You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently, installing skills is one-skill-one-platform-at-a-time. A skill sync command could reconcile state in bulk: install new, update changed, optionally remove stale.
Idea
Add skern skill sync to sync all skills from the registry to all detected platforms in one shot.
Open Questions — Decisions Needed
The value of a blanket sync is unclear given the project's strategic direction:
Dynamic skill loading vs bulk sync — The intended model is that agents dynamically install/uninstall skills as needed, keeping only the required skills at a platform's registry at any given time. A bulk sync contradicts this by pushing everything at once. Which model takes priority, or do both coexist?
Batch install/uninstall as an alternative — Rather than full sync, batch operations (skern skill install skill-a skill-b --platform all) may better serve the dynamic loading model. An agent could install multiple skills in one call without syncing everything.
Sync semantics — If sync is implemented, what happens with:
Skills installed on a platform but not in the registry (remove? ignore?)
Skills that were manually edited on the platform side (overwrite? skip? warn?)
Scope handling (sync user scope, project scope, or both?)
Context
Currently, installing skills is one-skill-one-platform-at-a-time. A
skill synccommand could reconcile state in bulk: install new, update changed, optionally remove stale.Idea
Add
skern skill syncto sync all skills from the registry to all detected platforms in one shot.Open Questions — Decisions Needed
The value of a blanket sync is unclear given the project's strategic direction:
Dynamic skill loading vs bulk sync — The intended model is that agents dynamically install/uninstall skills as needed, keeping only the required skills at a platform's registry at any given time. A bulk sync contradicts this by pushing everything at once. Which model takes priority, or do both coexist?
Batch install/uninstall as an alternative — Rather than full sync, batch operations (
skern skill install skill-a skill-b --platform all) may better serve the dynamic loading model. An agent could install multiple skills in one call without syncing everything.Sync semantics — If sync is implemented, what happens with:
Relationship to
--forceon install (Add --force flag to skill install for overwrite #33) — Does sync implicitly force-overwrite, or does it respect existing installations?These questions need resolution before implementation can proceed.
References
Analysis item 2.2