bc-telemetry-buddy-mcp 3.2.0 → 3.2.2

package/CHANGELOG.md

@@ -7,6 +7,18 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [3.2.2] - 2026-03-04
+
+### Fixed
+- **LLM API timeout + retry**: Agent API calls (`chatWithRetry`) now abort after 240 seconds via `AbortController` instead of hanging until the 300s hard gateway kill seen in build 76741. Timed-out requests are automatically retried using the existing retry/backoff loop (up to `maxRetries`, default 10). Configurable via `agents.defaults.retry.timeoutMs` in `.bctb-config.json`. The `timeoutMs` field was added to `RetryConfig` and `ChatOptions` (provider-level). Only the Anthropic provider is affected (the only current implementation).
+
+## [3.2.1] - 2026-03-04
+
+### Changed
+- **Percentile-based significant events**: `get_event_catalog` now identifies events covering the 90th percentile of total volume (instead of pointing to a single event). The response includes a `significantEvents` array with per-event percentage contribution and `uniqueEventIds` count. Events are deduplicated by eventId (summing counts across different shortMessage variants) before computing percentiles, eliminating duplicate noise (e.g., LC0156 appearing 10 times with different version strings).
+- **Improved `requiredNextStep` guidance**: Now lists all significant event IDs with their percentage contribution, telling agents to call `get_event_field_samples` for each — ensuring broad investigation that scales with the data distribution rather than cherry-picking one event.
+- **Updated server instructions**: Steps 1 and 2 now reference the `significantEvents` list and 90% volume threshold, explicitly guiding agents to investigate all significant events, not just the first one.
+
 ## [3.2.0] - 2026-03-04
 
 ### Added