feat(slurm): integrate Slurm execution for 1.0.0 release

.agents/recipes/dependencies/recipe.md

@@ -11,7 +11,7 @@ permissions:
 
 # Dependency Audit
 
-Audit the dependency graph across all three packages. Write findings to
+Audit the dependency graph across all four packages. Write findings to
 `/tmp/audit-{{suite}}.md`.
 
 Dependabot handles version bump PRs. This recipe focuses on what Dependabot
@@ -39,11 +39,12 @@ Read the `pyproject.toml` for each package:
 cat packages/data-designer-config/pyproject.toml
 cat packages/data-designer-engine/pyproject.toml
 cat packages/data-designer/pyproject.toml
+cat packages/data-designer-slurm/pyproject.toml
 ```
 
-Note: engine and interface packages use `uv-dynamic-versioning` to inject
-dependencies. Check both static declarations and the dynamic versioning
-config.
+Note: engine, interface, and Slurm packages use `uv-dynamic-versioning` to
+inject dependencies. Check both static declarations and the dynamic
+versioning config, including selected optional dependencies.
 
 ### 2. Direct dependency declaration gaps
 
@@ -79,7 +80,7 @@ intentionally deferred but still need to be declared as dependencies.
 
 Check that shared dependencies use consistent version constraints:
 ```bash
-# Extract dependency specs from all three pyproject.toml files
+# Extract dependency specs from all four pyproject.toml files
 grep -E "^\s+\"[a-zA-Z]" packages/*/pyproject.toml
 ```
 

.agents/recipes/structure/recipe.md

@@ -19,21 +19,24 @@ to `/tmp/audit-{{suite}}.md`.
 Before starting, read the authoritative sources for the structural rules:
 
 1. **`AGENTS.md`** (repo root) - "The Layering Is Structural" section defines
-   the three packages, their ownership, and the dependency direction rule.
+   the four packages, their ownership, and the dependency direction rule.
 2. **`architecture/overview.md`** - system architecture diagram, package layout,
    and the "no reverse imports" rule.
 
 The canonical rules:
 
 ```
-data-designer (interface) -> data-designer-engine -> data-designer-config
+data-designer-slurm -> data-designer (interface) -> data-designer-engine -> data-designer-config
 ```
 
 - `packages/data-designer-config/` must NOT import from `data_designer.engine`,
-  `data_designer.interface`, or `data_designer.cli`
+  `data_designer.interface`, `data_designer.cli`, or `data_designer.slurm`
 - `packages/data-designer-engine/` must NOT import from
-  `data_designer.interface` or `data_designer.cli`
-- `packages/data-designer/` CAN import from both engine and config
+  `data_designer.interface`, `data_designer.cli`, or `data_designer.slurm`
+- `packages/data-designer/` CAN import from engine and config but must NOT
+  import from `data_designer.slurm`
+- `packages/data-designer-slurm/` uses public interface/config APIs and must
+  NOT import engine internals
 
 **What CI already enforces**: ruff rule `TID` catches relative imports. The
 CI test matrix runs config, engine, and interface tests in isolation (separate

.agents/recipes/test-health/recipe.md

@@ -306,7 +306,7 @@ Write the report to `/tmp/audit-{{suite}}.md`:
 
 | Check | Status | Detail |
 |-------|--------|--------|
-| Package imports | OK/FAIL | All three packages import cleanly |
+| Package imports | OK/FAIL | All four packages import cleanly |
 | Import timing | OK/FAIL | X.XXs (budget: 3s) |
 | Registry completeness | OK/WARN | Column types resolve to config classes |
 

.github/workflows/agentic-ci-daily.yml

@@ -144,6 +144,7 @@ jobs:
         if: matrix.suite == 'dependencies'
         run: |
           .venv/bin/python scripts/audit_package_dependencies.py \
+            --extra data-designer:slurm \
             --output /tmp/dependency-inventory.json
           jq '.packages[] | {package, missing, unresolved_modules}' \
             /tmp/dependency-inventory.json

.github/workflows/ci.yml

@@ -129,6 +129,42 @@ jobs:
           uv run --with pytest --with pytest-asyncio --with pytest-httpx --with pytest-env \
             pytest packages/data-designer/tests
 
+  test-slurm-package:
+    name: ${{ matrix.name }}
+    needs: validate-dispatch
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - python-version: "3.10"
+            test-path: packages/data-designer-slurm/tests/runtime/test_proxy.py
+            name: Test Slurm proxy (Python 3.10)
+          - python-version: "3.11"
+            test-path: packages/data-designer-slurm/tests
+            name: Test Slurm package wheels
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@c771a70e6277c0a99b617c7a806ffedaca235ff9 # v9.0.0
+        with:
+          version: "latest"
+          python-version: ${{ matrix.python-version }}
+          enable-cache: true
+
+      - name: Install development dependencies
+        run: make install-dev
+
+      - name: Run package tests
+        run: .venv/bin/pytest ${{ matrix.test-path }}
+
+      - name: Run built-wheel installation tests
+        if: matrix.python-version == '3.11'
+        run: make test-slurm-wheel-install
+
   # ===========================================================================
   # Combined Coverage Check
   # Runs all tests together to verify overall coverage threshold
@@ -284,7 +320,7 @@ jobs:
   test-summary:
     name: Test (Python ${{ matrix.python-version }} on ${{ matrix.os }})
     runs-on: ubuntu-latest
-    needs: [validate-dispatch, test-config, test-engine, test-interface]
+    needs: [validate-dispatch, test-config, test-engine, test-interface, test-slurm-package]
     if: always()
     strategy:
       matrix:
@@ -297,12 +333,14 @@ jobs:
           if [[ "${{ needs.validate-dispatch.result }}" != "success" ]] || \
              [[ "${{ needs.test-config.result }}" != "success" ]] || \
              [[ "${{ needs.test-engine.result }}" != "success" ]] || \
-             [[ "${{ needs.test-interface.result }}" != "success" ]]; then
+             [[ "${{ needs.test-interface.result }}" != "success" ]] || \
+             [[ "${{ needs.test-slurm-package.result }}" != "success" ]]; then
             echo "One or more test jobs failed"
             echo "validate-dispatch: ${{ needs.validate-dispatch.result }}"
             echo "test-config: ${{ needs.test-config.result }}"
             echo "test-engine: ${{ needs.test-engine.result }}"
             echo "test-interface: ${{ needs.test-interface.result }}"
+            echo "test-slurm-package: ${{ needs.test-slurm-package.result }}"
             exit 1
           fi
           echo "All test jobs passed successfully"

AGENTS.md

@@ -7,15 +7,16 @@ If you are an agent helping a user **build a dataset**, use the [`data-designer`
 
 ## The Layering Is Structural
 
-The `data_designer` namespace is split across three installable packages that merge at runtime via PEP 420 implicit namespace packages (no top-level `__init__.py`).
+The `data_designer` namespace is split across four installable packages that merge at runtime via PEP 420 implicit namespace packages (no top-level `__init__.py`).
 
 | Package | Path | Owns |
 |---------|------|------|
 | `data-designer-config` | `packages/data-designer-config/` | `data_designer.config` — column configs, model configs, sampler params, builder API, plugin system, lazy imports |
 | `data-designer-engine` | `packages/data-designer-engine/` | `data_designer.engine` — column generators, dataset builders, DAG execution, model facade, validators, sampling |
 | `data-designer` | `packages/data-designer/` | `data_designer.interface` — public `DataDesigner` class, results, errors; `data_designer.cli` — CLI entry point; `data_designer.integrations` |
+| `data-designer-slurm` | `packages/data-designer-slurm/` | `data_designer.slurm` — optional Slurm execution |
 
-**Dependency direction (left depends on right):** interface → engine → config. Never import against this flow.
+**Import direction (left imports right):** Slurm → interface → engine → config. The `data-designer[slurm]` extra creates a packaging-only reverse edge; no code may import against this flow.
 
 ## Core Concepts
 
@@ -34,7 +35,7 @@ The `data_designer` namespace is split across three installable packages that me
 
 ## Structural Invariants
 
-- **Import direction** — interface → engine → config (left depends on right). No reverse imports.
+- **Import direction** — Slurm → interface → engine → config (left imports right). No reverse imports.
 - **Fast imports** — heavy third-party libraries are lazy-loaded via `data_designer.lazy_heavy_imports`. See [STYLEGUIDE.md](STYLEGUIDE.md) for the pattern.
 - **No relative imports** — absolute imports only, enforced by ruff rule `TID`.
 - **Typed code** — all functions, methods, and class attributes require type annotations. Modern syntax: `list[str]`, `str | None`.

DEVELOPMENT.md

@@ -105,16 +105,17 @@ uv run ruff format --check       # Check formatting
 
 ### Running Tests
 
-`make test` runs all three package test suites in sequence (config, engine, interface). When iterating on a single package, run its tests directly:
+`make test` runs all four package test suites in sequence (config, engine, interface, Slurm). When iterating on a single package, run its tests directly:
 
 ```bash
-# Run all tests (config + engine + interface)
+# Run all tests (config + engine + interface + Slurm)
 make test
 
 # Run a single package's tests
 make test-config       # data-designer-config
 make test-engine       # data-designer-engine
 make test-interface    # data-designer (interface)
+make test-slurm        # data-designer-slurm
 
 # Run a specific test file
 uv run pytest tests/config/test_sampler_constraints.py