diff --git a/src/brigade/model_inventory.py b/src/brigade/model_inventory.py
index 3617af65..e3f7dc79 100644
--- a/src/brigade/model_inventory.py
+++ b/src/brigade/model_inventory.py
@@ -3,6 +3,8 @@
 from __future__ import annotations
 
 import re
+import subprocess
+import tempfile
 from dataclasses import dataclass
 from typing import Literal
 
@@ -15,6 +17,8 @@
     "grok": ["grok", "models"],
     "ollama": ["ollama", "list"],
 }
+_INVENTORY_TIMEOUT_SECONDS = 15.0
+_CURSOR_COMPLETION_PREFIX = "Tip: use --model <id>"
 _EFFORT_SUFFIX = re.compile(r"-(?:none|low|medium|high|xhigh|extra-high|max)$")
 _OLLAMA_OPERATIONAL_ERRORS = (
     "authentication",
@@ -107,12 +111,17 @@ def _inventory(self, cli_ref: str) -> _HarnessInventory:
         cached = self._inventories.get(cli_ref)
         if cached is not None:
             return cached
-        result = proc.run(_LIST_COMMANDS[cli_ref], timeout=15.0)
+        result = (
+            _run_cursor_inventory()
+            if cli_ref == "cursor"
+            else proc.run(_LIST_COMMANDS[cli_ref], timeout=_INVENTORY_TIMEOUT_SECONDS)
+        )
         if result.code != 0:
             diagnostic = result.stderr.strip() or result.stdout.strip() or f"exit {result.code}"
             inventory = _HarnessInventory(error=diagnostic[:160])
         else:
-            recognized, models = _parse_model_list(cli_ref, f"{result.stdout}\n{result.stderr}")
+            output = result.stdout if cli_ref == "cursor" else f"{result.stdout}\n{result.stderr}"
+            recognized, models = _parse_model_list(cli_ref, output)
             if not recognized:
                 inventory = _HarnessInventory(error="command returned an unrecognized inventory shape")
             elif not models and cli_ref != "ollama":
@@ -161,17 +170,61 @@ def _inspect_ollama(self, requested: str) -> ModelInventoryResult:
         return ModelInventoryResult(state, requested, (), diagnostic[:200])
 
 
+def _run_cursor_inventory() -> proc.Result:
+    """Capture Cursor inventory through a file because its pipe output truncates at 8 KiB."""
+    with tempfile.TemporaryFile() as stdout:
+        try:
+            completed = subprocess.run(
+                _LIST_COMMANDS["cursor"],
+                stdout=stdout,
+                stderr=subprocess.PIPE,
+                stdin=subprocess.DEVNULL,
+                timeout=_INVENTORY_TIMEOUT_SECONDS,
+                check=False,
+            )
+        except subprocess.TimeoutExpired as exc:
+            stdout.seek(0)
+            output = stdout.read().decode("utf-8", errors="replace")
+            diagnostic = _decode_subprocess_output(exc.stderr)
+            if diagnostic and not diagnostic.endswith("\n"):
+                diagnostic += "\n"
+            return proc.Result(
+                124,
+                output,
+                f"{diagnostic}timeout after {_INVENTORY_TIMEOUT_SECONDS}s",
+            )
+        except OSError as exc:
+            return proc.Result(127 if isinstance(exc, FileNotFoundError) else 126, "", str(exc))
+        stdout.seek(0)
+        output = stdout.read().decode("utf-8", errors="replace")
+    return proc.Result(completed.returncode, output, _decode_subprocess_output(completed.stderr))
+
+
+def _decode_subprocess_output(value: str | bytes | None) -> str:
+    if value is None:
+        return ""
+    if isinstance(value, str):
+        return value
+    return value.decode("utf-8", errors="replace")
+
+
 def _parse_model_list(cli_ref: str, output: str) -> tuple[bool, tuple[str, ...]]:
     lines = output.splitlines()
     header_index = _inventory_header_index(cli_ref, lines)
     if header_index is None:
         return False, ()
     models: set[str] = set()
+    complete = cli_ref != "cursor"
     for line in lines[header_index + 1 :]:
         if cli_ref == "cursor":
-            if line.startswith("Tip:"):
+            if not line.strip():
+                continue
+            if line.startswith(_CURSOR_COMPLETION_PREFIX):
+                complete = True
                 break
             match = re.match(r"^([a-z0-9][a-z0-9._:/\[\],=-]*)\s+-\s+.+$", line)
+            if match is None:
+                return False, ()
         elif cli_ref == "grok":
             match = re.match(r"^\s*\*\s+([^\s(]+)", line)
         else:
@@ -184,7 +237,7 @@ def _parse_model_list(cli_ref: str, output: str) -> tuple[bool, tuple[str, ...]]
             continue
         if match is not None:
             models.add(match.group(1))
-    return True, tuple(sorted(models))
+    return complete, tuple(sorted(models))
 
 
 def _inventory_header_index(cli_ref: str, lines: list[str]) -> int | None:
diff --git a/tests/test_model_inventory.py b/tests/test_model_inventory.py
index 8c954eea..4eaf5353 100644
--- a/tests/test_model_inventory.py
+++ b/tests/test_model_inventory.py
@@ -1,9 +1,12 @@
+import subprocess
+
 from brigade import agents, model_inventory
 
 
 def _cursor_listing(*ids: str) -> str:
     lines = ["Available models", ""]
     lines.extend(f"{model_id} - Label for {model_id}" for model_id in ids)
+    lines.extend(["", "Tip: use --model <id> to switch."])
     return "\n".join(lines) + "\n"
 
 
@@ -19,12 +22,12 @@ def _ollama_listing(*ids: str) -> str:
     return "\n".join(lines) + "\n"
 
 
+def _patch_cursor_inventory(monkeypatch, result) -> None:
+    monkeypatch.setattr(model_inventory, "_run_cursor_inventory", lambda: result, raising=False)
+
+
 def test_cursor_inventory_exact_match(monkeypatch):
-    monkeypatch.setattr(
-        model_inventory.proc,
-        "run",
-        lambda argv, **kwargs: agents.proc.Result(0, _cursor_listing("composer-2.5"), ""),
-    )
+    _patch_cursor_inventory(monkeypatch, agents.proc.Result(0, _cursor_listing("composer-2.5"), ""))
 
     result = model_inventory.ModelInventoryInspector().inspect("cursor", "composer-2.5")
 
@@ -34,10 +37,9 @@ def test_cursor_inventory_exact_match(monkeypatch):
 
 
 def test_cursor_inventory_narrow_fuzzy_match(monkeypatch):
-    monkeypatch.setattr(
-        model_inventory.proc,
-        "run",
-        lambda argv, **kwargs: agents.proc.Result(
+    _patch_cursor_inventory(
+        monkeypatch,
+        agents.proc.Result(
             0,
             _cursor_listing("cursor-grok-4.5-low", "cursor-grok-4.5-high", "cursor-grok-4.6-high"),
             "",
@@ -52,10 +54,9 @@ def test_cursor_inventory_narrow_fuzzy_match(monkeypatch):
 
 
 def test_cursor_inventory_different_version_is_missing(monkeypatch):
-    monkeypatch.setattr(
-        model_inventory.proc,
-        "run",
-        lambda argv, **kwargs: agents.proc.Result(0, _cursor_listing("cursor-grok-4.6-high"), ""),
+    _patch_cursor_inventory(
+        monkeypatch,
+        agents.proc.Result(0, _cursor_listing("cursor-grok-4.6-high"), ""),
     )
 
     result = model_inventory.ModelInventoryInspector().inspect("cursor", "grok-4.5-xhigh")
@@ -66,10 +67,9 @@ def test_cursor_inventory_different_version_is_missing(monkeypatch):
 
 
 def test_cursor_inventory_requires_versioned_family_for_fuzzy_match(monkeypatch):
-    monkeypatch.setattr(
-        model_inventory.proc,
-        "run",
-        lambda argv, **kwargs: agents.proc.Result(0, _cursor_listing("cursor-auto-low"), ""),
+    _patch_cursor_inventory(
+        monkeypatch,
+        agents.proc.Result(0, _cursor_listing("cursor-auto-low"), ""),
     )
 
     result = model_inventory.ModelInventoryInspector().inspect("cursor", "auto-high")
@@ -80,11 +80,7 @@ def test_cursor_inventory_requires_versioned_family_for_fuzzy_match(monkeypatch)
 
 
 def test_cursor_inventory_command_failure_is_unavailable(monkeypatch):
-    monkeypatch.setattr(
-        model_inventory.proc,
-        "run",
-        lambda argv, **kwargs: agents.proc.Result(1, "", "not logged in"),
-    )
+    _patch_cursor_inventory(monkeypatch, agents.proc.Result(1, "", "not logged in"))
 
     result = model_inventory.ModelInventoryInspector().inspect("cursor", "composer-2.5")
 
@@ -94,11 +90,7 @@ def test_cursor_inventory_command_failure_is_unavailable(monkeypatch):
 
 
 def test_cursor_inventory_unrecognized_output_is_unavailable(monkeypatch):
-    monkeypatch.setattr(
-        model_inventory.proc,
-        "run",
-        lambda argv, **kwargs: agents.proc.Result(0, "model output changed\n", ""),
-    )
+    _patch_cursor_inventory(monkeypatch, agents.proc.Result(0, "model output changed\n", ""))
 
     result = model_inventory.ModelInventoryInspector().inspect("cursor", "composer-2.5")
 
@@ -108,11 +100,7 @@ def test_cursor_inventory_unrecognized_output_is_unavailable(monkeypatch):
 
 
 def test_cursor_inventory_error_shaped_success_is_unavailable(monkeypatch):
-    monkeypatch.setattr(
-        model_inventory.proc,
-        "run",
-        lambda argv, **kwargs: agents.proc.Result(0, "Warning - authentication required\n", ""),
-    )
+    _patch_cursor_inventory(monkeypatch, agents.proc.Result(0, "Warning - authentication required\n", ""))
 
     result = model_inventory.ModelInventoryInspector().inspect("cursor", "composer-2.5")
 
@@ -122,11 +110,7 @@ def test_cursor_inventory_error_shaped_success_is_unavailable(monkeypatch):
 
 def test_cursor_inventory_accepts_parameterized_exact_model(monkeypatch):
     base_model = "claude-opus-4-8-thinking-high"
-    monkeypatch.setattr(
-        model_inventory.proc,
-        "run",
-        lambda argv, **kwargs: agents.proc.Result(0, _cursor_listing(base_model), ""),
-    )
+    _patch_cursor_inventory(monkeypatch, agents.proc.Result(0, _cursor_listing(base_model), ""))
 
     result = model_inventory.ModelInventoryInspector().inspect(
         "cursor",
@@ -141,11 +125,11 @@ def test_cursor_inventory_accepts_parameterized_exact_model(monkeypatch):
 def test_cursor_inventory_is_loaded_once_for_repeated_seats(monkeypatch):
     calls = []
 
-    def fake_run(argv, **kwargs):
-        calls.append(argv)
+    def fake_run():
+        calls.append(["cursor-agent", "models"])
         return agents.proc.Result(0, _cursor_listing("composer-2.5", "gpt-5.5-high"), "")
 
-    monkeypatch.setattr(model_inventory.proc, "run", fake_run)
+    monkeypatch.setattr(model_inventory, "_run_cursor_inventory", fake_run, raising=False)
     inspector = model_inventory.ModelInventoryInspector()
 
     assert inspector.inspect("cursor", "composer-2.5").state == "exact"
@@ -153,6 +137,94 @@ def fake_run(argv, **kwargs):
     assert calls == [["cursor-agent", "models"]]
 
 
+def test_cursor_inventory_uses_regular_file_capture(monkeypatch):
+    listing = _cursor_listing("composer-2.5", "kimi-k2.7-code", "glm-5.2-high")
+
+    def fake_run(argv, **kwargs):
+        assert kwargs["stdout"] is not subprocess.PIPE
+        kwargs["stdout"].write(listing.encode())
+        return subprocess.CompletedProcess(argv, 0, stderr=b"")
+
+    monkeypatch.setattr(subprocess, "run", fake_run)
+
+    result = model_inventory._run_cursor_inventory()
+
+    assert result.code == 0
+    assert result.stdout == listing
+
+
+def test_cursor_inventory_is_stable_across_repeated_probes(monkeypatch):
+    calls = []
+    listing = _cursor_listing("composer-2.5", "kimi-k2.7-code", "glm-5.2-high")
+
+    def fake_run(argv, **kwargs):
+        calls.append(argv)
+        assert kwargs["stdout"] is not subprocess.PIPE
+        kwargs["stdout"].write(listing.encode())
+        return subprocess.CompletedProcess(argv, 0, stderr=b"")
+
+    monkeypatch.setattr(subprocess, "run", fake_run)
+
+    for requested in ("kimi-k2.7-code", "glm-5.2-high"):
+        for _ in range(3):
+            result = model_inventory.ModelInventoryInspector().inspect("cursor", requested)
+            assert result is not None
+            assert result.state == "exact"
+
+    assert calls == [["cursor-agent", "models"]] * 6
+
+
+def test_cursor_inventory_truncated_listing_is_unavailable(monkeypatch):
+    _patch_cursor_inventory(
+        monkeypatch,
+        agents.proc.Result(
+            0,
+            "Available models\n\ncomposer-2.5 - Composer 2.5\ngpt-5.",
+            "",
+        ),
+    )
+
+    result = model_inventory.ModelInventoryInspector().inspect("cursor", "kimi-k2.7-code")
+
+    assert result is not None
+    assert result.state == "unavailable"
+    assert "unrecognized inventory shape" in result.detail
+
+
+def test_cursor_inventory_completed_listing_with_malformed_row_is_unavailable(monkeypatch):
+    _patch_cursor_inventory(
+        monkeypatch,
+        agents.proc.Result(
+            0,
+            "Available models\n\ncomposer-2.5 - Composer 2.5\nbroken row\n\nTip: use --model <id> to switch.\n",
+            "",
+        ),
+    )
+
+    result = model_inventory.ModelInventoryInspector().inspect("cursor", "kimi-k2.7-code")
+
+    assert result is not None
+    assert result.state == "unavailable"
+    assert "unrecognized inventory shape" in result.detail
+
+
+def test_cursor_inventory_stderr_cannot_complete_truncated_stdout(monkeypatch):
+    _patch_cursor_inventory(
+        monkeypatch,
+        agents.proc.Result(
+            0,
+            "Available models\n\ncomposer-2.5 - Composer 2.5\ngpt-5.",
+            "Tip: use --model <id> to switch.\n",
+        ),
+    )
+
+    result = model_inventory.ModelInventoryInspector().inspect("cursor", "kimi-k2.7-code")
+
+    assert result is not None
+    assert result.state == "unavailable"
+    assert "unrecognized inventory shape" in result.detail
+
+
 def test_grok_inventory_parses_exact_and_fuzzy_models(monkeypatch):
     monkeypatch.setattr(
         model_inventory.proc,
diff --git a/tests/test_roster_cmd.py b/tests/test_roster_cmd.py
index f0034de8..9e949131 100644
--- a/tests/test_roster_cmd.py
+++ b/tests/test_roster_cmd.py
@@ -390,15 +390,18 @@ def test_roster_doctor_reuses_inventory_for_repeated_harness_seats(monkeypatch,
     monkeypatch.setattr(agents.proc, "which", lambda cmd: "/x/" + cmd)
     calls = []
 
-    def fake_run(argv, **kwargs):
-        calls.append(argv)
+    def fake_run():
+        calls.append(["cursor-agent", "models"])
         return agents.proc.Result(
             0,
-            "Available models\n\ncomposer-2.5 - Composer 2.5\ngpt-5.5-high - GPT-5.5 High\n",
+            "Available models\n\n"
+            "composer-2.5 - Composer 2.5\n"
+            "gpt-5.5-high - GPT-5.5 High\n\n"
+            "Tip: use --model <id> to switch.\n",
             "",
         )
 
-    monkeypatch.setattr(model_inventory.proc, "run", fake_run)
+    monkeypatch.setattr(model_inventory, "_run_cursor_inventory", fake_run)
 
     assert roster_cmd.doctor(tmp_target) == 0
     assert capsys.readouterr().out.count("model inventory") == 2