eve-learning-loop: agent-authored skills that resurface as native skills, plus self-curated memory (Turso/Postgres) #1014
BowTiedSwan
started this conversation in
Show and tell
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
I built an extension that closes the procedural self-improvement loop for eve agents: the agent writes its own skills, and they come back next session as native, load-on-demand skills.
Repo: https://github.com/BowTiedSwan/eve-learning-loop · npm:
npm i eve-learning-loopWhat it does
skill_managetool; a dynamic skill resolver surfaces them next session as<namespace>__<name>, loadable through the frameworkload_skilltool like any static skill.memorytool.session_searchreturns ranked snippets from past conversations.Everything is scoped per principal, and the store is bring-your-own-database: Turso/libSQL (FTS5) or Postgres (tsvector + GIN).
Two things I liked about eve while building it
skill.loadedevent for usage telemetry — turns out loads are first-class runtime actions, so a hook onactions.requested/action.resultnarrowingkind === "load-skill"gives per-skill use counts with no framework changes. The curator ages skills by last use or last edit, whichever is later.dependencies, not the consumer's.One gap worth flagging: extensions can't declare schedules, so the curator ships as copy-in templates (
cp node_modules/eve-learning-loop/templates/agent/schedules/*.ts agent/schedules/). If extension-contributed schedules land someday, that step disappears.Feedback very welcome, especially on the memory-as-prompt-injection guardrails and whether the frozen-per-session snapshot is the right default.
All reactions