lemming-cli 0.8.0__tar.gz → 0.8.1__tar.gz

{lemming_cli-0.8.0 → lemming_cli-0.8.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: lemming-cli
-Version: 0.8.0
+Version: 0.8.1
 Summary: An autonomous, iterative task runner for AI coding agents.
 Project-URL: Homepage, https://github.com/owahltinez/lemming
 Project-URL: Repository, https://github.com/owahltinez/lemming
@@ -226,8 +226,8 @@ lemming serve --tunnel cloudflare
   folders directly from the UI.
 - **Browse Files**: Quickly open your workspace in a separate window to inspect
   files and directory structure alongside the roadmap.
-- **Interactive Controls**: Add tasks, edit the goal, and manage the execution
-  loop from your browser.
+- **Interactive Controls**: Add tasks and pick their reviews, edit the goal, and
+  manage the execution loop from your browser.
 
 ---
 
@@ -387,7 +387,8 @@ These come before the subcommand and apply to all of them.
   `--model`, `--oneshot`, and `--hooks` (the task's reviews, comma-separated;
   inherited from `--parent` when omitted).
 - **`edit <id>`**: Modify a task's description, runner, model, position,
-  oneshot flag (`--oneshot/--no-oneshot`), or reviews (`--hooks`).
+  oneshot flag (`--oneshot/--no-oneshot`), or reviews (`--hooks`;
+  `--project-hooks` restores the project's).
 - **`brief <id> [text]`**: View or set a task's long-form brief. Unlike the
   description it has no length cap, and it is appended to the runner prompt
   automatically — the right home for measured timings, exact failing selectors,

{lemming_cli-0.8.0 → lemming_cli-0.8.1}/README.md

@@ -206,8 +206,8 @@ lemming serve --tunnel cloudflare
   folders directly from the UI.
 - **Browse Files**: Quickly open your workspace in a separate window to inspect
   files and directory structure alongside the roadmap.
-- **Interactive Controls**: Add tasks, edit the goal, and manage the execution
-  loop from your browser.
+- **Interactive Controls**: Add tasks and pick their reviews, edit the goal, and
+  manage the execution loop from your browser.
 
 ---
 
@@ -367,7 +367,8 @@ These come before the subcommand and apply to all of them.
   `--model`, `--oneshot`, and `--hooks` (the task's reviews, comma-separated;
   inherited from `--parent` when omitted).
 - **`edit <id>`**: Modify a task's description, runner, model, position,
-  oneshot flag (`--oneshot/--no-oneshot`), or reviews (`--hooks`).
+  oneshot flag (`--oneshot/--no-oneshot`), or reviews (`--hooks`;
+  `--project-hooks` restores the project's).
 - **`brief <id> [text]`**: View or set a task's long-form brief. Unlike the
   description it has no length cap, and it is appended to the runner prompt
   automatically — the right home for measured timings, exact failing selectors,

{lemming_cli-0.8.0 → lemming_cli-0.8.1}/docs/HOOKS.md

@@ -193,10 +193,14 @@ A task can pick its own reviews, which replace the project's for that task:
 
 ```bash
 lemming add "Redesign the settings page" --hooks ux,testing
-lemming edit <id> --hooks ''     # no reviews for this task
-lemming edit <id> --hooks '*'    # every review available now
+lemming edit <id> --hooks ''       # no reviews for this task
+lemming edit <id> --hooks '*'      # every review available now
+lemming edit <id> --project-hooks  # back to the project's reviews
 ```
 
+In the dashboard, tick **Custom reviews** when adding or editing a task to pick
+its reviews; leave it unticked to use the project's.
+
 `--hooks` accepts only reviews (priority below 90). Hooks at 90 and above, such
 as `roadmap`, orchestrate the queue and always follow the project's list. A task
 added with `--parent` and no `--hooks` inherits its parent's choice, so a split

{lemming_cli-0.8.0 → lemming_cli-0.8.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "lemming-cli"
-version = "0.8.0"
+version = "0.8.1"
 description = "An autonomous, iterative task runner for AI coding agents."
 readme = "README.md"
 authors = [

{lemming_cli-0.8.0 → lemming_cli-0.8.1}/src/lemming/api/tasks.py

@@ -3,7 +3,7 @@
 import fastapi
 import pydantic
 
-from .. import models, paths, persistence
+from .. import hooks, models, paths, persistence
 from ..tasks import lifecycle, operations, queries
 from . import context, loop
 
@@ -27,6 +27,11 @@ class AddTaskRequest(pydantic.BaseModel):
     index: int = -1
     parent: str | None = None
     parent_tasks_file: str | None = None
+    hooks: list[str] | None = None
+
+
+# Validates an update's hooks against the field they are stored in.
+_TASK_HOOKS = pydantic.TypeAdapter(models.Task.model_fields["hooks"].annotation)
 
 
 @router.post("/api/tasks")
@@ -54,6 +59,9 @@ def add_task(
             index=task.index,
             parent=task.parent,
             parent_tasks_file=task.parent_tasks_file,
+            hooks=None
+            if task.hooks is None
+            else hooks.parse_reviews(task.hooks, tasks_file),
         )
     except ValueError as e:
         raise fastapi.HTTPException(400, str(e))
@@ -85,7 +93,10 @@ def update_task(
     update: dict,
     project: str | None = None,
 ):
-    """Update a task's description, runner, position, status, or parent."""
+    """Update a task's description, runner, position, status, parent or hooks.
+
+    An explicit null for hooks restores the project's reviews.
+    """
     tasks_file = context.resolve_tasks_file(request.app.state, project)
     status = update.get("status")
     if status == models.TaskStatus.SUPERSEDED:
@@ -113,6 +124,9 @@ def update_task(
             require_progress = True
 
     try:
+        task_hooks = _TASK_HOOKS.validate_python(update.get("hooks"))
+        if task_hooks is not None:
+            task_hooks = hooks.parse_reviews(task_hooks, tasks_file)
         updated_task = operations.update_task(
             tasks_file,
             task_id,
@@ -122,6 +136,8 @@ def update_task(
             status=status,
             require_progress=require_progress,
             parent=update.get("parent"),
+            hooks=task_hooks,
+            project_hooks="hooks" in update and task_hooks is None,
         )
         return updated_task
     except ValueError as e:

{lemming_cli-0.8.0 → lemming_cli-0.8.1}/src/lemming/api/tasks_test.py

@@ -53,6 +53,53 @@ def test_add_task_with_runner(client, test_tasks):
     assert response.json()["runner"] == "claude"
 
 
+def test_add_task_with_hooks(client, test_tasks):
+    response = client.post(
+        "/api/tasks",
+        json={"description": "Reviewed", "hooks": ["ux", "readability"]},
+    )
+    assert response.status_code == 200
+    assert response.json()["hooks"] == ["readability", "ux"]
+
+
+def test_add_task_rejects_invalid_hooks(client, test_tasks):
+    for hooks, message in (
+        (["nope"], "Unknown review 'nope'"),
+        (["roadmap"], "orchestrates the roadmap"),
+    ):
+        response = client.post(
+            "/api/tasks", json={"description": "Bad", "hooks": hooks}
+        )
+        assert response.status_code == 400
+        assert message in response.json()["detail"]
+    assert len(persistence.load_tasks(test_tasks).tasks) == 3
+
+
+def test_update_task_hooks_and_reset(client, test_tasks):
+    url = "/api/tasks/task2/update"
+
+    response = client.post(url, json={"hooks": []})
+    assert response.status_code == 200
+    assert response.json()["hooks"] == []
+
+    response = client.post(url, json={"description": "Still custom"})
+    assert response.json()["hooks"] == []
+
+    response = client.post(url, json={"hooks": None})
+    assert response.status_code == 200
+    assert response.json()["hooks"] is None
+
+
+def test_update_task_rejects_invalid_hooks(client, test_tasks):
+    for hooks in (["nope"], ["roadmap"], "ux", [1]):
+        response = client.post("/api/tasks/task2/update", json={"hooks": hooks})
+        assert response.status_code == 400, hooks
+    task2 = next(
+        t for t in persistence.load_tasks(test_tasks).tasks if t.id == "task2"
+    )
+    assert task2.hooks is None
+
+
 def test_delete_completed_tasks(client, test_tasks):
     response = client.post("/api/tasks/delete-completed")
     assert response.status_code == 200

{lemming_cli-0.8.0 → lemming_cli-0.8.1}/src/lemming/cli/tasks.py

@@ -222,6 +222,11 @@ def add(
         "every review, '' for none)."
     ),
 )
+@click.option(
+    "--project-hooks",
+    is_flag=True,
+    help="Restore the project's reviews for this task.",
+)
 @click.pass_context
 def edit(
     ctx: click.Context,
@@ -235,8 +240,13 @@ def edit(
     parent_tasks_file: str | None,
     oneshot: bool | None,
     hooks_value: str | None,
+    project_hooks: bool,
 ):
     """Edits a task's description, runner, model, position, parent or flags."""
+    if project_hooks and hooks_value is not None:
+        click.echo("Error: Cannot provide both --hooks and --project-hooks.")
+        ctx.exit(1)
+
     if file:
         if description:
             click.echo("Error: Cannot provide both description and --file.")
@@ -254,11 +264,12 @@ def edit(
         and parent_tasks_file is None
         and oneshot is None
         and hooks_value is None
+        and not project_hooks
     ):
         click.echo(
             "Error: At least one of --description, --runner, --model,"
-            " --index, --parent, --parent-tasks-file, --oneshot, or --hooks"
-            " must be provided."
+            " --index, --parent, --parent-tasks-file, --oneshot, --hooks,"
+            " or --project-hooks must be provided."
         )
         ctx.exit(1)
 
@@ -276,6 +287,7 @@ def edit(
             parent_tasks_file=parent_tasks_file,
             oneshot=oneshot,
             hooks=_parse_task_hooks(hooks_value, tasks_file),
+            project_hooks=project_hooks,
         )
         click.echo(f"Task {target_task.id} updated.")
     except ValueError as e:

{lemming_cli-0.8.0 → lemming_cli-0.8.1}/src/lemming/cli/tasks_test.py

@@ -454,6 +454,30 @@ class TestCLITaskModel(unittest.TestCase):
         updated = persistence.load_tasks(self.test_tasks_file).tasks[0]
         self.assertEqual(updated.hooks, [])
 
+    def test_edit_project_hooks_restores_project_default(self):
+        task = self._add("--hooks", "ux")
+
+        result = self.cli_runner.invoke(
+            cli.cli, self.base_args + ["edit", task.id, "--project-hooks"]
+        )
+
+        self.assertEqual(result.exit_code, 0, result.output)
+        updated = persistence.load_tasks(self.test_tasks_file).tasks[0]
+        self.assertIsNone(updated.hooks)
+
+    def test_edit_rejects_hooks_with_project_hooks(self):
+        task = self._add("--hooks", "ux")
+
+        result = self.cli_runner.invoke(
+            cli.cli,
+            self.base_args
+            + ["edit", task.id, "--hooks", "", "--project-hooks"],
+        )
+
+        self.assertEqual(result.exit_code, 1, result.output)
+        updated = persistence.load_tasks(self.test_tasks_file).tasks[0]
+        self.assertEqual(updated.hooks, ["ux"])
+
     def test_add_without_hooks_uses_project_default(self):
         self.assertIsNone(self._add().hooks)
 

{lemming_cli-0.8.0 → lemming_cli-0.8.1}/src/lemming/evals/container.py

@@ -78,6 +78,43 @@ def prune_build_cache(docker: str = "docker") -> None:
         pass
 
 
+def uv_command(
+    uv_args: list[str],
+    options: list[str],
+    image: str = DEFAULT_IMAGE,
+    docker: str = "docker",
+) -> list[str]:
+    """Returns the docker argv that runs uv in the image's lemming project.
+
+    The image's own entrypoint serves the web UI, so it is replaced with uv
+    and the command runs with the lemming install's interpreter.
+
+    Args:
+        uv_args: Arguments following uv run --project /opt/lemming.
+        options: Extra docker run options (mounts, env, network).
+        image: Image tag to run.
+        docker: Docker-compatible CLI binary to invoke.
+
+    Returns:
+        The full docker run argv.
+    """
+    return [
+        docker,
+        "run",
+        "--rm",
+        *options,
+        "--workdir",
+        WORKSPACE_MOUNT,
+        "--entrypoint",
+        "uv",
+        image,
+        "run",
+        "--project",
+        "/opt/lemming",
+        *uv_args,
+    ]
+
+
 def trial_command(
     workspace: pathlib.Path,
     lemming_home: pathlib.Path,
@@ -103,42 +140,28 @@ def trial_command(
     Returns:
         The full docker run argv.
     """
-    command = [
-        docker,
-        "run",
-        "--rm",
+    options = [
         "--volume",
         f"{workspace}:{WORKSPACE_MOUNT}",
         "--volume",
         f"{lemming_home}:{HOME_MOUNT}",
         "--env",
         f"LEMMING_HOME={HOME_MOUNT}",
-        "--workdir",
-        WORKSPACE_MOUNT,
-        "--entrypoint",
-        "uv",
     ]
 
     # Bare --env NAME keeps the secret's value out of the argv.
     for name in forward_env:
         if os.environ.get(name):
-            command.extend(["--env", name])
+            options.extend(["--env", name])
     for spec in volumes:
-        command.extend(["--volume", spec])
+        options.extend(["--volume", spec])
 
-    command.extend(
-        [
-            image,
-            "run",
-            "--project",
-            "/opt/lemming",
-            "python",
-            "-m",
-            "lemming.evals.trial",
-            *trial_args,
-        ]
+    return uv_command(
+        ["python", "-m", "lemming.evals.trial", *trial_args],
+        options,
+        image=image,
+        docker=docker,
     )
-    return command
 
 
 def run_trial(

{lemming_cli-0.8.0 → lemming_cli-0.8.1}/src/lemming/evals/harness.py

@@ -7,6 +7,7 @@ docker-based runner from the container module, tests use in-process fakes.
 """
 
 import concurrent.futures
+import contextvars
 import dataclasses
 import json
 import os
@@ -106,6 +107,10 @@ class HarnessConfig:
     volumes: tuple[str, ...] = ()
 
 
+# The running trial's config, for graders that execute workspace code.
+TRIAL_CONFIG = contextvars.ContextVar("trial_config", default=HarnessConfig())
+
+
 @dataclasses.dataclass
 class TrialResult:
     """Outcome of a single graded trial.
@@ -356,6 +361,7 @@ def _execute_trial(
     lemming_home = trial_dir / "home"
     lemming_home.mkdir(parents=True, exist_ok=True)
 
+    TRIAL_CONFIG.set(config)
     started = time.monotonic()
     error = ""
     try:

{lemming_cli-0.8.0 → lemming_cli-0.8.1}/src/lemming/evals/harness_test.py

@@ -71,6 +71,27 @@ class TestRunSuite(HarnessTestCase):
 
         self.assertEqual([r.passed for r in results], [True, True])
 
+    def test_graders_see_the_run_config(self):
+        # Graders that run workspace code must use the run's own image.
+        scenario = scenarios.Scenario(
+            name="config-demo",
+            summary="Reports the image it is graded with.",
+            build=lambda workspace: workspace.mkdir(parents=True),
+            grade=lambda workspace: [
+                scenarios.Check(
+                    name=harness.TRIAL_CONFIG.get().image, passed=True
+                )
+            ],
+        )
+        config = dataclasses.replace(self.config, image="custom-image")
+
+        results = harness.run_suite(
+            [scenario], self.run_dir, config, lambda *args: None
+        )
+
+        names = {check.name for r in results for check in r.checks}
+        self.assertEqual(names, {"custom-image"})
+
     def test_infrastructure_errors_are_captured(self):
         def broken_runner(scenario, workspace, lemming_home, config):
             raise RuntimeError("docker daemon unreachable")

{lemming_cli-0.8.0 → lemming_cli-0.8.1}/src/lemming/evals/readability.py

@@ -23,7 +23,7 @@ import subprocess
 import sys
 
 from .. import models
-from . import container, fixtures, metrics, scenarios
+from . import container, fixtures, harness, metrics, scenarios
 
 _GOAL = (
     "Build a small calculator CLI in calc/ with add, subtract, and "
@@ -95,26 +95,22 @@ def _save_finished_task(workspace: pathlib.Path, progress: list[str]) -> None:
 
 def _check_tests_pass(workspace: pathlib.Path) -> scenarios.Check:
     """Checks that the fixture's unit tests still pass after the hook."""
-    if shutil.which("docker"):
-        cmd = [
-            "docker",
-            "run",
-            "--rm",
-            "--network=none",
-            "-v",
-            f"{workspace}:{container.WORKSPACE_MOUNT}:ro",
-            "-w",
-            container.WORKSPACE_MOUNT,
-            container.DEFAULT_IMAGE,
-            "python3",
-            "-m",
-            "unittest",
-            "discover",
-            "-p",
-            "*_test.py",
-        ]
+    unittest_args = ["-m", "unittest", "discover", "-p", "*_test.py"]
+    config = harness.TRIAL_CONFIG.get()
+    if shutil.which(config.docker):
+        # Offline, so uv must not sync the dev group the image lacks.
+        cmd = container.uv_command(
+            ["--no-sync", "python", *unittest_args],
+            [
+                "--network=none",
+                "--volume",
+                f"{workspace}:{container.WORKSPACE_MOUNT}:ro",
+            ],
+            image=config.image,
+            docker=config.docker,
+        )
     else:
-        cmd = [sys.executable, "-m", "unittest", "discover", "-p", "*_test.py"]
+        cmd = [sys.executable, *unittest_args]
     result = subprocess.run(
         cmd,
         cwd=workspace,

{lemming_cli-0.8.0 → lemming_cli-0.8.1}/src/lemming/evals/readability_test.py

@@ -5,7 +5,7 @@ import unittest
 import unittest.mock
 
 from lemming import models
-from lemming.evals import fixtures, readability, scenarios, suites
+from lemming.evals import fixtures, harness, readability, scenarios, suites
 from lemming.tasks import operations, progress
 
 
@@ -380,12 +380,15 @@ class TestFixtureIntegrity(ScenarioTestCase):
                 check = readability._check_tests_pass(workspace)
                 self.assertTrue(check.passed, check.detail)
 
-    def test_check_tests_pass_uses_docker_when_available(self):
+    def test_check_tests_pass_runs_offline_in_the_run_image(self):
         completed = unittest.mock.Mock(returncode=0, stderr="")
+        config = harness.HarnessConfig(image="custom-image", docker="podman")
+        token = harness.TRIAL_CONFIG.set(config)
+        self.addCleanup(harness.TRIAL_CONFIG.reset, token)
         with (
             unittest.mock.patch(
                 "lemming.evals.readability.shutil.which",
-                return_value="/usr/bin/docker",
+                return_value="/usr/bin/podman",
             ),
             unittest.mock.patch(
                 "lemming.evals.readability.subprocess.run",
@@ -396,19 +399,19 @@ class TestFixtureIntegrity(ScenarioTestCase):
 
         self.assertTrue(check.passed)
         cmd = mock_run.call_args[0][0]
+        self.assertEqual(cmd[0], "podman")
+        self.assertIn("--network=none", cmd)
+        self.assertIn(f"{self.workspace}:/workspace:ro", cmd)
+        self.assertEqual(cmd[cmd.index("--entrypoint") + 1], "uv")
+        # The image ships no dev dependencies, and offline uv cannot sync.
         self.assertEqual(
-            cmd,
+            cmd[cmd.index("custom-image") + 1 :],
             [
-                "docker",
                 "run",
-                "--rm",
-                "--network=none",
-                "-v",
-                f"{self.workspace}:/workspace:ro",
-                "-w",
-                "/workspace",
-                "lemming-evals",
-                "python3",
+                "--project",
+                "/opt/lemming",
+                "--no-sync",
+                "python",
                 "-m",
                 "unittest",
                 "discover",

{lemming_cli-0.8.0 → lemming_cli-0.8.1}/src/lemming/tasks/operations.py

@@ -273,6 +273,7 @@ def update_task(
     parent_tasks_file: str | None = None,
     oneshot: bool | None = None,
     hooks: list[str] | None = None,
+    project_hooks: bool = False,
     force: bool = False,
 ) -> models.Task:
     """Updates an existing task.
@@ -291,6 +292,7 @@ def update_task(
         parent_tasks_file: New parent tasks file path.
         oneshot: New oneshot flag; None leaves it unchanged.
         hooks: New reviews to run after the task; None leaves them unchanged.
+        project_hooks: If True, the task goes back to the project's reviews.
         force: If True, force status transition even if task is in progress.
 
     Returns:
@@ -326,7 +328,9 @@ def update_task(
             target.model = model or None
         if oneshot is not None:
             target.oneshot = oneshot
-        if hooks is not None:
+        if project_hooks:
+            target.hooks = None
+        elif hooks is not None:
             target.hooks = hooks
         if parent is not None:
             if parent == "":

{lemming_cli-0.8.0 → lemming_cli-0.8.1}/src/lemming/tasks/operations_test.py

@@ -55,6 +55,9 @@ def test_update_task_hooks(tmp_path):
     cleared = operations.update_task(tasks_file, task.id, hooks=[])
     assert cleared.hooks == []
 
+    reset = operations.update_task(tasks_file, task.id, project_hooks=True)
+    assert reset.hooks is None
+
 
 def test_add_task(tmp_path):
     tasks_file = tmp_path / "tasks.yml"