Hermes Agent completes the skill retirement cycle: use it and remember, don't use it for 30 days and it will be downgraded, archive after 90 days

AIMPACT News, April 30 (UTC+8), according to Dongcha Beating monitoring, Nous Research’s open-source AI agent framework Hermes Agent has added a Curator feature, completing the final step in skill lifecycle management. Previously, one of Hermes Agent’s core selling points was that agents could automatically create skill documents after completing complex tasks, accumulating experience for future reuse, but skills only entered and did not exit. Some users reported that after accumulating 146 self-built agent skills, the skill list in system prompts for each conversation took up about 4,400 tokens, with useless skills piling up and no elimination mechanism. The Curator handles this in two steps. The first step is a deterministic state machine that does not call the model: each self-built skill has counters such as usage count and last used time, with skills unused for 30 days marked as stale, and those unused for 90 days moved to an archive directory, from which they can be restored at any time. The second step is model review, where every 7 days, during agent idle times, an auxiliary model scans the remaining skills, merging overlapping functions and patching outdated content, with up to 8 iterations. The review runs in an independent prompt cache, not affecting ongoing user conversations. The Curator only manages self-built skills, not built-in or installed from the skill marketplace. It does not perform automatic deletion; the worst case is archiving. Users can lock important skills with the pin command; locked skills cannot be modified even by the agent’s own skill_manage tool. It is enabled by default but can be turned off. The PR is about 2,200 lines of code, with all 56 tests passing, merged into main on April 29, but not yet released with the official version. (Source: BlockBeats)