chore(provider): increase model cache TTL to one week

crates/forge_app/src/orch_spec/orch_setup.rs

@@ -95,6 +95,7 @@ impl Default for TestContext {
                 workspace_server_url: Url::parse("http://localhost:8080").unwrap(),
                 max_extensions: 15,
                 parallel_file_reads: 64,
+                model_cache_ttl: 604_800,
             },
             title: Some("test-conversation".into()),
             agent: Agent::new(

crates/forge_domain/src/env.rs

@@ -100,6 +100,9 @@ pub struct Environment {
     /// Controlled by FORGE_PARALLEL_FILE_READS environment variable.
     /// Caps the `buffer_unordered` concurrency to avoid EMFILE errors.
     pub parallel_file_reads: usize,
+    /// TTL in seconds for the model API list cache.
+    /// Controlled by FORGE_MODEL_CACHE_TTL environment variable.
+    pub model_cache_ttl: u64,
 }
 
 /// The output format used when auto-dumping a conversation on task completion.
@@ -339,6 +342,7 @@ fn test_command_path() {
         max_extensions: 15,
         auto_dump: None,
         parallel_file_reads: 64,
+        model_cache_ttl: 604_800,
     };
 
     let actual = fixture.command_path();
@@ -381,6 +385,7 @@ fn test_command_cwd_path() {
         max_extensions: 15,
         auto_dump: None,
         parallel_file_reads: 64,
+        model_cache_ttl: 604_800,
     };
 
     let actual = fixture.command_cwd_path();
@@ -423,6 +428,7 @@ fn test_command_cwd_path_independent_from_command_path() {
         max_extensions: 15,
         auto_dump: None,
         parallel_file_reads: 64,
+        model_cache_ttl: 604_800,
     };
 
     let command_path = fixture.command_path();

crates/forge_infra/src/env.rs

@@ -104,6 +104,7 @@ impl ForgeEnvironmentInfra {
                         .unwrap_or(32)
                 },
             ),
+            model_cache_ttl: parse_env::<u64>("FORGE_MODEL_CACHE_TTL").unwrap_or(604_800), /* 1 week */
         }
     }
 

crates/forge_repo/src/provider/chat.rs

@@ -15,9 +15,6 @@ use crate::provider::google::GoogleResponseRepository;
 use crate::provider::openai::OpenAIResponseRepository;
 use crate::provider::openai_responses::OpenAIResponsesResponseRepository;
 
-/// TTL for cached model lists: 2 hours in seconds.
-const MODEL_CACHE_TTL_SECS: u128 = 7200;
-
 /// Repository responsible for routing chat requests to the appropriate provider
 /// implementation based on the provider's response type.
 pub struct ForgeChatRepository<F> {
@@ -48,7 +45,7 @@ impl<F: EnvironmentInfra + HttpInfra> ForgeChatRepository<F> {
 
         let model_cache = Arc::new(CacacheStorage::new(
             env.cache_dir().join("model_cache"),
-            Some(MODEL_CACHE_TTL_SECS),
+            Some(env.model_cache_ttl as u128),
         ));
 
         Self {