fix(wrap/claude): keep --1m effective when an explicit --model is passed through

Ensure explicit Claude model arguments retain the 1M context suffix (#2915).

headroom/cli/wrap.py

@@ -303,6 +303,33 @@ def _resolve_1m_model(current: str | None) -> str:
     return base if base.endswith(_CONTEXT_1M_SUFFIX) else f"{base}{_CONTEXT_1M_SUFFIX}"
 
 
+def _apply_1m_to_claude_args(args: tuple[str, ...]) -> tuple[tuple[str, ...], str | None]:
+    """Add the ``[1m]`` suffix to an explicit ``--model`` in pass-through args.
+
+    Claude Code gives the ``--model`` CLI flag precedence over the
+    ``ANTHROPIC_MODEL`` env var, so when a user passes both ``--1m`` and
+    ``--model X`` the env-var suffix is silently shadowed and the session caps at
+    200k (#2915). Rewriting the flag's value the same way ``_resolve_1m_model``
+    rewrites the env var keeps ``--1m`` effective on the higher-precedence flag.
+
+    Handles ``--model VALUE`` and ``--model=VALUE`` (the first occurrence only, as
+    Claude Code honours the first). Idempotent via ``_resolve_1m_model``. Returns
+    ``(new_args, rewritten_value)``; ``rewritten_value`` is ``None`` when no
+    ``--model`` was present (the env-var path already covers that case).
+    """
+    out = list(args)
+    for i, arg in enumerate(out):
+        if arg == "--model" and i + 1 < len(out):
+            rewritten = _resolve_1m_model(out[i + 1])
+            out[i + 1] = rewritten
+            return tuple(out), rewritten
+        if arg.startswith("--model="):
+            rewritten = _resolve_1m_model(arg.split("=", 1)[1])
+            out[i] = f"--model={rewritten}"
+            return tuple(out), rewritten
+    return tuple(out), None
+
+
 def _normalize_tool_search_mode(value: str) -> str:
     """Validate an ``ENABLE_TOOL_SEARCH`` value and return it normalized.
 
@@ -4728,10 +4755,18 @@ def claude(
         # force it via ANTHROPIC_MODEL on the launched process.
         if context_1m:
             env[_ANTHROPIC_MODEL_ENV] = _resolve_1m_model(env.get(_ANTHROPIC_MODEL_ENV))
-            click.echo(
-                f"  {_ANTHROPIC_MODEL_ENV}={env[_ANTHROPIC_MODEL_ENV]} "
-                "(1M context window; issue #1158)"
-            )
+            # An explicit pass-through --model outranks ANTHROPIC_MODEL in Claude
+            # Code, so add the suffix there too or the env var is silently
+            # shadowed and the window stays 200k (#2915). Report what will
+            # actually take effect rather than the shadowed env value.
+            claude_args, _model_flag_1m = _apply_1m_to_claude_args(claude_args)
+            if _model_flag_1m is not None:
+                click.echo(f"  --model {_model_flag_1m} (1M context window; issue #1158)")
+            else:
+                click.echo(
+                    f"  {_ANTHROPIC_MODEL_ENV}={env[_ANTHROPIC_MODEL_ENV]} "
+                    "(1M context window; issue #1158)"
+                )
 
         result = subprocess.run([claude_bin, *claude_args], env=env)
         raise SystemExit(result.returncode)

tests/test_cli/test_wrap_claude_vertex_proxy_env.py

@@ -160,6 +160,46 @@ def test_wrap_claude_sibling_note_accurate_under_1m_and_tool_search_optouts(
     assert "kept on" not in output
 
 
+def test_wrap_claude_1m_adds_suffix_to_passthrough_model_flag(
+    runner: CliRunner, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    # #2915: Claude Code's --model CLI flag outranks ANTHROPIC_MODEL, so with
+    # both --1m and an explicit --model the env-var [1m] suffix is shadowed and
+    # the window silently caps at 200k. The wrapper must add the suffix to the
+    # pass-through flag so the 1M window actually activates.
+    captured, output = _invoke_wrap_claude(
+        runner,
+        monkeypatch,
+        env={},
+        extra_args=("--1m", "--model", "opusplan"),
+    )
+    assert captured["child_cmd"] == ["/usr/bin/claude", "--model", "opusplan[1m]"]
+    # The banner reports what actually takes effect, not the shadowed env value.
+    assert "--model opusplan[1m]" in output
+
+
+def test_wrap_claude_1m_adds_suffix_to_equals_model_flag(
+    runner: CliRunner, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    captured, _output = _invoke_wrap_claude(
+        runner,
+        monkeypatch,
+        env={},
+        extra_args=("--1m", "--model=opusplan"),
+    )
+    assert captured["child_cmd"] == ["/usr/bin/claude", "--model=opusplan[1m]"]
+
+
+def test_wrap_claude_1m_without_model_flag_still_uses_env(
+    runner: CliRunner, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    # No pass-through --model: ANTHROPIC_MODEL carries the suffix as before, and
+    # the launched command is untouched.
+    captured, _output = _invoke_wrap_claude(runner, monkeypatch, env={}, extra_args=("--1m",))
+    assert captured["child_cmd"] == ["/usr/bin/claude"]
+    assert captured["child_env"]["ANTHROPIC_MODEL"].endswith("[1m]")
+
+
 def test_wrap_claude_tool_search_banner_line_still_accurate_when_active(
     runner: CliRunner, monkeypatch: pytest.MonkeyPatch
 ) -> None:

tests/test_cli/test_wrap_helpers.py

@@ -109,6 +109,35 @@ def test_wrap_claude_allows_claude_print_short_flag_in_passthrough_args() -> Non
     assert result.exit_code == 0, result.output
 
 
+# ---------------------------------------------------------------------------
+# _apply_1m_to_claude_args — add the [1m] suffix to an explicit pass-through
+# --model so it survives Claude Code's CLI-over-env precedence (#2915).
+# ---------------------------------------------------------------------------
+def test_apply_1m_rewrites_model_flag_value() -> None:
+    args, rewritten = wrap_mod._apply_1m_to_claude_args(("--model", "opusplan"))
+    assert args == ("--model", "opusplan[1m]")
+    assert rewritten == "opusplan[1m]"
+
+
+def test_apply_1m_rewrites_equals_model_flag() -> None:
+    args, rewritten = wrap_mod._apply_1m_to_claude_args(("--model=opusplan",))
+    assert args == ("--model=opusplan[1m]",)
+    assert rewritten == "opusplan[1m]"
+
+
+def test_apply_1m_is_idempotent_on_already_suffixed_model() -> None:
+    args, rewritten = wrap_mod._apply_1m_to_claude_args(("--model", "opusplan[1m]"))
+    assert args == ("--model", "opusplan[1m]")
+    assert rewritten == "opusplan[1m]"
+
+
+def test_apply_1m_noop_without_model_flag() -> None:
+    original = ("--permission-mode", "auto", "--resume")
+    args, rewritten = wrap_mod._apply_1m_to_claude_args(original)
+    assert args == original
+    assert rewritten is None
+
+
 # ---------------------------------------------------------------------------
 # _run_proxy_only_watcher — must print banner, call setup callback, install
 # signal handlers, and clean up. Heavily mocked since the real watcher