plicara labs researchhttps://plicara.ai/research/Studies, tools and perspectives from Adrian Tame’s independent AI research lab.enThe harness is the signalhttps://plicara.ai/research/harness-is-signal/As models improve, some of today's agent machinery will disappear. The parts that govern permissions, execution, recovery and accountability probably will not.https://plicara.ai/research/harness-is-signal/Wed, 02 Sep 2026 00:00:00 GMTWhat agent skills are actually made ofhttps://plicara.ai/research/agent-skill-programming-languages/1.9 million agent skills on GitHub. Only 12 in 100 come with any code attached, and when they do it is usually Python, whatever the skill is about.https://plicara.ai/research/agent-skill-programming-languages/Sun, 30 Aug 2026 00:00:00 GMTA safety filter moved our benchmark by seven placeshttps://plicara.ai/research/refused-at-random/A provider content filter silently deleted 29.3% of one model's answers on our regex benchmark, and the blanked calls were not missing at random. Depending on how a harness handles them, the same model finishes 3rd, 10th or 4th out of 11. Running k=3 rather than a single sample is what caught it.https://plicara.ai/research/refused-at-random/Thu, 27 Aug 2026 00:00:00 GMTWhat language are agent skills written in?https://plicara.ai/research/agent-skill-languages/3.8 million AI agent instruction files on GitHub, read by a model that speaks every language. So why are 85.3% of them in English, and what is changing?https://plicara.ai/research/agent-skill-languages/Mon, 24 Aug 2026 00:00:00 GMTwhether a regular expression is dangerous depends on whether anyone ever ran ithttps://plicara.ai/research/whether-anyone-ever-ran-it/Eleven models write regular expressions, scored three ways, and the safety screen then points at half a million patterns from shipped packages. Vulnerability tracks whether a pattern was ever executed, not whether a human or a model wrote it.https://plicara.ai/research/whether-anyone-ever-ran-it/Fri, 21 Aug 2026 00:00:00 GMT