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
{{ message }}
Repository navigation
Refresh global, external, and HTTP skill sources without restart #37426
Project .opencode/skills content currently invalidates through the Location filesystem watcher. Global compatibility roots, explicitly configured external directories, non-git Locations, and already cached HTTP catalogs do not have a continuous invalidation path. Replaying an unchanged source registration also does not clear its cached content.
Acceptance criteria
Global OpenCode, .claude, and .agents skill edits refresh in loaded Locations.
Explicit external skill directories refresh without requiring an unrelated config edit or restart.
HTTP catalogs have a defined refresh trigger and observe a changed version while OpenCode remains running.
Skill removal clears model-facing guidance and slash visibility on the next safe boundary.
Tests document and cover project, global, external, non-git, and HTTP source behavior.
Partly done: #40954 added watchers for directory skill sources, including global and external folders outside the project and symlink targets, so changes there are picked up without a restart. HTTP skill catalogs are still only fetched on load. Closing this tracker. A focused issue for refreshing HTTP catalogs is easier to pick up if it's still needed.
Activity
fix(core): close skill-refresh review findings
Shalin-Shah-2002 commentedon Jul 24, 2026
Hi! I'd like to work on this issue if it's available.
I'll investigate the current skill refresh flow and submit a PR.
neriousy commentedon Oct 21, 2026
Partly done: #40954 added watchers for directory skill sources, including global and external folders outside the project and symlink targets, so changes there are picked up without a restart. HTTP skill catalogs are still only fetched on load. Closing this tracker. A focused issue for refreshing HTTP catalogs is easier to pick up if it's still needed.