Hermes-Jev-Skills reaches 853 stars as an agent decision skill library
A skill set iterated through daily use with 40+ agents, covering model selection, skill selection, memory filtering, screening unsafe instructions, and guiding browser and desktop actions.
Wow! 🤯 My Hermes-Jev-Skills repo is 853 stars!! https://github.com/kerpopule/hermes-jev-skills I’m continuing to test/iterate/merge daily as I burn through both my ChatGPT Pro and Claude Max plans with my 40+ agents daily…just want you to know these skills are constantly battle tested and iterated upon. I feed new repos and articles to my agents daily to make this better and better. Here’s what we got so far: - Choose models: Pick an appropriate model for the task. - Choose skills: Find the right instructions or workflow. - Filter memory: Pick relevant information from retrieved notes. - Screen unsafe instructions: Flag attempts to manipulate the agent through retrieved content. - Guide research: Rank search results, choose the next search, and judge whether enough evidence exists. - Trim long conversations: Choose what to keep, summarize, or drop when needed. Not our default handoff method. - Triage messages: Identify urgency and whether human attention is needed. - Sort email: Separate replies, updates, promotions, sales, and spam. - Guide browser actions: Choose the next page action. - Guide computer actions: Choose the next desktop action for a separate driver to execute. - Choose stronger help: Select a frontier model for genuinely hard work. - Monitor delegated work: Identify progress, stalls, questions, or completion.