Apodex just put a name on the agent skill benchmarks keep missing. The team calls it working capability. Sustained, verifiable progress toward a real objective, which is the part demo reels rarely show. Their technical report hit Hugging Face Daily Papers this week, and the framing deserves a minute. Real work runs for days across files, searches, and code, where step 40 depends on step 12 surviving. Most benchmarks were built to grade the final answer. An agent that reasons brilliantly and…