fix(infra,cowork): remove broken Cosmos emulator; harden IPC bridge

docker-compose:
  - Drop the cosmos-emulator service block. Both image variants we
    tried were unfit for the prototype: `:vnext-preview` returned
    plain-text PGCosmosError strings that crashed @azure/cosmos at
    JSON.parse, and `:latest` core-dumped under load. The container
    has been Exited(255) for weeks and was blocking depends_on chains.
  - Real Azure Cosmos account `cosmos-mywisprai` (db `bytelyst`,
    West US 2) is now the single source of truth; all services pick
    up COSMOS_ENDPOINT/KEY/DATABASE from `.env` (already mounted via
    `env_file: .env`).
  - extraction-service: drop hardcoded `COSMOS_ENDPOINT=…cosmos-emulator…`,
    `NODE_TLS_REJECT_UNAUTHORIZED=0`, and `depends_on: cosmos-emulator`.
  - cowork-service: same cleanup.

cowork-service IPC bridge:
  - Add `error` listeners to the spawned child's stdin/stdout/stderr.
    Without them, an EPIPE on stdin (child died mid-write) or a
    teardown-time stream error surfaced as an unhandled error and
    crashed vitest after all 140 tests had passed.
  - Removes the only failing recursive test in the workspace.

Test status after this commit:
  - 94 workspace packages, all green
  - cowork-service: 19 passed | 1 skipped (140 tests)
  - platform-service: 131 test files passed
  - extraction-service: 13 test files passed
  - All other packages: passing

Generated with [Devin](https://cli.devin.ai/docs)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

docker-compose.yml

@@ -33,38 +33,28 @@ services:
       retries: 6
     restart: unless-stopped
 
-  # ── Azure Cosmos DB Emulator (prototype only) ─────────────────
+  # ── Azure Cosmos DB Emulator — REMOVED 2026-05-30 ─────────────
-  # 2026-05-30: switched off `:vnext-preview` (Postgres-backed experimental
+  #
-  # image) because cross-partition `queryFeed` returned plain-text
+  # Both image variants we tried were unfit for the prototype:
-  # `PGCosmosError(3, "Database query failed: PostgresError(SqlState(EXX000))...")`
+  #   - `:vnext-preview` (Postgres-backed experimental): cross-partition
-  # for every query, which crashes the @azure/cosmos SDK at JSON.parse. Affected
+  #     `queryFeed` returned plain-text PGCosmosError strings instead of
-  # every collection — point reads worked, but anything filtering by non-PK
+  #     JSON, crashing @azure/cosmos at JSON.parse on every login,
-  # fields failed (login, register, OAuth, feature-flag list, etc.).
+  #     register, OAuth, and feature-flag list call
-  # `:latest` is the stable Linux port of the Windows Cosmos emulator and
+  #   - `:latest` (stable Linux port of Windows emulator): HTTPS-only
-  # returns proper JSON errors. It is HTTPS-only with a self-signed cert,
+  #     with a self-signed cert and core-dumped under modest load,
-  # so consumers in this compose file are gated by NODE_TLS_REJECT_UNAUTHORIZED=0
+  #     leaving services hung waiting on never-resolving futures
-  # (dev-prototype only — never set that in production).
+  #
-  cosmos-emulator:
+  # Replacement: real Azure Cosmos DB account `cosmos-mywisprai` in
-    image: mcr.microsoft.com/cosmosdb/linux/azure-cosmos-emulator:latest
+  # `rg-mywisprai` (West US 2), database `bytelyst`. All services pick
-    ports:
+  # up the connection from `.env` (`COSMOS_ENDPOINT`, `COSMOS_KEY`,
-      - '8081:8081'
+  # `COSMOS_DATABASE`) via their `env_file: .env` entries below.
-      - '1234:1234'
+  #
-    environment:
+  # If you need a local-only stack for offline development, prefer:
-      - AZURE_COSMOS_EMULATOR_PARTITION_COUNT=10
+  #   1. Mocked Cosmos in tests (already wired across the workspace), or
-      - AZURE_COSMOS_EMULATOR_ENABLE_DATA_PERSISTENCE=false
+  #   2. A scoped Cosmos account on a free Azure subscription with a
-      - ENABLE_EXPLORER=true
+  #      throwaway database
-      - GATEWAY_PUBLIC_ENDPOINT=cosmos-emulator
+  # Do NOT resurrect the emulator service block without verifying both
-    healthcheck:
+  # of the above failure modes have been fixed upstream.
-      test:
-        [
-          'CMD-SHELL',
-          'curl -sk --max-time 3 https://127.0.0.1:8081/_explorer/emulator.pem > /dev/null',
-        ]
-      interval: 10s
-      timeout: 5s
-      retries: 30
-      start_period: 60s
-    restart: unless-stopped
 
   # ── Loki (Log Aggregation) ────────────────────────────────────
   loki:
@@ -179,13 +169,7 @@ services:
     environment:
       - PORT=4005
       - PYTHON_SIDECAR_URL=http://localhost:4006
-      # See cosmos-emulator service block: stable image is HTTPS-only with
+      # COSMOS_* come from `.env` (real Cosmos account; see top of file).
-      # a self-signed cert. Dev-prototype only.
-      - COSMOS_ENDPOINT=https://cosmos-emulator:8081
-      - NODE_TLS_REJECT_UNAUTHORIZED=0
-    depends_on:
-      cosmos-emulator:
-        condition: service_healthy
     labels:
       - 'traefik.enable=true'
       - 'traefik.http.routers.extraction.rule=PathPrefix(`/api/extract`) || PathPrefix(`/api/tasks`)'
@@ -249,14 +233,10 @@ services:
       - PORT=4009
       - NODE_ENV=development
       - PRODUCT_ID=clawcowork
-      - COSMOS_ENDPOINT=https://cosmos-emulator:8081
+      # COSMOS_* come from `.env` (real Cosmos account; see top of file).
-      # Stable emulator's self-signed cert — dev-prototype only.
-      - NODE_TLS_REJECT_UNAUTHORIZED=0
       - PLATFORM_SERVICE_URL=http://platform-service:4003
       - EXTRACTION_SERVICE_URL=http://extraction-service:4005
     depends_on:
-      cosmos-emulator:
-        condition: service_healthy
       platform-service:
         condition: service_healthy
     labels:

services/cowork-service/src/lib/ipc-bridge.ts

@@ -77,6 +77,19 @@ export class IpcBridge {
       this.log.error(`IPC child process error: ${err.message}`);
     });
 
+    // Without these listeners, an EPIPE on stdin (child died mid-write) or
+    // a stdout/stderr error during teardown becomes an unhandled stream
+    // error and crashes the test runner / parent process.
+    this.child.stdin?.on('error', err => {
+      this.log.error(`IPC child stdin error: ${err.message}`);
+    });
+    this.child.stdout?.on('error', err => {
+      this.log.error(`IPC child stdout error: ${err.message}`);
+    });
+    this.child.stderr?.on('error', err => {
+      this.log.error(`IPC child stderr error: ${err.message}`);
+    });
+
     this.child.on('exit', (code, signal) => {
       this.log.info(`IPC child process exited: code=${code} signal=${signal}`);
       this.rejectAllPending(new Error(`IPC child process exited (code=${code})`));