feat(codex): pick, clear and auto-assign each LazyCodex role's reasoning effort

docs-site/src/content/docs/guides/integrations.md

@@ -683,14 +683,16 @@ role's model is actually decided. LazyCodex counts as installed when the `omo@si
 Codex plugin is enabled in `$CODEX_HOME/config.toml` and an installed copy under
 `$CODEX_HOME/plugins/cache/sisyphuslabs/omo/` carries its `lazycodex-install.json`. A `~/.omo`
 folder on its own does not count, because Pi-based omo creates it too. Without LazyCodex the
-section is hidden and the command line reports it as not installed. Pick a model on a row and
-press Save:
-
-- opencodex rewrites only the root `model = "..."` line of that role's file. The role's
-  instructions, comments, and other keys are left exactly as they were. A role with no pin gets
-  one added near the top of the file.
-- The same value is written to `[codex].agents.<role>.model` in `~/.omo/omo.jsonc`, which
-  LazyCodex 5.1.1 and later reads. If that file does not exist it is not created. If it contains
+section is hidden and the command line reports it as not installed. Pick a model and a reasoning
+effort on a row and press Save:
+
+- opencodex updates only the root `model = "..."` and `model_reasoning_effort = "..."` keys of that role's file.
+  Missing pins are inserted near the top. Choosing **Inherit effort**, available on every row,
+  sends `"effort": null` to clear the effort pin, removing its line and trailing comment.
+  The role's instructions, other comments, and other keys are left exactly as they were.
+- The model is written to `[codex].agents.<role>.model` in `~/.omo/omo.jsonc`, which
+  LazyCodex 5.1.1 and later reads, and a changed effort LazyCodex supports goes to
+  `[codex].agents.<role>.reasoning` in the same entry. If that file does not exist it is not created. If it contains
   comments it is left untouched, because saving would remove them; the tab says so, and you can
   set the value there by hand. Symlinks and non-regular files are rejected; on macOS and Linux,
   a FIFO is rejected without waiting for a writer. A skipped mirror does not undo the role-file save.
@@ -704,9 +706,15 @@ ocx agent roles set explorer xai/grok-4.5
 ```
 
 `ocx agent roles set <role> <model> --effort <level>` also sets the role's reasoning effort: it
-rewrites the role file's `model_reasoning_effort` line and writes `[codex].agents.<role>.reasoning`
+inserts or replaces the role file's `model_reasoning_effort` value and writes `[codex].agents.<role>.reasoning`
 in omo.jsonc when LazyCodex has that level (`ultra` stays in the role file only, and an older
-`reasoning` is removed). A Save that changes only the model leaves both effort values as they were.
+`reasoning` is removed). On `PUT /api/codex-agent-roles/{role}`, omitting `effort` keeps the pin,
+a level sets it, and `null` removes the role file's `model_reasoning_effort` line, its trailing
+comment, and the omo.jsonc `reasoning`. A Save that changes only the model leaves both effort values as they were.
+Only an existing one-line string value can be removed; a multiline or non-string value is refused
+with 409 `unsupported_model_value`, leaving both files unchanged. Clearing removes only the
+role's own override: Codex then resolves the effort as it does for any role without one (an explicit
+spawn value, the `[agents]` default, or the parent session), not a fixed model default.
 
 ### Auto-assign
 
@@ -729,9 +737,11 @@ picks a model. opencodex then picks the cheapest model from your picker list tha
 Each proposal shows the model, the tier, the reasoning effort, a one-line reason, and what would move
 it up or down. A role the model could not size clearly is shown as not sized, with the reason, and
 cannot be applied. Nothing is written until you press Apply on a row or Apply all. Applying uses the
-same save as picking by hand, and also rewrites the role's `model_reasoning_effort` when the file already has
-one. The effort is placed on the chosen model's own levels: its lowest, its default, one above the
-default, or its highest.
+same save as picking by hand. Every sized role whose chosen model has a non-empty effort ladder gets
+an effort proposal, and Apply or `--apply` inserts or replaces its `model_reasoning_effort` value.
+The effort is placed on the chosen model's own levels: its lowest, its default, one above the
+default, or its highest. An empty ladder gives no effort proposal (`proposedEffort: null`);
+Apply omits `effort` and leaves any existing effort pin as it was. A null proposal never clears a pin.
 
 ```bash
 ocx agent roles suggest

gui/src/i18n/de.ts

@@ -1761,11 +1761,14 @@ export const de: Record<TKey, string> = {
   "integrations.semantics.raycast": "Fügt einen OpenCodex-Provider-Eintrag in die providers.yaml von Raycast ein, damit jedes geroutete Modell in der Modellauswahl von Raycast AI erscheint. Raycast Pro erforderlich.",
   "integrations.semantics.omo": "Verwaltet ausschließlich providers.opencodex in der models.json von omo — ~/.omo/agent, sofern nicht OMO_CODING_AGENT_DIR, SENPI_CODING_AGENT_DIR oder PI_CODING_AGENT_DIR sie umleitet. Ihre übrigen Provider bleiben unverändert. Gilt ab neuen Sitzungen.",
   "integrations.lazycodexRoles.title": "omo (Codex / LazyCodex) · Modelle der Codex-Agentenrollen",
-  "integrations.lazycodexRoles.hint": "Wählen Sie das Modell, mit dem Codex jede Agentenrolle ausführt. Beim Speichern ändert sich nur die model-Zeile in der Datei dieser Rolle unter $CODEX_HOME/agents. In omo.jsonc wird die Auswahl für LazyCodex nur übernommen, wenn sich die Datei sicher neu schreiben lässt.",
+  "integrations.lazycodexRoles.hint": "Wählen Sie Modell und Reasoning-Aufwand, mit denen Codex jede Agentenrolle ausführt. Beim Speichern ändern sich nur die model- und effort-Zeilen in der Datei dieser Rolle unter $CODEX_HOME/agents. In omo.jsonc wird die Auswahl für LazyCodex nur übernommen, wenn sich die Datei sicher neu schreiben lässt.",
   "integrations.lazycodexRoles.role": "Rolle",
   "integrations.lazycodexRoles.current": "Aktuelles Modell",
   "integrations.lazycodexRoles.model": "Neues Modell",
   "integrations.lazycodexRoles.modelFor": "Modell für {role}",
+  "integrations.lazycodexRoles.effort": "Reasoning-Aufwand",
+  "integrations.lazycodexRoles.effortFor": "Reasoning-Aufwand ({role})",
+  "integrations.lazycodexRoles.effortDefault": "Aufwand erben",
   "integrations.lazycodexRoles.none": "Kein Modell festgelegt",
   "integrations.lazycodexRoles.choose": "Modell wählen",
   "integrations.lazycodexRoles.empty": "Keine Codex-Agentenrollen gefunden.",

gui/src/i18n/en.ts

@@ -2372,11 +2372,14 @@ export const en = {
   "integrations.semantics.raycast": "Adds an OpenCodex provider entry to Raycast's providers.yaml so every routed model appears in the Raycast AI model picker. Raycast Pro required.",
   "integrations.semantics.omo": "Manages only providers.opencodex in omo's models.json — ~/.omo/agent unless OMO_CODING_AGENT_DIR, SENPI_CODING_AGENT_DIR or PI_CODING_AGENT_DIR redirects it. Your other providers stay unchanged. Applies to new sessions.",
   "integrations.lazycodexRoles.title": "omo (Codex / LazyCodex) · Codex agent role models",
-  "integrations.lazycodexRoles.hint": "Pick the model Codex runs for each agent role. Saving changes only the model line in that role's file under $CODEX_HOME/agents, and mirrors the choice into omo.jsonc for LazyCodex only when that file can be safely rewritten.",
+  "integrations.lazycodexRoles.hint": "Pick the model and reasoning effort Codex runs for each agent role. Saving changes only the model and effort lines in that role's file under $CODEX_HOME/agents, and mirrors the choice into omo.jsonc for LazyCodex only when that file can be safely rewritten.",
   "integrations.lazycodexRoles.role": "Role",
   "integrations.lazycodexRoles.current": "Current model",
   "integrations.lazycodexRoles.model": "New model",
   "integrations.lazycodexRoles.modelFor": "Model for {role}",
+  "integrations.lazycodexRoles.effort": "Reasoning effort",
+  "integrations.lazycodexRoles.effortFor": "Reasoning effort ({role})",
+  "integrations.lazycodexRoles.effortDefault": "Inherit effort",
   "integrations.lazycodexRoles.none": "No model pin",
   "integrations.lazycodexRoles.choose": "Choose a model",
   "integrations.lazycodexRoles.empty": "No Codex agent roles were found.",

gui/src/i18n/fr.ts

@@ -2299,11 +2299,14 @@ export const fr: Record<TKey, string> = {
   "integrations.semantics.raycast": "Ajoute une entrée de fournisseur OpenCodex dans le providers.yaml de Raycast afin que chaque modèle routé apparaisse dans le sélecteur de modèles de Raycast AI. Raycast Pro requis.",
   "integrations.semantics.omo": "Gère uniquement providers.opencodex dans le models.json d'omo — ~/.omo/agent, sauf redirection par OMO_CODING_AGENT_DIR, SENPI_CODING_AGENT_DIR ou PI_CODING_AGENT_DIR. Vos autres fournisseurs restent inchangés. S'applique aux nouvelles sessions.",
   "integrations.lazycodexRoles.title": "omo (Codex / LazyCodex) · Modèles des rôles d’agent Codex",
-  "integrations.lazycodexRoles.hint": "Choisissez le modèle que Codex utilise pour chaque rôle d’agent. L’enregistrement ne modifie que la ligne model du fichier de ce rôle dans $CODEX_HOME/agents, et ne reporte le choix dans omo.jsonc pour LazyCodex que si ce fichier peut être réécrit sans risque.",
+  "integrations.lazycodexRoles.hint": "Choisissez le modèle et l’effort de raisonnement que Codex utilise pour chaque rôle d’agent. L’enregistrement ne modifie que les lignes model et effort du fichier de ce rôle dans $CODEX_HOME/agents, et ne reporte le choix dans omo.jsonc pour LazyCodex que si ce fichier peut être réécrit sans risque.",
   "integrations.lazycodexRoles.role": "Rôle",
   "integrations.lazycodexRoles.current": "Modèle actuel",
   "integrations.lazycodexRoles.model": "Nouveau modèle",
   "integrations.lazycodexRoles.modelFor": "Modèle pour {role}",
+  "integrations.lazycodexRoles.effort": "Effort de raisonnement",
+  "integrations.lazycodexRoles.effortFor": "Effort de raisonnement ({role})",
+  "integrations.lazycodexRoles.effortDefault": "Hériter de l’effort",
   "integrations.lazycodexRoles.none": "Aucun modèle épinglé",
   "integrations.lazycodexRoles.choose": "Choisir un modèle",
   "integrations.lazycodexRoles.empty": "Aucun rôle d’agent Codex trouvé.",

gui/src/i18n/ja.ts

@@ -2194,11 +2194,14 @@ export const ja: Record<TKey, string> = {
   "integrations.semantics.raycast": "Raycast の providers.yaml に OpenCodex のプロバイダーエントリを追加し、ルーティングされたすべてのモデルを Raycast AI のモデル選択に表示します。Raycast Pro が必要です。",
   "integrations.semantics.omo": "omo の models.json にある providers.opencodex のみを管理します。場所は ~/.omo/agent で、OMO_CODING_AGENT_DIR・SENPI_CODING_AGENT_DIR・PI_CODING_AGENT_DIR のいずれかが設定されている場合はそちらが優先されます。他のプロバイダーは変更しません。新しいセッションから適用されます。",
   "integrations.lazycodexRoles.title": "omo (Codex / LazyCodex) · Codex エージェントロールのモデル",
-  "integrations.lazycodexRoles.hint": "Codex が各エージェントロールで使うモデルを選びます。保存すると $CODEX_HOME/agents にあるそのロールのファイルの model 行だけが変わり、omo.jsonc を安全に書き換えられる場合に限り LazyCodex 用に同じ値を反映します。",
+  "integrations.lazycodexRoles.hint": "Codex が各エージェントロールで使うモデルと推論負荷を選びます。保存すると $CODEX_HOME/agents にあるそのロールのファイルの model 行と effort 行だけが変わり、omo.jsonc を安全に書き換えられる場合に限り LazyCodex 用に同じ値を反映します。",
   "integrations.lazycodexRoles.role": "ロール",
   "integrations.lazycodexRoles.current": "現在のモデル",
   "integrations.lazycodexRoles.model": "新しいモデル",
   "integrations.lazycodexRoles.modelFor": "{role} のモデル",
+  "integrations.lazycodexRoles.effort": "推論負荷",
+  "integrations.lazycodexRoles.effortFor": "推論負荷 ({role})",
+  "integrations.lazycodexRoles.effortDefault": "推論エフォートを継承",
   "integrations.lazycodexRoles.none": "モデル指定なし",
   "integrations.lazycodexRoles.choose": "モデルを選択",
   "integrations.lazycodexRoles.empty": "Codex エージェントロールが見つかりません。",

gui/src/i18n/ko.ts

@@ -1805,11 +1805,14 @@ export const ko: Record<TKey, string> = {
   "integrations.semantics.raycast": "Raycast의 providers.yaml에 OpenCodex 프로바이더 항목을 추가해 라우팅된 모든 모델이 Raycast AI 모델 선택기에 표시되도록 합니다. Raycast Pro가 필요합니다.",
   "integrations.semantics.omo": "omo의 models.json에서 providers.opencodex만 관리합니다. 위치는 ~/.omo/agent이며 OMO_CODING_AGENT_DIR, SENPI_CODING_AGENT_DIR, PI_CODING_AGENT_DIR 중 설정된 값이 있으면 그쪽이 우선합니다. 다른 프로바이더는 그대로 유지됩니다. 새 세션부터 적용됩니다.",
   "integrations.lazycodexRoles.title": "omo (Codex / LazyCodex) · Codex 에이전트 역할 모델",
-  "integrations.lazycodexRoles.hint": "Codex가 에이전트 역할마다 쓰는 모델을 고릅니다. 저장하면 $CODEX_HOME/agents에 있는 그 역할 파일의 model 줄만 바뀌고, omo.jsonc를 안전하게 다시 쓸 수 있을 때만 LazyCodex용으로 같은 값을 반영합니다.",
+  "integrations.lazycodexRoles.hint": "Codex가 에이전트 역할마다 쓰는 모델과 추론 강도를 고릅니다. 저장하면 $CODEX_HOME/agents에 있는 그 역할 파일의 model 줄과 effort 줄만 바뀌고, omo.jsonc를 안전하게 다시 쓸 수 있을 때만 LazyCodex용으로 같은 값을 반영합니다.",
   "integrations.lazycodexRoles.role": "역할",
   "integrations.lazycodexRoles.current": "현재 모델",
   "integrations.lazycodexRoles.model": "새 모델",
   "integrations.lazycodexRoles.modelFor": "{role} 모델",
+  "integrations.lazycodexRoles.effort": "추론 강도",
+  "integrations.lazycodexRoles.effortFor": "추론 강도 ({role})",
+  "integrations.lazycodexRoles.effortDefault": "추론 강도 상속",
   "integrations.lazycodexRoles.none": "지정된 모델 없음",
   "integrations.lazycodexRoles.choose": "모델 선택",
   "integrations.lazycodexRoles.empty": "Codex 에이전트 역할을 찾지 못했습니다.",

gui/src/i18n/pt.ts

@@ -2345,11 +2345,14 @@ export const pt: Record<TKey, string> = {
   "integrations.semantics.raycast": "Adiciona uma entrada de provedor OpenCodex ao providers.yaml do Raycast, para que todos os modelos roteados apareçam no seletor de modelos do Raycast AI. Requer Raycast Pro.",
   "integrations.semantics.omo": "Gerencia apenas providers.opencodex no models.json do omo — ~/.omo/agent, a menos que OMO_CODING_AGENT_DIR, SENPI_CODING_AGENT_DIR ou PI_CODING_AGENT_DIR o redirecione. Seus outros provedores permanecem inalterados. Vale para novas sessões.",
   "integrations.lazycodexRoles.title": "omo (Codex / LazyCodex) · Modelos das funções de agente do Codex",
-  "integrations.lazycodexRoles.hint": "Escolha o modelo que o Codex executa em cada função de agente. Ao salvar, só a linha do modelo no arquivo da função em $CODEX_HOME/agents é alterada, e a escolha é espelhada em omo.jsonc para o LazyCodex apenas quando esse arquivo puder ser reescrito com segurança.",
+  "integrations.lazycodexRoles.hint": "Escolha o modelo e o esforço de raciocínio que o Codex usa em cada função de agente. Ao salvar, só as linhas de modelo e esforço no arquivo da função em $CODEX_HOME/agents são alteradas, e a escolha é espelhada em omo.jsonc para o LazyCodex apenas quando esse arquivo puder ser reescrito com segurança.",
   "integrations.lazycodexRoles.role": "Função",
   "integrations.lazycodexRoles.current": "Modelo atual",
   "integrations.lazycodexRoles.model": "Novo modelo",
   "integrations.lazycodexRoles.modelFor": "Modelo para {role}",
+  "integrations.lazycodexRoles.effort": "Esforço de raciocínio",
+  "integrations.lazycodexRoles.effortFor": "Esforço de raciocínio ({role})",
+  "integrations.lazycodexRoles.effortDefault": "Herdar esforço",
   "integrations.lazycodexRoles.none": "Nenhum modelo fixado",
   "integrations.lazycodexRoles.choose": "Escolha um modelo",
   "integrations.lazycodexRoles.empty": "Nenhuma função de agente do Codex foi encontrada.",

gui/src/i18n/ru.ts

@@ -2285,11 +2285,14 @@ export const ru: Record<TKey, string> = {
   "integrations.semantics.raycast": "Добавляет запись провайдера OpenCodex в providers.yaml Raycast, чтобы каждая маршрутизируемая модель появилась в выборе моделей Raycast AI. Требуется Raycast Pro.",
   "integrations.semantics.omo": "Управляет только providers.opencodex в models.json omo — ~/.omo/agent, если только OMO_CODING_AGENT_DIR, SENPI_CODING_AGENT_DIR или PI_CODING_AGENT_DIR не перенаправляет путь. Остальные провайдеры остаются без изменений. Применяется к новым сессиям.",
   "integrations.lazycodexRoles.title": "omo (Codex / LazyCodex) · Модели ролей агентов Codex",
-  "integrations.lazycodexRoles.hint": "Выберите модель, на которой Codex запускает каждую роль агента. При сохранении меняется только строка model в файле этой роли в $CODEX_HOME/agents, а в omo.jsonc для LazyCodex выбор копируется, только если этот файл можно безопасно перезаписать.",
+  "integrations.lazycodexRoles.hint": "Выберите модель и уровень рассуждений, с которыми Codex запускает каждую роль агента. При сохранении меняются только строки model и effort в файле этой роли в $CODEX_HOME/agents, а в omo.jsonc для LazyCodex выбор копируется, только если этот файл можно безопасно перезаписать.",
   "integrations.lazycodexRoles.role": "Роль",
   "integrations.lazycodexRoles.current": "Текущая модель",
   "integrations.lazycodexRoles.model": "Новая модель",
   "integrations.lazycodexRoles.modelFor": "Модель для {role}",
+  "integrations.lazycodexRoles.effort": "Уровень рассуждений",
+  "integrations.lazycodexRoles.effortFor": "Уровень рассуждений ({role})",
+  "integrations.lazycodexRoles.effortDefault": "Наследовать усилие",
   "integrations.lazycodexRoles.none": "Модель не закреплена",
   "integrations.lazycodexRoles.choose": "Выберите модель",
   "integrations.lazycodexRoles.empty": "Роли агентов Codex не найдены.",