Workspace Indexing Plan
Current bottlenecks
Large monorepos slow down context gathering. File reads, path resolution, and repeated parsing add latency before the agent can plan or edit.
Proposed changes
Add semantic file cache
src/workbench/orbit/common/contextCache.ts
LRU cache keyed by workspace revision. Store parsed symbols and imports so agents skip redundant reads.
Parallel index workers
src/workbench/orbit/browser/indexWorker.ts
Shard indexing by top-level folders. Stream progress to the agent panel in real time.
