fix(cli_driver): place extra_args before prompt positional

Bugbot flagged that `run_claude` placed the prompt as cmd[7] and then
appended extra_args after it. `claude --print` takes the prompt as the
final positional argument; flags appearing after it (e.g.
`--allowed-tools Bash`, `--image <path>`) are swallowed by the prompt
parser, which silently breaks the tool_use and vision cells.

Build the flag list first, then append the prompt last. Add a unit
test that pins the ordering.

Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>

tests/claude_code/_driver_unit_tests/test_cli_driver.py

@@ -63,6 +63,29 @@ def test_run_claude_assembles_command_correctly():
     assert cmd[-1] == "hello"
 
 
+def test_run_claude_places_extra_args_before_prompt():
+    """`claude --print` expects the prompt as the final positional arg.
+
+    Flags appearing after the prompt are ignored or eaten by the prompt
+    parser, which silently broke the tool_use and vision cells before the
+    fix. Pin the ordering: every flag (including caller-supplied
+    `extra_args`) must precede the prompt.
+    """
+    runner, captured = _make_runner(stdout="")
+    run_claude(
+        prompt="say hi",
+        model="claude-haiku-4-5",
+        base_url="http://localhost:4000",
+        api_key="sk-test",
+        extra_args=["--allowed-tools", "Bash"],
+        runner=runner,
+    )
+    cmd = captured["cmd"]
+    assert cmd[-1] == "say hi"
+    prompt_idx = cmd.index("say hi")
+    assert cmd[prompt_idx - 2 : prompt_idx] == ["--allowed-tools", "Bash"]
+
+
 def test_run_claude_overlays_proxy_env():
     runner, captured = _make_runner(stdout="")
     run_claude(

tests/claude_code/cli_driver.py

@@ -78,6 +78,11 @@ def run_claude(
     if not api_key:
         raise ValueError("api_key must be a non-empty string")
 
+    # `claude --print` takes the prompt as the **last positional argument**.
+    # Flags must come before it, otherwise they're parsed as part of the
+    # prompt (or silently dropped, depending on the CLI version) and the
+    # tool_use / vision cells fail with confusing "no tool_use observed"
+    # errors. Build the flag list first, then append the prompt last.
     cmd: List[str] = [
         cli_path,
         "--print",
@@ -86,10 +91,10 @@ def run_claude(
         "--verbose",
         "--model",
         model,
-        prompt,
     ]
     if extra_args:
         cmd.extend(extra_args)
+    cmd.append(prompt)
 
     env = {**os.environ, **(extra_env or {})}
     env["ANTHROPIC_BASE_URL"] = base_url