mlx-stack 0.3.7__tar.gz → 0.3.8__tar.gz

mlx_stack-0.3.8/.factory/init.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+set -e
+
+cd /Users/weae1504/Projects/mlx-stack
+
+# Install dev dependencies (idempotent)
+uv sync --dev

mlx_stack-0.3.8/.factory/library/architecture.md

@@ -0,0 +1,80 @@
+# Architecture
+
+How the mlx-stack system works at a high level.
+
+## Overview
+
+mlx-stack is a CLI tool that manages local LLM infrastructure on Apple Silicon. It orchestrates vllm-mlx model servers behind a LiteLLM proxy, providing a unified OpenAI-compatible API endpoint.
+
+## Layers
+
+```
+CLI Layer (src/mlx_stack/cli/)
+  ├── Commands: setup, up, down, status, models, pull, bench, logs, config, watch, install, uninstall
+  └── Each command is a Click command registered in main.py
+
+Core Layer (src/mlx_stack/core/)
+  ├── hardware.py      — Apple Silicon detection (chip, GPU cores, memory, bandwidth)
+  ├── catalog.py       — YAML catalog loading, validation, querying (15 curated models)
+  ├── discovery.py     — Live HuggingFace API query for mlx-community models
+  ├── scoring.py       — Hardware-aware model recommendation engine
+  ├── onboarding.py    — Setup wizard orchestration (scoring variant for DiscoveredModel)
+  ├── stack_init.py    — Stack definition generation (stack.yaml + litellm.yaml)
+  ├── litellm_gen.py   — LiteLLM proxy config generation
+  ├── stack_up.py      — Process management (start/stop vllm-mlx + LiteLLM)
+  ├── pull.py          — Model download (HuggingFace snapshot_download)
+  ├── benchmark.py     — Performance benchmarking
+  ├── watchdog.py      — Health monitoring + auto-restart
+  ├── launchd.py       — macOS LaunchAgent management
+  ├── config.py        — User config (~/.mlx-stack/config.yaml)
+  ├── paths.py         — Path resolution for data/config/stacks
+  └── process.py       — Low-level process management
+
+Data Layer (src/mlx_stack/data/)
+  ├── catalog/*.yaml   — Curated model entries (15 files)
+  └── benchmark_data.json — Static performance overlay from mlx_transformers_benchmark
+```
+
+## Data Flow
+
+1. **Hardware detection** → `HardwareProfile` (chip, memory, bandwidth, GPU cores)
+2. **Model discovery** → `CatalogEntry` (from YAML catalog) or `DiscoveredModel` (from HF API)
+3. **Scoring** → `ScoredModel` / `ScoredDiscoveredModel` with composite scores
+4. **Tier assignment** → `TierAssignment` (model → tier name mapping)
+5. **Config generation** → `stack.yaml` (tier definitions) + `litellm.yaml` (proxy config)
+6. **Process management** → vllm-mlx subprocesses + LiteLLM proxy process
+
+## Stack Tier Field Semantics
+
+- `stack.yaml` tier objects use:
+  - `name`: tier identifier (e.g., `standard`, `fast`, `reasoning`)
+  - `model`: canonical model identifier used by mlx-stack logic
+  - `source`: concrete model source for runtime/download
+- For catalog-backed tiers, keep `model` as the catalog model ID (for example `qwen3.5-8b`) rather than a display label, and keep the resolved Hugging Face repo in `source`.
+
+## Key Files for This Mission
+
+- `cli/main.py` — Command registration, `_COMMAND_CATEGORIES`, welcome screen, help formatting
+- `cli/pull.py` — Pull command (being ungated to accept HF repos)
+- `cli/status.py` — Status command (absorbing hardware display from profile)
+- `cli/models.py` — Models command (absorbing recommend functionality)
+- `cli/setup.py` — Setup command (gaining modification flags)
+- `cli/profile.py` — Being DELETED
+- `cli/recommend.py` — Being DELETED
+- `cli/init.py` — Being DELETED
+- `core/pull.py` — Download infrastructure (already accepts arbitrary HF repos)
+- `core/stack_init.py` — Config generation (preserved for internal use by setup)
+- `core/onboarding.py` — Setup wizard orchestration
+
+## Testing Patterns
+
+- All CLI tests use Click's `CliRunner().invoke(cli, ["command", ...])`
+- Core functions mocked via `@patch("mlx_stack.core.module.function")` or `monkeypatch.setattr`
+- `FakeServiceLayer` test double for stack_up/watchdog tests
+- Test factories in `tests/factories.py` for creating test data
+- No real HF downloads, no real hardware detection in unit tests
+
+## Operational Constraint: Service Name Safety
+
+- Service names are reused as PID/log filename stems by `core/process.py` (`pid_file` and log path construction).
+- Any dynamically generated `service_name` must be filesystem-safe (no path separators like `/`), or temp process startup can fail before health checks run.

mlx_stack-0.3.8/.factory/library/environment.md

@@ -0,0 +1,38 @@
+# Environment
+
+Environment variables, external dependencies, and setup notes.
+
+**What belongs here:** Required env vars, external API keys/services, dependency quirks, platform-specific notes.
+**What does NOT belong here:** Service ports/commands (use `.factory/services.yaml`).
+
+---
+
+## Python Environment
+
+- Python 3.14+ via `uv`
+- All dependencies managed by `uv sync --dev`
+- Virtual environment at `.venv/` (created by uv)
+
+## Key Dependencies
+
+- `click` — CLI framework
+- `rich` — Terminal UI (tables, colors, progress)
+- `pyyaml` — YAML parsing
+- `huggingface_hub` — HF API + model downloads
+- `pytest` + `pytest-cov` — Testing
+- `ruff` — Linting
+- `pyright` — Type checking
+
+## Environment Variables
+
+- `MLX_STACK_HOME` — Override data directory (default: `~/.mlx-stack/`). Used extensively in tests via `mlx_stack_home` fixture.
+
+## Data Directories
+
+- `~/.mlx-stack/` — User data home
+- `~/.mlx-stack/stacks/default.yaml` — Stack definition
+- `~/.mlx-stack/litellm.yaml` — LiteLLM proxy config
+- `~/.mlx-stack/profile.json` — Hardware profile
+- `~/.mlx-stack/config.yaml` — User configuration
+- `~/.mlx-stack/models/` — Downloaded model files
+- `~/.mlx-stack/benchmarks/` — Saved benchmark results

mlx_stack-0.3.8/.factory/library/user-testing.md

@@ -0,0 +1,57 @@
+# User Testing
+
+Testing surface, required tools, and validation approach.
+
+## Validation Surface
+
+**Primary surface:** CLI commands via pytest CliRunner (unit-level) and shell invocation (smoke-level).
+
+This is a CLI-only mission with no browser UI, no running services, and no external API dependencies during testing. All HuggingFace API calls and model downloads are mocked in tests.
+
+### Tools
+
+- **pytest** with Click's `CliRunner` — primary test executor
+- **Shell invocation** — for smoke tests that verify real subprocess CLI behavior
+- **pyright** — type checking gate
+- **ruff** — linting gate
+
+### Test Commands
+
+```bash
+uv run pytest --cov=src/mlx_stack -x -q --tb=short   # unit tests
+uv run python -m pyright                                # type check
+uv run ruff check src/ tests/                           # lint
+```
+
+## Validation Concurrency
+
+**Max concurrent validators: 5**
+
+Rationale: CLI tests are lightweight (no browser, no services). Each pytest invocation uses ~100MB RAM. Machine has 128GB RAM and 18 CPU cores. Even 5 concurrent test runs would use <1GB total. No infrastructure contention.
+
+## Testing Patterns
+
+- CLI commands tested via `CliRunner().invoke(cli, ["command", "--flag", "arg"])`
+- Exit codes checked: 0 for success, non-zero for errors
+- Output checked via `result.output` string matching
+- Side effects verified via mock assertions (`mock_download.assert_called_once()`, etc.)
+- File system effects checked via `tmp_path` fixtures
+- Test factories in `tests/factories.py` for creating test data consistently
+
+## Flow Validator Guidance: CLI
+
+- Surface is CLI-only; do not use browser automation.
+- Stay within repository-local and mission-local paths only:
+  - Repo: `/Users/weae1504/Projects/mlx-stack`
+  - Mission evidence: `/Users/weae1504/.factory/missions/7fc62a3d-138f-4cd2-a601-3f6d1b174b53/evidence/ungate-pull/<group-id>/`
+- Prefer assertion-targeted checks first (specific pytest tests and direct CLI invocations), then add broader checks only when needed to disambiguate failures.
+- Do not edit source code while validating; only create report/evidence artifacts requested for user-testing flows.
+- If any assertion is blocked by environment/tooling, capture exact blocking command output and mark as blocked rather than guessing.
+
+## Validation Notes: absorb-recommend-remove-init
+
+- `mlx-stack models --available` currently degrades to catalog-backed fallback when HF API is unreachable and still exits `0`. To reproduce the outage path in validation, set:
+  - `HTTP_PROXY=http://127.0.0.1:9 HTTPS_PROXY=http://127.0.0.1:9 ALL_PROXY=http://127.0.0.1:9 NO_PROXY=`
+  - or `HF_ENDPOINT=http://127.0.0.1:9`
+- `mlx-stack pull <model> --bench` now prints generic saved-results scoring guidance
+  without referencing removed commands (`recommend`/`init`).

mlx_stack-0.3.8/.factory/services.yaml

@@ -0,0 +1,8 @@
+commands:
+  install: uv sync --dev
+  test: uv run pytest --cov=src/mlx_stack -x -q --tb=short
+  typecheck: uv run python -m pyright
+  lint: uv run ruff check src/ tests/
+  check: uv run ruff check src/ tests/ && uv run python -m pyright && uv run pytest --cov=src/mlx_stack -x -q --tb=short
+
+services: {}

mlx_stack-0.3.8/.factory/skills/cli-worker/SKILL.md

@@ -0,0 +1,145 @@
+---
+name: cli-worker
+description: Implements CLI command changes, module refactoring, and test updates for mlx-stack
+---
+
+# CLI Worker
+
+NOTE: Startup and cleanup are handled by `worker-base`. This skill defines the WORK PROCEDURE.
+
+## When to Use This Skill
+
+Use for features that involve:
+- Adding, removing, or modifying Click CLI commands
+- Updating command registration in `main.py`
+- Modifying core modules called by CLI commands
+- Writing or rewriting pytest unit tests for CLI commands
+- Updating help text, command categories, error messages
+
+## Required Skills
+
+None — all work uses standard file editing and shell commands (pytest, pyright, ruff).
+
+## Work Procedure
+
+### Step 1: Understand the Feature
+
+Read the feature description, preconditions, expectedBehavior, and verificationSteps carefully. Read AGENTS.md for conventions and boundaries. Read `.factory/library/architecture.md` for system structure.
+
+### Step 2: Read Affected Files
+
+Before writing any code, read ALL files that will be affected:
+- The CLI command file(s) being changed
+- The core module(s) being called
+- The test file(s) being updated
+- `cli/main.py` if command registration changes
+- Any test files that import from affected modules
+
+Understand the existing patterns, mock strategies, and test structure.
+
+### Step 3: Write Tests First (TDD)
+
+Write failing tests BEFORE implementing changes:
+1. Create or update the test file with new test cases
+2. Run `uv run pytest tests/unit/<test_file> -x -q --tb=short` to confirm tests fail (red)
+3. Each test should test ONE specific behavior from the feature's expectedBehavior
+
+Test patterns to follow:
+```python
+from click.testing import CliRunner
+from mlx_stack.cli.main import cli
+
+def test_example(mlx_stack_home):
+    runner = CliRunner()
+    with patch("mlx_stack.core.module.function") as mock_fn:
+        result = runner.invoke(cli, ["command", "--flag", "arg"])
+    assert result.exit_code == 0
+    assert "expected output" in result.output
+    mock_fn.assert_called_once_with(...)
+```
+
+### Step 4: Implement Changes
+
+Make the minimum changes needed to make all tests pass:
+1. Modify CLI command files
+2. Modify core modules if needed
+3. Update `cli/main.py` command registration if needed
+
+Follow existing patterns:
+- Use `console = Console(stderr=True)` for errors, `out = Console()` for output
+- Catch domain exceptions, print user-friendly errors, `raise SystemExit(1)`
+- Use absolute imports: `from mlx_stack.core.module import Class`
+
+### Step 5: Run Tests (Green)
+
+1. Run the specific test file: `uv run pytest tests/unit/<test_file> -x -q --tb=short`
+2. Run the FULL test suite: `uv run pytest --cov=src/mlx_stack -x -q --tb=short`
+3. Fix any failures in other test files caused by your changes
+
+### Step 6: Run Validators
+
+1. Type check: `uv run python -m pyright`
+2. Lint: `uv run ruff check src/ tests/`
+3. Fix any issues
+
+### Step 7: Verify Manually
+
+For each changed command, run a quick manual check:
+```bash
+uv run mlx-stack --help                    # verify help output
+uv run mlx-stack <command> --help          # verify command help
+```
+
+If the feature removes a command, verify it's gone:
+```bash
+uv run mlx-stack <removed-command>          # should show error
+```
+
+### Step 8: Clean Up
+
+- Remove any deleted test files from disk
+- Remove any deleted CLI command files from disk
+- Ensure no orphaned imports remain
+- Run the full test suite one final time
+
+## Example Handoff
+
+```json
+{
+  "salientSummary": "Ungated pull command to accept HF repo strings. Added slash-based routing (contains '/' = HF repo, no '/' = catalog ID). Wrote 12 new tests in test_cli_pull.py covering HF repo acceptance, error handling, and flag combinations. All 1400+ tests pass, pyright clean, ruff clean.",
+  "whatWasImplemented": "Modified cli/pull.py to detect HF repo strings (containing '/') and bypass catalog lookup, routing directly to download_model(). Updated core/pull.py pull_model() to accept hf_repo_override parameter. Updated help text to document both input types. Added 12 new test cases and updated 3 existing tests.",
+  "whatWasLeftUndone": "",
+  "verification": {
+    "commandsRun": [
+      { "command": "uv run pytest tests/unit/test_cli_pull.py -x -q --tb=short", "exitCode": 0, "observation": "77 passed (12 new + 65 existing)" },
+      { "command": "uv run pytest --cov=src/mlx_stack -x -q --tb=short", "exitCode": 0, "observation": "1412 passed, 0 failed" },
+      { "command": "uv run python -m pyright", "exitCode": 0, "observation": "0 errors, 0 warnings" },
+      { "command": "uv run ruff check src/ tests/", "exitCode": 0, "observation": "All checks passed" },
+      { "command": "uv run mlx-stack pull --help", "exitCode": 0, "observation": "Help text mentions HF repo and catalog ID" }
+    ],
+    "interactiveChecks": [
+      { "action": "Ran 'uv run mlx-stack pull --help'", "observed": "Help text now says 'MODEL is a catalog model ID (e.g., qwen3.5-8b) or HuggingFace repo (e.g., mlx-community/Phi-5-Mini-4bit)'" }
+    ]
+  },
+  "tests": {
+    "added": [
+      {
+        "file": "tests/unit/test_cli_pull.py",
+        "cases": [
+          { "name": "test_pull_hf_repo_downloads_directly", "verifies": "HF repo string bypasses catalog lookup" },
+          { "name": "test_pull_hf_repo_with_quant_stores_metadata", "verifies": "--quant flag stores metadata for HF repo" },
+          { "name": "test_pull_hf_repo_nonexistent_shows_error", "verifies": "Invalid HF repo shows user-friendly error" }
+        ]
+      }
+    ]
+  },
+  "discoveredIssues": []
+}
+```
+
+## When to Return to Orchestrator
+
+- Feature depends on changes that haven't been made yet (e.g., needs a core module that another feature creates)
+- Test failures in unrelated areas that can't be resolved without understanding broader context
+- Ambiguity in feature requirements that can't be resolved from AGENTS.md or feature description
+- A boundary violation would be needed to complete the feature (e.g., need to change scoring.py)

mlx_stack-0.3.8/.factory/validation/absorb-profile/scrutiny/reviews/absorb-profile-into-status.json

@@ -0,0 +1,33 @@
+{
+  "featureId": "absorb-profile-into-status",
+  "reviewedAt": "2026-04-04T20:13:14Z",
+  "commitId": "f684034",
+  "transcriptSkeletonReviewed": true,
+  "diffReviewed": true,
+  "status": "fail",
+  "codeReview": {
+    "summary": "The profile command removal and status hardware integration are mostly correct, but VAL-STATUS-008 is not met in real usage: estimated bandwidth is never marked as estimated when status reads persisted profile.json data.",
+    "issues": [
+      {
+        "file": "src/mlx_stack/core/hardware.py",
+        "line": 301,
+        "severity": "blocking",
+        "description": "load_profile() always reconstructs HardwareProfile with is_estimate=False, so status cannot emit '(estimate)' for unknown chips from saved profile data. status.py only adds the indicator when hw.is_estimate is true (src/mlx_stack/cli/status.py:72), which makes VAL-STATUS-008 fail for persisted profiles."
+      }
+    ]
+  },
+  "sharedStateObservations": [
+    {
+      "area": "conventions",
+      "observation": "Mission guidance and validation contract are misaligned for estimate handling: AGENTS.md says core/hardware.py must remain unchanged, but VAL-STATUS-008 expects estimate signaling that the current load_profile format/path does not preserve.",
+      "evidence": "AGENTS.md:14 mandates core/hardware.py unchanged; validation-contract.md:317-318 requires estimate indicator; core/hardware.py:301 hardcodes is_estimate=False."
+    },
+    {
+      "area": "skills",
+      "observation": "The worker marked skill procedure as followed, but did not follow cli-worker's TDD step ordering (tests-first).",
+      "evidence": "cli-worker/SKILL.md:40-45 requires writing failing tests before implementation; transcript skeleton for session 916ec186-7707-4ffc-80b2-72d2134197f5 shows code edits/deletions were performed before the later 'Now I need to update the existing tests and add new ones' step."
+    }
+  ],
+  "addressesFailureFrom": null,
+  "summary": "Review result: FAIL. The feature correctly removes `profile` and adds hardware output to `status`, but it does not satisfy VAL-STATUS-008 in real persisted-profile flows because estimate metadata is dropped by load_profile()."
+}

mlx_stack-0.3.8/.factory/validation/absorb-profile/scrutiny/reviews/fix-status-estimate-indicator.json

@@ -0,0 +1,15 @@
+{
+  "featureId": "fix-status-estimate-indicator",
+  "reviewedAt": "2026-04-04T20:24:27Z",
+  "commitId": "a477711",
+  "transcriptSkeletonReviewed": true,
+  "diffReviewed": true,
+  "status": "pass",
+  "codeReview": {
+    "summary": "The fix directly resolves the prior VAL-STATUS-008 failure by preserving `is_estimate` through profile serialization/deserialization and exposing that metadata in `status --json`. Compared with the original failing commit (`f684034`), which hardcoded `is_estimate=False` in `load_profile()`, commit `a477711` now reads `is_estimate` from profile JSON (with a backward-compatible default), so table output can correctly show `(estimate)` for estimated bandwidth.",
+    "issues": []
+  },
+  "sharedStateObservations": [],
+  "addressesFailureFrom": "/Users/weae1504/Projects/mlx-stack/.factory/validation/absorb-profile/scrutiny/reviews/absorb-profile-into-status.json",
+  "summary": "Review result: PASS. The fix adequately addresses the original failure by preserving estimate metadata end-to-end and leaving no blocking issues for estimate indication behavior."
+}

mlx_stack-0.3.8/.factory/validation/absorb-profile/scrutiny/synthesis.json

@@ -0,0 +1,33 @@
+{
+  "milestone": "absorb-profile",
+  "round": 2,
+  "status": "pass",
+  "validatorsRun": {
+    "test": {
+      "passed": true,
+      "command": "uv run pytest --cov=src/mlx_stack -x -q --tb=short",
+      "exitCode": 0
+    },
+    "typecheck": {
+      "passed": true,
+      "command": "uv run python -m pyright",
+      "exitCode": 0
+    },
+    "lint": {
+      "passed": true,
+      "command": "uv run ruff check src/ tests/",
+      "exitCode": 0
+    }
+  },
+  "reviewsSummary": {
+    "total": 1,
+    "passed": 1,
+    "failed": 0,
+    "failedFeatures": []
+  },
+  "blockingIssues": [],
+  "appliedUpdates": [],
+  "suggestedGuidanceUpdates": [],
+  "rejectedObservations": [],
+  "previousRound": ".factory/validation/absorb-profile/scrutiny/synthesis.round1.json"
+}

mlx_stack-0.3.8/.factory/validation/absorb-profile/scrutiny/synthesis.round1.json

@@ -0,0 +1,53 @@
+{
+  "milestone": "absorb-profile",
+  "round": 1,
+  "status": "fail",
+  "validatorsRun": {
+    "test": {
+      "passed": true,
+      "command": "uv run pytest --cov=src/mlx_stack -x -q --tb=short",
+      "exitCode": 0
+    },
+    "typecheck": {
+      "passed": true,
+      "command": "uv run python -m pyright",
+      "exitCode": 0
+    },
+    "lint": {
+      "passed": true,
+      "command": "uv run ruff check src/ tests/",
+      "exitCode": 0
+    }
+  },
+  "reviewsSummary": {
+    "total": 1,
+    "passed": 0,
+    "failed": 1,
+    "failedFeatures": [
+      "absorb-profile-into-status"
+    ]
+  },
+  "blockingIssues": [
+    {
+      "featureId": "absorb-profile-into-status",
+      "severity": "blocking",
+      "description": "VAL-STATUS-008 is not satisfied for persisted profiles: load_profile() reconstructs HardwareProfile with is_estimate=false, so status output cannot display '(estimate)' for unknown chips loaded from profile.json."
+    }
+  ],
+  "appliedUpdates": [],
+  "suggestedGuidanceUpdates": [
+    {
+      "target": "AGENTS.md",
+      "suggestion": "Clarify the boundary that core/hardware.py can be minimally updated when required to preserve factual hardware metadata (for example estimate signaling) needed by validation assertions.",
+      "evidence": "AGENTS.md currently says core/hardware.py must remain unchanged, while VAL-STATUS-008 requires estimate signaling and review found load_profile() currently drops that metadata.",
+      "isSystemic": false
+    }
+  ],
+  "rejectedObservations": [
+    {
+      "observation": "Feature worker did not follow cli-worker TDD ordering strictly.",
+      "reason": "already-documented"
+    }
+  ],
+  "previousRound": null
+}

mlx_stack-0.3.8/.factory/validation/absorb-profile/user-testing/flows/status-core-cli.json

@@ -0,0 +1,106 @@
+{
+  "groupId": "status-core-cli",
+  "testedAt": "2026-04-04T20:29:01.801098+00:00",
+  "isolation": {
+    "surface": "CLI",
+    "repoRoot": "/Users/weae1504/Projects/mlx-stack",
+    "missionDir": "/Users/weae1504/.factory/missions/7fc62a3d-138f-4cd2-a601-3f6d1b174b53",
+    "mlxStackHome": "/Users/weae1504/.factory/missions/7fc62a3d-138f-4cd2-a601-3f6d1b174b53/evidence/absorb-profile/status-core-cli-home"
+  },
+  "toolsUsed": [
+    "shell",
+    "uv run mlx-stack"
+  ],
+  "assertionResults": [
+    {
+      "id": "VAL-STATUS-001",
+      "status": "pass",
+      "evidence": {
+        "files": [
+          "absorb-profile/status-core-cli/VAL-STATUS-001-profile-command.txt"
+        ],
+        "observed": "exit=2; contains_no_such_command=True; deprecated_present=False"
+      }
+    },
+    {
+      "id": "VAL-STATUS-002",
+      "status": "pass",
+      "evidence": {
+        "files": [
+          "absorb-profile/status-core-cli/VAL-STATUS-002-main-help.txt"
+        ],
+        "observed": "profile_listed=False; status_listed=True"
+      }
+    },
+    {
+      "id": "VAL-STATUS-003",
+      "status": "pass",
+      "evidence": {
+        "files": [
+          "absorb-profile/status-core-cli/VAL-STATUS-003-status-with-profile.txt"
+        ],
+        "observed": "exit=0; has_chip=True; has_gpu=True; has_memory=True; has_bandwidth=True; has_service_table=True"
+      }
+    },
+    {
+      "id": "VAL-STATUS-004",
+      "status": "pass",
+      "evidence": {
+        "files": [
+          "absorb-profile/status-core-cli/VAL-STATUS-004-status-without-profile.txt"
+        ],
+        "observed": "exit=0; traceback_present=False; has_service_table=True; hardware_section_present=False"
+      }
+    },
+    {
+      "id": "VAL-STATUS-007",
+      "status": "pass",
+      "evidence": {
+        "files": [
+          "absorb-profile/status-core-cli/VAL-STATUS-003-status-with-profile.txt"
+        ],
+        "observed": "missing_columns=[]"
+      }
+    },
+    {
+      "id": "VAL-STATUS-009",
+      "status": "pass",
+      "evidence": {
+        "files": [
+          "absorb-profile/status-core-cli/VAL-STATUS-009-status-help.txt"
+        ],
+        "observed": "exit=0; mentions_hardware_or_chip=True; has_json_flag=True"
+      }
+    }
+  ],
+  "commandsRun": [
+    {
+      "command": "uv run --project /Users/weae1504/Projects/mlx-stack mlx-stack profile",
+      "exitCode": 2,
+      "keyObservation": "Error: No such command 'profile'."
+    },
+    {
+      "command": "uv run --project /Users/weae1504/Projects/mlx-stack mlx-stack --help",
+      "exitCode": 0,
+      "keyObservation": "mlx-stack \u2014 CLI control plane for local LLM infrastructure on Apple Silicon"
+    },
+    {
+      "command": "uv run --project /Users/weae1504/Projects/mlx-stack mlx-stack status",
+      "exitCode": 0,
+      "keyObservation": "Hardware          "
+    },
+    {
+      "command": "uv run --project /Users/weae1504/Projects/mlx-stack mlx-stack status",
+      "exitCode": 0,
+      "keyObservation": "Service Status                              "
+    },
+    {
+      "command": "uv run --project /Users/weae1504/Projects/mlx-stack mlx-stack status --help",
+      "exitCode": 0,
+      "keyObservation": "Usage: mlx-stack status [OPTIONS]"
+    }
+  ],
+  "frictions": [],
+  "blockers": [],
+  "summary": "Tested 6 assertions: 6 passed, 0 failed, 0 blocked."
+}