@ai-matrx/agents 0.35.0 → 0.37.0

package/CHANGELOG.md

@@ -1,5 +1,20 @@
 # Changelog
 
+## 0.37.0
+
+- **The request pipeline moves in** (chat package independence P9b — one `callApi`, not two). New `./matrx` module `call`: `buildMatrxRequestUrl`, `buildMatrxRequestBody` (scope injection, UI-only strip, fail-closed body-vs-context organization check), `executeMatrxCall` (JSON or NDJSON over the v2 → v1 fallback; stream callbacks, `consumeStream`, 204 = no data), `parseMatrxNdjsonResponse` (broken body → resumable `StreamTransportError`, with `onEvent` / `onTransportError` hooks), `bareStatusSentence`, `isBareTransportCode`, `honestTransportMessage`, `buildSafeRequestLog`, `redactUrlForRequestLog`, `shouldReportMatrxCallError`, and the types `MatrxCallScope`, `MatrxCallResult`, `MatrxHttpMethod`, `MatrxStreamParse`, `ExecuteMatrxCallRequest`, `ExecuteMatrxCallHooks`.
+- `normalizeMatrxError` is now THE classifier matrx-frontend's `callApi` used: typed `BackendApiError` / `StreamTransportError` keep their `code` and `details` (as `serverDetail`), and no message is ever a bare status line ("HTTP 503" becomes the status sentence; the status rides `status`). `MatrxCallError` gains `raw?`.
+- Consumer action: build host request clients on `executeMatrxCall` + `normalizeMatrxError`; supply only base URL, credential/org headers, and diagnostics hooks.
+
+## 0.36.1
+
+- Build fix (0.36.0 never reached npm: the publish job's DTS worker ran out of heap). `./message-parts` is now built in the generated group, so the 6.8 MB `generated/api-types` declaration is rolled up by exactly one DTS pass. No API change from 0.36.0.
+
+## 0.36.0
+
+- **`./message-parts`** (chat package independence P15 batch 2): `UserInputPart` — the exact generated user-input union every agent route accepts — and its derived part types (`TextBlock`, `ImageBlock`, `AudioBlock`, `VideoBlock`, `DocumentBlock`, `YouTubeVideoBlock`, the structured input blocks, table/list bookmarks, data refs, `ContentBlock`, `UserInput`). Moved from matrx-frontend chat `agents/types/message-types.ts` + `request.types.ts` so chat and the files feature depend on agents, not on each other. Pure; derived from `./generated/api-types` only.
+- Consumer action: import these from `@ai-matrx/agents/message-parts`; delete host copies.
+
 ## 0.35.0
 
 - **The server's client contract lives here** (chat package independence P15, chair ruling 2026-10-02). New pure subpaths `./generated/{api-types,stream-events,llm-enums,llm-params-enums,source-attribution,workflow-events,canvas-chat,provision-offers,hr-contracts}` — emitted by aidream `scripts/generate_types.py` / `scripts/mandates_generate.py` straight into `generated/`, so a server change and its types land in one commit and ship on the publish train. `api-types` and `hr-contracts` are type-only (empty runtime modules).

package/dist/content-transfer/index.cjs

@@ -709,18 +709,6 @@ function extractMatrxErrorCode(serverDetail) {
   return null;
 }
 
-// matrx/conversation.ts
-function mintMatrxConversationId() {
-  return crypto.randomUUID();
-}
-function newStoredConversationStart(conversationId) {
-  return {
-    conversation_id: conversationId ?? mintMatrxConversationId(),
-    is_new: true,
-    store: true
-  };
-}
-
 // stream/ndjson.ts
 var DEFAULT_MATRX_NDJSON_READ_AHEAD = 64;
 function isRecord2(value) {
@@ -941,6 +929,18 @@ async function* readMatrxNdjsonStream(body, options = {}) {
   }
 }
 
+// matrx/conversation.ts
+function mintMatrxConversationId() {
+  return crypto.randomUUID();
+}
+function newStoredConversationStart(conversationId) {
+  return {
+    conversation_id: conversationId ?? mintMatrxConversationId(),
+    is_new: true,
+    store: true
+  };
+}
+
 // matrx/internal.ts
 function encodePathSegment(value) {
   return encodeURIComponent(value);