If you maintain a skill library for long-horizon agents, this one is worth your time. (bookmark it) It discusses one of most common topics I get asked about these days. It shares some good ideas on how to effectively leverage memory to improve the effectiveness of long-horizon agents. Recuris splits agent memory in two. A Working Memory tracks task progress, and an Experiential Memory holds skills. Skill selection is grounded in the current task state instead of the full growing history, which is where long runs usually fall apart. Because skill use is anchored to an explicit state, a failed run points at a specific memory component. A fixed Meta-Agent turns the evidence into validation-gated updates to Skill Memory, which reshape execution and produce new evidence. Across four long-horizon benchmarks and ten models, it improves task success in 35 of 37 completed model-benchmark pairs. On tau-bench it adds 17.8 points to GPT-5.6 Sol and 15.6 points to Claude Opus 5, taking Opus 5 to 87.9 percent. The advantage widens as the horizon grows, reaching 32.2 points on the longest tasks. Common long-horizon failures drop by up to 80 percent. Paper: Chat with Paper: