fix: add explicit permissions to CI and integration workflows

.github/workflows/ci.yml

@@ -5,6 +5,9 @@ on:
     branches: [main]
   pull_request:
 
+permissions:
+  contents: read
+
 concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref || github.ref_name }}
   cancel-in-progress: true

.github/workflows/integration-nightly.yml

@@ -9,6 +9,9 @@ on:
         description: "Specific model ID to test (blank = all non-gated models that fit in memory)"
         required: false
 
+permissions:
+  contents: read
+
 concurrency:
   group: nightly-integration
   cancel-in-progress: true

.github/workflows/integration-prerelease.yml

@@ -5,6 +5,9 @@ on:
     types: [created]
   workflow_dispatch:
 
+permissions:
+  contents: read
+
 concurrency:
   group: prerelease-integration
   cancel-in-progress: true

README.md

@@ -14,8 +14,7 @@ Most local LLM tools serve **one model at a time** and leave you to figure out w
 
 ```bash
 uv tool install mlx-stack
-mlx-stack init --accept-defaults   # detects hardware, picks models, generates configs
-mlx-stack up                       # 3 model servers + API gateway, one command
+mlx-stack setup                    # detects hardware, picks models, pulls, starts — one command
 # → OpenAI-compatible API at http://localhost:4000/v1
 ```
 
@@ -141,6 +140,33 @@ uvx mlx-stack profile
 
 ## Quick Start
 
+The fastest way to get running is the interactive setup command:
+
+```bash
+mlx-stack setup
+```
+
+This walks you through hardware detection, model selection, downloading, and starting all services in one guided flow. For CI or scripting, pass `--accept-defaults` to skip all prompts:
+
+```bash
+mlx-stack setup --accept-defaults
+```
+
+The OpenAI-compatible API is now available at `http://localhost:4000/v1`.
+
+```bash
+# Check service health
+mlx-stack status
+
+# Stop everything when done
+mlx-stack down
+```
+
+<details>
+<summary>Manual step-by-step setup</summary>
+
+If you prefer full control over each step:
+
 ```bash
 # 1. Detect your hardware
 mlx-stack profile
@@ -158,17 +184,20 @@ mlx-stack up
 mlx-stack status
 ```
 
-The OpenAI-compatible API is now available at `http://localhost:4000/v1`.
-
-```bash
-# Stop everything when done
-mlx-stack down
-```
+</details>
 
 ## CLI Reference
 
 ### Setup & Configuration
 
+**`mlx-stack setup`** — Interactive guided setup: detects hardware, selects models, pulls weights, and starts the stack in one command.
+
+| Option | Description |
+|--------|-------------|
+| `--accept-defaults` | Skip all prompts and use recommended defaults |
+| `--intent <balanced\|agent-fleet>` | Use case intent (prompted if not provided) |
+| `--budget-pct <10-90>` | Memory budget as percentage of unified memory (default: 40) |
+
 | Command | Description |
 |---------|-------------|
 | `mlx-stack profile` | Detect Apple Silicon hardware and save profile to `~/.mlx-stack/profile.json` |
@@ -294,7 +323,7 @@ mlx-stack is designed to run unattended on always-on hardware like a Mac Mini.
 ### Quick setup
 
 ```bash
-mlx-stack init --accept-defaults
+mlx-stack setup --accept-defaults
 mlx-stack install
 ```
 
@@ -407,14 +436,12 @@ See [DEVELOPING.md](DEVELOPING.md) for the full developer guide, including proje
 # Install dev dependencies
 uv sync
 
-# Run tests
-uv run pytest
-
-# Type checking
-uv run python -m pyright
+# Run all checks (lint + typecheck + tests) — same as CI
+make check
 
-# Linting
-uv run ruff check src/ tests/
+# Or individually
+make lint    # ruff + pyright
+make test    # pytest with coverage
 ```
 
 ## Contributing

tests/unit/test_ops_cross_area.py

@@ -410,6 +410,15 @@ def follow_thread() -> None:
                     output_callback=lambda text: captured.append(text),
                 )
 
+        def wait_for_content(marker: str, timeout: float = 5.0) -> bool:
+            """Wait until marker appears in captured output."""
+            end = time.monotonic() + timeout
+            while time.monotonic() < end:
+                if any(marker in c for c in captured):
+                    return True
+                time.sleep(0.05)
+            return False
+
         thread = threading.Thread(target=follow_thread, daemon=True)
         thread.start()
 
@@ -423,7 +432,9 @@ def follow_thread() -> None:
         with open(log, "a") as f:
             f.write("after-truncation\n")
 
-        time.sleep(1.0)
+        assert wait_for_content("after-truncation"), (
+            f"Timed out waiting for 'after-truncation' in captured output: {captured}"
+        )
 
         import ctypes
 
@@ -435,9 +446,6 @@ def follow_thread() -> None:
             )
         thread.join(timeout=3)
 
-        combined = "\n".join(captured)
-        assert "after-truncation" in combined
-
     def test_follow_continues_after_multiple_rotations(
         self, mlx_stack_home: Path, logs_dir: Path
     ) -> None: