2

Source document

[2026-01-30 19:58] lisamegawatts: could do a weid abstraction where the agent gets skill badges by actually doing a task and then commiting the exact workflow to a file, like you complete quest and the archivist writes your tale of glory into the legend, which in this case is actually a list of skills that now any agent can perform, the skill is committed at time of completion and maybe programmatically analyzes tools used and sequence to pair them with skill as well as maybe time to completion for a measure of skill quality? and a description of how to perform the skill and maybe an intended use field [2026-01-30 20:00] lisamegawatts: dependencies on tools can be sidequests then, like you can't perform the spell or skill until you acquire the x tool [2026-01-30 20:00] lisamegawatts: https://tenor.com/view/think-winnie-the-pooh-thinking-pooh-think-think-think-gif-7743943814219302717 [2026-01-30 20:06] lisamegawatts: you could also have levels, so lets say omegas attempt to use browser takes 30 seconds because it tries multiple tools and fails, omega is a level 1 browser mage with one tool equipped. a level 10 browser mage maybe completes in 5 seconds because it has abstractions for every browser, programmatically detects environment or configs and equips tool, and can do browser automation rather than just fetch [2026-01-30 20:06] lisamegawatts: the agents could also compete and prove their worth by executing a skill more efficiently than other agent and gain rank [2026-01-30 20:10] lisamegawatts: so you can see both how well it functions and what tools it requires equipped. maybe i only use brave so a well equipped brave mage is for me, but if i do ui testing or something i might want a crossplatform level 10 browser mage [2026-01-30 20:14] lisamegawatts: probably should have both time and cost of execution as metrics, how many tokens, it seems to me you also need to specify the model so maybe that is an equipped tool. like the frontier model can do the skill quickly with these tools and this prompt, but the local one can't or does it much slower [2026-01-30 20:14] ajaxdavis: we still on google meets