Prompt 9 AI platform

.env.example

@@ -21,6 +21,15 @@ NEO4J_PASSWORD=changeme_neo4j
 REDIS_HOST=redis
 REDIS_PORT=6379
 REDIS_URL=redis://redis:6379/0
+REDIS_OPERATION_TIMEOUT_SECONDS=10
+AGENT_JOB_STREAM=agents:jobs
+AGENT_JOB_GROUP=agents-workers
+AGENT_WORKER_RECLAIM_IDLE_SECONDS=60
+AGENT_WORKER_EXECUTION_TIMEOUT_SECONDS=300
+AGENT_WORKER_LOCK_TTL_SECONDS=390
+AGENT_MAX_RETRIES=3
+# Required in production; use a generated Fernet-compatible key.
+EXECUTION_PAYLOAD_KEY=
 
 # ── OpenSearch ───────────────────────────────────────────────────────────
 OPENSEARCH_URL=http://opensearch:9200

.github/workflows/ci.yml

@@ -20,6 +20,9 @@ jobs:
         with: { version: "${{ env.PNPM_VERSION }}" }
       - uses: actions/setup-node@v4
         with: { node-version: 20, cache: pnpm }
+      - uses: actions/setup-python@v5
+        with: { python-version: "${{ env.PYTHON_VERSION }}" }
+      - run: python -m pip install ruff mypy
       - run: pnpm install --frozen-lockfile
       - run: pnpm lint
       - run: pnpm typecheck
@@ -57,9 +60,10 @@ jobs:
       - working-directory: ${{ matrix.service }}
         run: |
           pip install -r requirements.txt
-          pip install ruff pytest
-          ruff check app || true
-          pytest -q || true
+          pip install ruff mypy
+          ruff check app tests
+          mypy app --ignore-missing-imports
+          pytest -q
 
   docker-build:
     name: Docker — build all images
@@ -78,12 +82,20 @@ jobs:
             docker build -f "$dockerfile" -t "ai-rxos/$tag:ci" .
             echo "::endgroup::"
           done
+      - name: Verify agents image
+        run: |
+          image=ai-rxos/services-agents:ci
+          docker run --rm "$image" python -m pip check
+          docker run --rm "$image" python -c "import fastapi, langgraph, cryptography, redis, jwt, opentelemetry, opentelemetry.sdk, opentelemetry.exporter.otlp.proto.http"
+          docker run --rm "$image" python -m compileall -q app tests
+          docker run --rm "$image" python -m pytest -q
 
   helm-lint:
     name: Helm — lint chart
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
       - uses: azure/setup-helm@v4
+      - run: helm dependency build infra/helm/ai-rxos
       - run: helm lint infra/helm/ai-rxos
       - run: helm template ai-rxos infra/helm/ai-rxos -f infra/helm/ai-rxos/values-dev.yaml > /dev/null

Makefile

@@ -1,4 +1,4 @@
-.PHONY: bootstrap dev build lint test up down logs helm-lint helm-template
+.PHONY: bootstrap dev build lint test test-python up down logs helm-lint helm-template
 
 bootstrap:
 	pnpm install
@@ -16,6 +16,16 @@ lint:
 test:
 	pnpm test
 
+test-python:
+	cd apps/ai-services && python -m pytest -q
+	cd apps/knowledge-service && python -m pytest -q
+	cd services/agents && python -m pytest -q
+	cd services/docking && python -m pytest -q
+	cd services/kg && python -m pytest -q
+	cd services/literature && python -m pytest -q
+	cd services/reports && python -m pytest -q
+	cd services/workflows && python -m pytest -q
+
 up:
 	docker compose up --build -d
 

README.md

@@ -65,6 +65,42 @@ out to the native toolchain (`go build`, `uvicorn`, `pytest`, ...), so Turbo
 can orchestrate the whole polyglot repo (`pnpm build`, `pnpm dev --parallel`,
 etc.) rather than just the JS packages.
 
+## Testing
+
+Run the full workspace test command through Turbo:
+
+```bash
+pnpm test
+```
+
+Do not run `python -m pytest` from the repository root. The Python services
+are independent projects with separate `app` packages, dependency sets, and
+test configuration. A single root pytest process mixes those import roots and
+causes errors such as `ModuleNotFoundError: No module named 'app'`, missing
+service-specific dependencies, and duplicate `test_health` module names.
+
+To run Python tests together from a shell that supports Make:
+
+```bash
+make test-python
+```
+
+Or run a service directly from its own directory, for example:
+
+```bash
+cd services/agents
+python -m pytest -q
+```
+
+The `bezs-pipeline` tests are a separate Python project. Install its project
+and development dependencies from `bezs-pipeline` before running them:
+
+```bash
+cd bezs-pipeline
+python -m pip install -e ".[dev]"
+python -m pytest -q
+```
+
 ## Quickstart — Docker Compose (fastest path to a running system)
 
 ```bash

architecture/06-ai-agent-orchestration.md

@@ -1,5 +1,27 @@
 # AI Agent Orchestration Architecture - AI-RxOS
 
+> **Implementation note (Prompt 9):** The current implementation is in
+> `services/agents`. It uses a custom `StateGraph`/`AgentGraph` runtime and an
+> MCP client; it does not install the external LangGraph package, OpenAI
+> Agents SDK, or expose the illustrative MCP server shown later in this
+> document. The code-backed behavior takes precedence over the conceptual
+> examples below.
+
+The current Prompt 9 service also uses Redis Streams for durable asynchronous
+jobs, consumer-group recovery, tenant-scoped idempotency keys, bounded worker
+retries, and opaque request/correlation IDs. `AuthorizationService` applies
+deny-by-default agent and tool policies from authenticated JWT context. The
+repository does not currently expose a separate Better Auth permission API or
+OpenTelemetry exporter for this service.
+
+> **Implementation note (Prompt 9):** The current implementation is in
+> `services/agents`. It uses a custom `StateGraph`/`AgentGraph` runtime and an
+> MCP client; it does not install the external LangGraph package, OpenAI
+> Agents SDK, or expose the illustrative MCP server shown later in this
+> document. The code-backed behavior takes precedence over the conceptual
+> examples below.
+# AI Agent Orchestration Architecture - AI-RxOS
+
 ## Overview
 AI-RxOS uses a multi-agent orchestration system based on the Model Context Protocol (MCP) and OpenAI Agents SDK. Agents are autonomous AI entities that use tools to accomplish tasks, coordinate with other agents, and maintain conversation context.