Main layout redesign

.github/workflows/ci.yml

@@ -147,6 +147,10 @@ jobs:
         working-directory: e2e_tests/${{ matrix.name }}
         run: just install
 
+      - name: Install Playwright browsers
+        working-directory: e2e_tests/${{ matrix.name }}
+        run: just install-playwright
+
       - name: Run Playwright tests
         working-directory: e2e_tests/${{ matrix.name }}
         run: just test

README.md

@@ -26,6 +26,7 @@ modAI-chat/
 ├── backend/
 │   ├── omni/              # Python FastAPI backend
 │   └── tools/dice-roller/ # Example tool microservice
+├── docs/                  # User, admin & developer documentation
 ├── frontend/
 │   └── omni/              # Svelte TypeScript SPA
 └── e2e_tests/
@@ -43,9 +44,18 @@ Why more than one frontend: because there are different use cases. e.g. one full
 ### Prerequisites
 - Python 3.13+
 - Node.js 24+
-- UV package manager
-- Docker (for NanoIDP identity provider)
-- Optional: [just](https://github.com/casey/just)
+- [pnpm](http://pnpm.io)
+- [uv](https://docs.astral.sh/uv)
+- Docker
+- [just](https://github.com/casey/just)
+
+For easy dependency installation of the whole repo, run
+
+```bash
+just install
+```
+
+this installes the python/npm dependencies for all subprojects.
 
 ### NanoIDP Setup (Identity Provider)
 Start the lightweight local OIDC identity provider (runs on port 9000):
@@ -59,16 +69,16 @@ nanoidp Dashboard: http://localhost:9000
 ```bash
 cd backend/omni
 cp .env.sample .env
-uv sync
-uv run uvicorn modai.main:app --reload
+just install
+just start
 ```
 
 ### Frontend Setup
 ```bash
 cd frontend/omni
 ln -sf modules_with_backend.json public/modules.json
-pnpm install
-pnpm dev
+just install
+just start
 ```
 
 Browse to http://localhost:5173/

backend/omni/docs/learnings/INSTRUCTION_UPDATES.md

@@ -28,20 +28,7 @@ This file tracks corrections provided by the user to improve future performance.
 - **Correction**: Tests should only verify observable behavior (e.g. construction succeeds/fails, method returns expected result), never assert on internal instance attributes.
 - **New Rule**: Never assert on internal object fields/state in tests. Only verify behavior: does it raise? Does the return value match expectations? Does it produce the correct side-effects?
 
-### 2026-03-04 - No `patch` in tests — use testcontainers + pytest-httpserver
-- **Mistake**: Used `unittest.mock.patch` to spy on `_create_agent` and mock `httpx.Client` to test tool handler behaviour.
-- **Correction**: `patch` must not be used. The file under test should be modular enough that all external dependencies can be configured. For the LLM side use an llmock testcontainer; for tool HTTP endpoints use `pytest-httpserver`.
-- **New Rule**: Never use `patch`. For tool invocation tests, configure llmock with `ToolCallStrategy` and point `ToolDefinition.url` at a `pytest-httpserver` instance.
-
 ### 2026-03-04 - Explicit user directive: no whitebox testing anywhere
 - **Mistake**: Writing tests that target individual private/internal functions to achieve coverage.
 - **Correction**: All tests must exercise only the public interface. If internal logic needs coverage, improve public-API tests, not private-function tests.
 - **New Rule**: NO WHITEBOX TESTING. Never test `_prefixed` functions or assert on private object state. A test that does so is incorrect by definition and must be rewritten to go through the public API. Updated `AGENTS.md`.
-
-### 2026-03-13 - Injected metadata in tool params uses `_` prefix
-- **Convention**: When the caller needs to pass transport-level metadata (e.g. bearer token) into `Tool.run`, inject it as a `_`-prefixed key in the `params` dict (e.g. `_bearer_token`). The implementation pops those keys before building the request body; they are never forwarded as JSON payload.
-- **New Rule**: Any caller-injected, non-payload property passed via `params` MUST use a `_`-prefixed key. Document new keys in `docs/architecture/tools.md` under "Reserved `_`-prefixed keys".
-
-- **Mistake**: Passed `base_url = f"{root_url}/v1"` — the updated llmock no longer mounts routes under `/v1`.
-- **Correction**: All llmock endpoints are now at the root (`/chat/completions`, `/models`, `/health`). Pass `base_url = f"{root_url}/"` (trailing slash) so the OpenAI SDK does not append `/v1`.
-- **New Rule**: Always use `base_url = "http://<host>:<port>/"` (trailing slash) when connecting to llmock. The SKILL.md has been updated.

backend/omni/pyproject.toml

@@ -27,7 +27,7 @@ dev = [
     "pytest-asyncio",
     "pytest-httpserver",
     "ruff",
-    "testcontainers",
+    "testcontainers>=4.15.0rc2",
 ]
 
 [tool.pytest.ini_options]