fix: project model and reasoning_effort into Codex CLI args, stop rewriting .codex/config.toml

- Add model_reasoning_effort field to AgentProfile (domain layer)
- Remove model parameter from codex_config_toml, stop rewriting model in TOML
- Pass model and model_reasoning_effort via Codex CLI -c overrides on every exec
- Update CodexExecSession with new_with_policy_and_overrides factory
- Cover new conversation and resume flows with tests

crates/domain/src/permission.rs

@@ -851,7 +851,7 @@ pub struct ProjectionContext<'a> {
     /// Absolute isolated run dir of the agent (`.ideai/run/<agent-id>/`).
     pub run_dir: &'a str,
     /// Optional model selected by the agent profile. Orthogonal to permissions:
-    /// projectors may still materialise it even when `eff == None`.
+    /// projectors may still materialise or forward it when appropriate.
     pub model: Option<&'a str>,
 }
 

crates/domain/src/profile.rs

@@ -910,6 +910,10 @@ pub struct AgentProfile {
     /// CLI. OpenCode garde ses champs dédiés.
     #[serde(default, skip_serializing_if = "Option::is_none")]
     pub model: Option<String>,
+    /// Effort de raisonnement Codex explicitement configuré par le profil. `None`
+    /// conserve le défaut natif de la CLI ; seules les sessions Codex le consomment.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub model_reasoning_effort: Option<String>,
     /// Capacité **MCP** (ARCHITECTURE §14.3, orchestration v3, Décision 1).
     /// `None` ⇒ repli fichier `.ideai/requests` + prose (comportement actuel).
     /// `Some(_)` ⇒ IdeA matérialise la config MCP de cette CLI au lancement et
@@ -1117,6 +1121,7 @@ impl AgentProfile {
             opencode: None,
             opencode_provider: None,
             model: None,
+            model_reasoning_effort: None,
             mcp: None,
             liveness: None,
             rate_limit_pattern: None,
@@ -1174,6 +1179,13 @@ impl AgentProfile {
         self
     }
 
+    /// Builder : fixe l'effort de raisonnement Codex direct (ticket #99 follow-up).
+    #[must_use]
+    pub fn with_model_reasoning_effort(mut self, effort: impl Into<String>) -> Self {
+        self.model_reasoning_effort = Some(effort.into());
+        self
+    }
+
     /// Builder : fixe la [`McpCapability`] (§14.3, orchestration v3) et renvoie le
     /// profil. Laisse [`AgentProfile::new`] stable (zéro régression d'appel) : les
     /// profils sans MCP ne l'appellent simplement pas.
@@ -1488,10 +1500,15 @@ mod mcp_tests {
     fn profile_model_round_trips_without_codex_or_claude_provider_config() {
         let profile = profile_without_mcp()
             .with_structured_adapter(StructuredAdapter::Codex)
-            .with_model("gpt-5-codex");
+            .with_model("gpt-5-codex")
+            .with_model_reasoning_effort("medium");
 
         let json = serde_json::to_string(&profile).expect("serialise");
         assert!(json.contains("\"model\":\"gpt-5-codex\""), "got: {json}");
+        assert!(
+            json.contains("\"modelReasoningEffort\":\"medium\""),
+            "got: {json}"
+        );
         assert!(!json.contains("codexProvider"), "got: {json}");
         assert!(!json.contains("claudeProvider"), "got: {json}");
         assert!(!json.contains("apiKeyRef"), "got: {json}");
@@ -1499,6 +1516,7 @@ mod mcp_tests {
 
         let back: AgentProfile = serde_json::from_str(&json).expect("deserialise");
         assert_eq!(back.model.as_deref(), Some("gpt-5-codex"));
+        assert_eq!(back.model_reasoning_effort.as_deref(), Some("medium"));
     }
 
     #[test]