SKILL: Cache Sweep Activation Fallback

Domain: extensions-mcp
Trigger: Apply when newly materialized extension sources must be activated immediately, but the live refresh can fail and needs a recoverable fallback. Source Pattern: Distilled from reviewed extension lifecycle and source-reconciliation patterns.

Core Method

Treat installation and activation as separate phases. After a background reconcile changes what is on disk, clear the marketplace cache, attempt a full activation refresh, and degrade gracefully if that refresh throws. The fallback path should clear plugin caches again, log the failure, and set a sticky manual-refresh bit so the system admits that runtime state is stale instead of pretending activation succeeded.

Key Rules

• Only attempt automatic activation after a materialized change that needs live components to catch up; steady-state or metadata-only work should not enter this path.
• Clear upstream caches before activation so the refresh reads fresh marketplace and plugin state rather than replaying stale memoized data.
• Wrap the activation step in a catch boundary that logs the error and preserves the newly materialized state on disk.
• If activation fails, clear the plugin cache with a specific reason string and raise an idempotent manual-refresh flag instead of retrying blindly.
• Keep the fallback sticky until a deliberate refresh consumes it; repeated failures should not thrash the state tree with duplicate writes.

Example Application

If a background extension installer clones a new command marketplace during startup, clear the source cache, try a full runtime reload, and if the reload fails mark the UI with a single “reload extensions” prompt while preserving the newly cloned files for recovery.

Anti-Patterns (What NOT to do)

• Do not swallow activation failures after install; users will assume the new extension is live while the runtime still points at stale caches.
• Do not skip cache clearing before the activation attempt or the refresh may rehydrate old data and fail in harder-to-debug ways.
• Do not loop automatic retries without a user-visible fallback; repeated refresh crashes should turn into a manual recovery prompt.