fix(mcp): flow_server i_will_work_on forwards steps — completes #172

#172 (4c397e1) added IWillWorkOnRequest.steps, the flow_dev route
threading, the _dev_steps_gate, and the developer prompt — but never
updated the roboco-flow MCP tool. flow_server.i_will_work_on exposed
only (task_id, plan) and posted only those, so the agent's tool could
not transmit steps. Every fresh dev claim hit the gate's
incomplete_input missing=['steps'] with no way to satisfy it →
permanent wedge for every code task (observed in smoke-16: be-dev-1
looped ~20 times, then deadlocked — could not claim, block, or idle).

Add the steps parameter and forward it, mirroring the existing
i_will_plan/sub_tasks pattern (which is why PM i_will_plan was never
affected). Update the two body-shape tests and add a steps-passthrough
regression test.

roboco/mcp/flow_server.py

@@ -210,9 +210,28 @@ def give_me_work() -> dict[str, Any]:
     return _post(_role_path("give_me_work"), {})
 
 
-def i_will_work_on(task_id: str, plan: str | None = None) -> dict[str, Any]:
+def i_will_work_on(
-    """Claim/start/recover a task. Works for pending, claimed, needs_revision."""
+    task_id: str,
-    return _post(_role_path("i_will_work_on"), {"task_id": task_id, "plan": plan})
+    plan: str | None = None,
+    steps: list[dict[str, str]] | None = None,
+) -> dict[str, Any]:
+    """Claim/start/recover a task. Works for pending, claimed, needs_revision.
+
+    Args:
+        task_id: UUID of the task you are claiming.
+        plan: One-paragraph narrative of how you'll execute the task.
+        steps: Ordered execution checklist — list of
+            ``{"title": "...", "description": "..."}``. Required on a
+            FRESH claim: the gateway's ``_dev_steps_gate`` rejects an
+            empty or thin list, and the same list is reused as the
+            progress checklist (#173 — completing a step advances %).
+            Re-entry / recovery claims (already-claimed or
+            needs_revision) do not need steps re-supplied.
+    """
+    return _post(
+        _role_path("i_will_work_on"),
+        {"task_id": task_id, "plan": plan, "steps": steps or []},
+    )
 
 
 def open_pr(task_id: str) -> dict[str, Any]:

tests/unit/mcp_servers/test_flow_server.py

@@ -140,7 +140,11 @@ def test_i_will_work_on_passes_plan(flow_module: types.ModuleType) -> None:
         flow_module.i_will_work_on("task-uuid", plan="my plan")
 
     args, kwargs = fake_client.post.call_args
-    assert kwargs["json"] == {"task_id": "task-uuid", "plan": "my plan"}
+    assert kwargs["json"] == {
+        "task_id": "task-uuid",
+        "plan": "my plan",
+        "steps": [],
+    }
     assert "/api/v2/flow/developer/i_will_work_on" in args[0]
 
 
@@ -151,7 +155,28 @@ def test_i_will_work_on_plan_defaults_to_none(flow_module: types.ModuleType) ->
         flow_module.i_will_work_on("task-uuid")
 
     _, kwargs = fake_client.post.call_args
-    assert kwargs["json"] == {"task_id": "task-uuid", "plan": None}
+    assert kwargs["json"] == {"task_id": "task-uuid", "plan": None, "steps": []}
+
+
+def test_i_will_work_on_passes_steps(flow_module: types.ModuleType) -> None:
+    """#172: the MCP tool must forward the steps checklist; the
+    server-side _dev_steps_gate rejects a fresh claim without it, so a
+    tool that drops steps wedges every code task."""
+    fake_client = _make_fake_client({"status": "in_progress"})
+    steps = [
+        {"title": "Checkout", "description": "create branch and read README"},
+        {"title": "Edit", "description": "append the timestamp comment"},
+    ]
+
+    with patch("httpx.Client", return_value=fake_client):
+        flow_module.i_will_work_on("task-uuid", plan="p", steps=steps)
+
+    _, kwargs = fake_client.post.call_args
+    assert kwargs["json"] == {
+        "task_id": "task-uuid",
+        "plan": "p",
+        "steps": steps,
+    }
 
 
 def test_i_am_done_sends_task_id_and_notes(flow_module: types.ModuleType) -> None: