fix(provider): implement chat_with_history for OpenAI Codex and Gemini

Both providers only implemented chat_with_system, so the default
chat_with_history trait method was discarding all conversation history
except the last user message. This caused the Telegram bot to lose
context between messages.

Changes:
- OpenAiCodexProvider: extract send_responses_request helper, add
  chat_with_history that maps full ChatMessage history to ResponsesInput
- GeminiProvider: extract send_generate_content helper, add
  chat_with_history that maps ChatMessage history to Gemini Content
  (with assistant→model role mapping)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

src/providers/gemini.rs

@@ -3,7 +3,7 @@
 //! - Gemini CLI OAuth tokens (reuse existing ~/.gemini/ authentication)
 //! - Google Cloud ADC (`GOOGLE_APPLICATION_CREDENTIALS`)
 
-use crate::providers::traits::Provider;
+use crate::providers::traits::{ChatMessage, Provider};
 use async_trait::async_trait;
 use directories::UserDirs;
 use reqwest::Client;
@@ -326,12 +326,11 @@ impl GeminiProvider {
     }
 }
 
-#[async_trait]
-impl Provider for GeminiProvider {
-    async fn chat_with_system(
+impl GeminiProvider {
+    async fn send_generate_content(
         &self,
-        system_prompt: Option<&str>,
-        message: &str,
+        contents: Vec<Content>,
+        system_instruction: Option<Content>,
         model: &str,
         temperature: f64,
     ) -> anyhow::Result<String> {
@@ -345,21 +344,8 @@ impl Provider for GeminiProvider {
             )
         })?;
 
-        // Build request
-        let system_instruction = system_prompt.map(|sys| Content {
-            role: None,
-            parts: vec![Part {
-                text: sys.to_string(),
-            }],
-        });
-
         let request = GenerateContentRequest {
-            contents: vec![Content {
-                role: Some("user".to_string()),
-                parts: vec![Part {
-                    text: message.to_string(),
-                }],
-            }],
+            contents,
             system_instruction,
             generation_config: GenerationConfig {
                 temperature,
@@ -382,12 +368,10 @@ impl Provider for GeminiProvider {
 
         let result: GenerateContentResponse = response.json().await?;
 
-        // Check for API error in response body
         if let Some(err) = result.error {
             anyhow::bail!("Gemini API error: {}", err.message);
         }
 
-        // Extract text from response
         result
             .candidates
             .and_then(|c| c.into_iter().next())
@@ -395,6 +379,84 @@ impl Provider for GeminiProvider {
             .and_then(|p| p.text)
             .ok_or_else(|| anyhow::anyhow!("No response from Gemini"))
     }
+}
+
+#[async_trait]
+impl Provider for GeminiProvider {
+    async fn chat_with_system(
+        &self,
+        system_prompt: Option<&str>,
+        message: &str,
+        model: &str,
+        temperature: f64,
+    ) -> anyhow::Result<String> {
+        let system_instruction = system_prompt.map(|sys| Content {
+            role: None,
+            parts: vec![Part {
+                text: sys.to_string(),
+            }],
+        });
+
+        let contents = vec![Content {
+            role: Some("user".to_string()),
+            parts: vec![Part {
+                text: message.to_string(),
+            }],
+        }];
+
+        self.send_generate_content(contents, system_instruction, model, temperature)
+            .await
+    }
+
+    async fn chat_with_history(
+        &self,
+        messages: &[ChatMessage],
+        model: &str,
+        temperature: f64,
+    ) -> anyhow::Result<String> {
+        let mut system_parts: Vec<&str> = Vec::new();
+        let mut contents: Vec<Content> = Vec::new();
+
+        for msg in messages {
+            match msg.role.as_str() {
+                "system" => {
+                    system_parts.push(&msg.content);
+                }
+                "user" => {
+                    contents.push(Content {
+                        role: Some("user".to_string()),
+                        parts: vec![Part {
+                            text: msg.content.clone(),
+                        }],
+                    });
+                }
+                "assistant" => {
+                    // Gemini API uses "model" role instead of "assistant"
+                    contents.push(Content {
+                        role: Some("model".to_string()),
+                        parts: vec![Part {
+                            text: msg.content.clone(),
+                        }],
+                    });
+                }
+                _ => {}
+            }
+        }
+
+        let system_instruction = if system_parts.is_empty() {
+            None
+        } else {
+            Some(Content {
+                role: None,
+                parts: vec![Part {
+                    text: system_parts.join("\n\n"),
+                }],
+            })
+        };
+
+        self.send_generate_content(contents, system_instruction, model, temperature)
+            .await
+    }
 
     async fn warmup(&self) -> anyhow::Result<()> {
         if let Some(auth) = self.auth.as_ref() {

src/providers/openai_codex.rs

@@ -1,6 +1,6 @@
 use crate::auth::openai_oauth::extract_account_id_from_jwt;
 use crate::auth::AuthService;
-use crate::providers::traits::Provider;
+use crate::providers::traits::{ChatMessage, Provider};
 use crate::providers::ProviderRuntimeOptions;
 use async_trait::async_trait;
 use reqwest::Client;
@@ -335,14 +335,12 @@ async fn decode_responses_body(response: reqwest::Response) -> anyhow::Result<St
     extract_responses_text(&parsed).ok_or_else(|| anyhow::anyhow!("No response from OpenAI Codex"))
 }
 
-#[async_trait]
-impl Provider for OpenAiCodexProvider {
-    async fn chat_with_system(
+impl OpenAiCodexProvider {
+    async fn send_responses_request(
         &self,
-        system_prompt: Option<&str>,
-        message: &str,
+        input: Vec<ResponsesInput>,
+        instructions: String,
         model: &str,
-        _temperature: f64,
     ) -> anyhow::Result<String> {
         let profile = self
             .auth
@@ -368,14 +366,8 @@ impl Provider for OpenAiCodexProvider {
 
         let request = ResponsesRequest {
             model: normalized_model.to_string(),
-            input: vec![ResponsesInput {
-                role: "user".to_string(),
-                content: vec![ResponsesInputContent {
-                    kind: "input_text".to_string(),
-                    text: message.to_string(),
-                }],
-            }],
-            instructions: resolve_instructions(system_prompt),
+            input,
+            instructions,
             store: false,
             stream: true,
             text: ResponsesTextOptions {
@@ -411,6 +403,64 @@ impl Provider for OpenAiCodexProvider {
     }
 }
 
+#[async_trait]
+impl Provider for OpenAiCodexProvider {
+    async fn chat_with_system(
+        &self,
+        system_prompt: Option<&str>,
+        message: &str,
+        model: &str,
+        _temperature: f64,
+    ) -> anyhow::Result<String> {
+        let input = vec![ResponsesInput {
+            role: "user".to_string(),
+            content: vec![ResponsesInputContent {
+                kind: "input_text".to_string(),
+                text: message.to_string(),
+            }],
+        }];
+        self.send_responses_request(input, resolve_instructions(system_prompt), model)
+            .await
+    }
+
+    async fn chat_with_history(
+        &self,
+        messages: &[ChatMessage],
+        model: &str,
+        _temperature: f64,
+    ) -> anyhow::Result<String> {
+        let mut system_parts: Vec<&str> = Vec::new();
+        let mut input: Vec<ResponsesInput> = Vec::new();
+
+        for msg in messages {
+            match msg.role.as_str() {
+                "system" => {
+                    system_parts.push(&msg.content);
+                }
+                "user" | "assistant" => {
+                    input.push(ResponsesInput {
+                        role: msg.role.clone(),
+                        content: vec![ResponsesInputContent {
+                            kind: "input_text".to_string(),
+                            text: msg.content.clone(),
+                        }],
+                    });
+                }
+                _ => {}
+            }
+        }
+
+        let instructions = if system_parts.is_empty() {
+            DEFAULT_CODEX_INSTRUCTIONS.to_string()
+        } else {
+            system_parts.join("\n\n")
+        };
+
+        self.send_responses_request(input, instructions, model)
+            .await
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;