fix: validate script_path exists + tell agent to call generate_job_file

Two problems with the prior prepare_submit_job flow:

  1. The agent could pass a script_path that only existed in its own
     context (LLM-generated code not yet on disk) or a path on the host
     filesystem that's invisible inside the container. The new check
     surfaces this as a 400 with a clear remediation hint instead of
     letting spark-submit fail later with an opaque FileNotFoundError -> 500.

  2. The container-isolation issue: any path the agent gives is interpreted
     inside the container. The two ways a file can legitimately exist there
     are (a) generate_job_file(code=...) just wrote it to
     SPARK_EXECUTOR_JOBS_DIR (the default ./data/jobs/ is the only
     gitignored dir that survives restarts), or (b) a host dir was
     mounted via -v. The error message spells both out so an agent can
     self-correct.

Implementation:
  - submit.py: new _check_script_path() that raises ValueError (-> 400)
    when the path is missing, empty, or a directory. Called in both
    prepare_submit_job and confirm_submit_job (defense in depth).
  - prepare_submit_job checks the connection FIRST (KeyError -> 404)
    before the script (ValueError -> 400), so an agent with both problems
    sees the more fundamental 'unknown connection' error first.
  - server.py / requests.py: route description and Pydantic field
    description spell out the generate_job_file pattern so an LLM
    reading the tool schema learns the right next step.

8 new tests; 8 existing tests adjusted to create real files (they used
synthetic /tmp/*.py paths that don't exist).

tests/unit/test_submit_tool.py

@@ -26,6 +26,14 @@ def _fresh(tmp_path: Path, monkeypatch):
     ))
 
 
+@pytest.fixture
+def real_script(tmp_path: Path) -> Path:
+    """A real .py file on disk that satisfies prepare_submit_job's path check."""
+    p = tmp_path / "demo.py"
+    p.write_text("print('hi from test')\n")
+    return p
+
+
 def _last_pending_id() -> str:
     pending = submit.pending_store.list_all()
     assert pending, "no pending submission was created"
@@ -34,16 +42,16 @@ def _last_pending_id() -> str:
 
 # --- prepare_submit_job ---
 
-def test_prepare_does_not_invoke_spark_submit(monkeypatch):
+def test_prepare_does_not_invoke_spark_submit(monkeypatch, real_script):
     called = []
     monkeypatch.setattr(submit, "run_spark_submit", lambda cmd: called.append(cmd))
-    out = submit.prepare_submit_job(connection="prod", script_path="/tmp/j.py")
+    out = submit.prepare_submit_job(connection="prod", script_path=str(real_script))
     assert called == []  # never called
     assert out["status"] == "PENDING"
     assert out["pending_id"].startswith("p_")
 
 
-def test_prepare_persists_pending_with_snapshot(monkeypatch):
+def test_prepare_persists_pending_with_snapshot(monkeypatch, real_script):
     monkeypatch.setattr(submit, "run_spark_submit", lambda cmd: None)
     connection_store.store.save(
         Connection(
@@ -54,7 +62,7 @@ def test_prepare_persists_pending_with_snapshot(monkeypatch):
             spark_conf={"spark.sql.shuffle.partitions": "200"},
         )
     )
-    submit.prepare_submit_job(connection="prod", script_path="/tmp/j.py", queue="research")
+    submit.prepare_submit_job(connection="prod", script_path=str(real_script), queue="research")
     p = submit.pending_store.get(_last_pending_id())
     assert p is not None
     assert p.connection == "prod"
@@ -63,18 +71,57 @@ def test_prepare_persists_pending_with_snapshot(monkeypatch):
     assert p.yarn_rm_url == "http://rm:8088"
     assert p.spark_conf == {"spark.sql.shuffle.partitions": "200"}
     assert p.queue == "research"
-    assert p.script_path == "/tmp/j.py"
+    assert p.script_path == str(real_script)
 
 
-def test_prepare_raises_for_unknown_connection():
+def test_prepare_raises_for_unknown_connection(real_script):
     with pytest.raises(KeyError, match="missing"):
-        submit.prepare_submit_job(connection="missing", script_path="/tmp/j.py")
+        submit.prepare_submit_job(connection="missing", script_path=str(real_script))
 
 
-def test_prepare_snapshots_connection_at_prepare_time(monkeypatch):
+# --- script_path validation ---
+
+def test_prepare_rejects_nonexistent_script_path(tmp_path):
+    missing = str(tmp_path / "does_not_exist.py")
+    with pytest.raises(ValueError, match="does not exist or is not a file"):
+        submit.prepare_submit_job(connection="prod", script_path=missing)
+
+
+def test_prepare_rejects_directory_as_script_path(tmp_path):
+    """A directory is not a file, even if it exists."""
+    with pytest.raises(ValueError, match="does not exist or is not a file"):
+        submit.prepare_submit_job(connection="prod", script_path=str(tmp_path))
+
+
+def test_prepare_rejects_empty_script_path():
+    with pytest.raises(ValueError, match="does not exist or is not a file"):
+        submit.prepare_submit_job(connection="prod", script_path="")
+
+
+def test_prepare_error_message_mentions_generate_job_file(tmp_path):
+    """The agent must be told to call generate_job_file first."""
+    with pytest.raises(ValueError, match="generate_job_file"):
+        submit.prepare_submit_job(
+            connection="prod", script_path=str(tmp_path / "x.py")
+        )
+
+
+def test_confirm_rejects_if_script_was_deleted_after_prepare(tmp_path, monkeypatch):
+    """Defense in depth: file present at prepare, gone by confirm -> 400."""
+    script = tmp_path / "demo.py"
+    script.write_text("print('hi')\n")
+    submit.prepare_submit_job(connection="prod", script_path=str(script))
+    pid = _last_pending_id()
+    # Simulate the file being deleted (e.g. cleanup job) between prepare and confirm
+    script.unlink()
+    with pytest.raises(ValueError, match="does not exist or is not a file"):
+        submit.confirm_submit_job(pending_id=pid)
+
+
+def test_prepare_snapshots_connection_at_prepare_time(monkeypatch, real_script):
     """Editing the connection between prepare and confirm must NOT silently retarget."""
     monkeypatch.setattr(submit, "run_spark_submit", lambda cmd: None)
-    submit.prepare_submit_job(connection="prod", script_path="/tmp/j.py")
+    submit.prepare_submit_job(connection="prod", script_path=str(real_script))
     p = submit.pending_store.get(_last_pending_id())
     # Now mutate the connection
     connection_store.store.save(
@@ -88,8 +135,8 @@ def test_prepare_snapshots_connection_at_prepare_time(monkeypatch):
 
 # --- confirm_submit_job ---
 
-def test_confirm_invokes_spark_submit_and_marks_submitted(monkeypatch):
-    submit.prepare_submit_job(connection="prod", script_path="/tmp/j.py")
+def test_confirm_invokes_spark_submit_and_marks_submitted(monkeypatch, real_script):
+    submit.prepare_submit_job(connection="prod", script_path=str(real_script))
     pid = _last_pending_id()
     fake_proc = type("P", (), {
         "returncode": 0,
@@ -100,7 +147,7 @@ def test_confirm_invokes_spark_submit_and_marks_submitted(monkeypatch):
     cmd = m.call_args.args[0]
     assert "yarn" in cmd
     assert "cluster" in cmd
-    assert cmd[-1] == "/tmp/j.py"
+    assert cmd[-1] == str(real_script)
     assert result.application_id == "application_17400000001"
     # pending updated
     p = submit.pending_store.get(pid)
@@ -116,8 +163,8 @@ def test_confirm_raises_for_unknown_pending_id():
         submit.confirm_submit_job(pending_id="missing")
 
 
-def test_confirm_refuses_non_pending_status(monkeypatch):
-    submit.prepare_submit_job(connection="prod", script_path="/tmp/j.py")
+def test_confirm_refuses_non_pending_status(monkeypatch, real_script):
+    submit.prepare_submit_job(connection="prod", script_path=str(real_script))
     pid = _last_pending_id()
     # Mark it CANCELLED first
     p = submit.pending_store.get(pid)
@@ -127,9 +174,9 @@ def test_confirm_refuses_non_pending_status(monkeypatch):
         submit.confirm_submit_job(pending_id=pid)
 
 
-def test_confirm_marks_failed_on_spark_submit_error(monkeypatch):
+def test_confirm_marks_failed_on_spark_submit_error(monkeypatch, real_script):
     from spark_executor.core.spark_submit import SparkSubmitError
-    submit.prepare_submit_job(connection="prod", script_path="/tmp/j.py")
+    submit.prepare_submit_job(connection="prod", script_path=str(real_script))
     pid = _last_pending_id()
     def _raise(_cmd):
         raise SparkSubmitError("boom")
@@ -147,20 +194,23 @@ def test_list_pending_jobs_empty():
     assert submit.list_pending_jobs() == []
 
 
-def test_list_pending_jobs_returns_all(monkeypatch):
+def test_list_pending_jobs_returns_all(monkeypatch, tmp_path):
+    a = tmp_path / "a.py"
+    b = tmp_path / "b.py"
+    a.write_text("a\n"); b.write_text("b\n")
     monkeypatch.setattr(submit, "run_spark_submit", lambda cmd: None)
-    submit.prepare_submit_job(connection="prod", script_path="/tmp/a.py")
-    submit.prepare_submit_job(connection="prod", script_path="/tmp/b.py")
+    submit.prepare_submit_job(connection="prod", script_path=str(a))
+    submit.prepare_submit_job(connection="prod", script_path=str(b))
     out = submit.list_pending_jobs()
-    assert {p["script_path"] for p in out} == {"/tmp/a.py", "/tmp/b.py"}
+    assert {p["script_path"] for p in out} == {str(a), str(b)}
     assert all(p["status"] == "PENDING" for p in out)
 
 
 # --- get_pending_job ---
 
-def test_get_pending_job_returns_dump(monkeypatch):
+def test_get_pending_job_returns_dump(monkeypatch, real_script):
     monkeypatch.setattr(submit, "run_spark_submit", lambda cmd: None)
-    submit.prepare_submit_job(connection="prod", script_path="/tmp/j.py")
+    submit.prepare_submit_job(connection="prod", script_path=str(real_script))
     pid = _last_pending_id()
     out = submit.get_pending_job(pid)
     assert out["pending_id"] == pid
@@ -175,9 +225,9 @@ def test_get_pending_job_unknown_raises():
 
 # --- cancel_pending_job ---
 
-def test_cancel_pending_job_marks_cancelled(monkeypatch):
+def test_cancel_pending_job_marks_cancelled(monkeypatch, real_script):
     monkeypatch.setattr(submit, "run_spark_submit", lambda cmd: None)
-    submit.prepare_submit_job(connection="prod", script_path="/tmp/j.py")
+    submit.prepare_submit_job(connection="prod", script_path=str(real_script))
     pid = _last_pending_id()
     out = submit.cancel_pending_job(pid)
     assert out == {"pending_id": pid, "status": "CANCELLED"}
@@ -189,8 +239,8 @@ def test_cancel_pending_job_unknown_raises():
         submit.cancel_pending_job("missing")
 
 
-def test_cancel_pending_job_refuses_submitted(monkeypatch):
-    submit.prepare_submit_job(connection="prod", script_path="/tmp/j.py")
+def test_cancel_pending_job_refuses_submitted(monkeypatch, real_script):
+    submit.prepare_submit_job(connection="prod", script_path=str(real_script))
     pid = _last_pending_id()
     p = submit.pending_store.get(pid)
     p.status = "SUBMITTED"