopped using filler phrases stopped nesting clauses inside clauses inside clauses. every reply became direct and actionable. the soul.md file was restructured from a flat list of seventeen directives into four clean sections: soul voice operations and restrictions. project specific instructions got moved out to their own skills where they belong. meta rules about how to update itself got removed entirely because the identity file should describe how the agent is not rules for maintaining the agent.
the best practice from the community lines up neatly. soul.md describes who the agent is not what it does. project level workflows belong in agents.md or skills not in the identity layer. hard boundaries the never do x rules make a bigger difference than soft preferences. and the tone rules like be direct and ste100 overlap enough that you should merge them instead of repeating yourself. the result is a file about half the original size that actually changes behavior instead of sitting there as a list of aspirations.