Source Intelligence

DisclaimerUnofficial, and not affiliated with Anthropic. Nearly all of this is read straight out of what ships: npm bundles, captured prompts, published docs. Anthropic's own notes go in verbatim, marked as theirs. The rest is my reading, and every entry carries the strings behind it. If one looks wrong, vote it down and say why.

All of v2.1.187 Home All releases olderv2.1.186 v2.1.190newer
Claude Code v2.1.187

Per-context bootstrap data cache [In Development]

What

Bootstrap data (client configuration, model options, costs) will be cached on a per-context key (entrypoint + model + org UUID + CC version) rather than a single global cache slot, allowing different sessions or entrypoints to carry their own cached data without overwriting each other.

Status

Active — tengu_client_data_cache_key is a telemetry-only gate, not a feature gate. The per-context caching is written and read regardless; the gate only controls whether cache hit/miss/stale/changed stats are reported to telemetry.

Details
  • Cache key is composed of: entrypoint + model + org UUID + CC version — so CLI and IDE extension sessions never clobber each other's cached model/org config
  • New clientDataCacheSlots config key stores keyed entries; legacy clientDataCache (global single slot) is still read as a fallback when no matching slot exists (legacy_fallback: true in telemetry)
  • Telemetry tracks four outcomes per fetch: slot_hit, slot_changed (remote data differs from cached), legacy_fallback, and slot_stale (cache older than the stale threshold)
  • If cache hit and data is unchanged and not stale, the bootstrap write is skipped entirely
Evidence

New slot-keyed cache (search for "clientDataCacheSlots" or "tengu_client_data_cache_key")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

See this entry in the whole of v2.1.187 →