fix(llm): add stop_sequences parity for tool completions (#1170)

* fix(llm): add stop_sequences parity for tool completions

* refactor(web-openai): dedupe request builders and satisfy no-panics gate

* test(llm): mark multiline assert with safety comment for CI gate

* test(llm): make safety-marked assert formatting-stable

src/channels/web/openai_compat.rs

@@ -419,6 +419,44 @@ fn parse_stop(val: &serde_json::Value) -> Option<Vec<String>> {
     }
 }
 
+fn build_completion_request(
+    req: &OpenAiChatRequest,
+    messages: Vec<ChatMessage>,
+) -> CompletionRequest {
+    let mut comp_req = CompletionRequest::new(messages).with_model(req.model.clone());
+    if let Some(t) = req.temperature {
+        comp_req = comp_req.with_temperature(t);
+    }
+    if let Some(mt) = req.max_tokens {
+        comp_req = comp_req.with_max_tokens(mt);
+    }
+    if let Some(stops) = req.stop.as_ref().and_then(parse_stop) {
+        comp_req.stop_sequences = Some(stops);
+    }
+    comp_req
+}
+
+fn build_tool_request(
+    req: &OpenAiChatRequest,
+    messages: Vec<ChatMessage>,
+) -> ToolCompletionRequest {
+    let tools = convert_tools(req.tools.as_deref().unwrap_or(&[]));
+    let mut tool_req = ToolCompletionRequest::new(messages, tools).with_model(req.model.clone());
+    if let Some(t) = req.temperature {
+        tool_req = tool_req.with_temperature(t);
+    }
+    if let Some(mt) = req.max_tokens {
+        tool_req = tool_req.with_max_tokens(mt);
+    }
+    if let Some(stops) = req.stop.as_ref().and_then(parse_stop) {
+        tool_req = tool_req.with_stop_sequences(stops);
+    }
+    if let Some(choice) = req.tool_choice.as_ref().and_then(normalize_tool_choice) {
+        tool_req = tool_req.with_tool_choice(choice);
+    }
+    tool_req
+}
+
 // ---------------------------------------------------------------------------
 // Handlers
 // ---------------------------------------------------------------------------
@@ -476,19 +514,7 @@ pub async fn chat_completions_handler(
     let created = unix_timestamp();
 
     if has_tools {
-        let tools = convert_tools(req.tools.as_deref().unwrap_or(&[]));
-        let mut tool_req = ToolCompletionRequest::new(messages, tools).with_model(req.model);
-        if let Some(t) = req.temperature {
-            tool_req = tool_req.with_temperature(t);
-        }
-        if let Some(mt) = req.max_tokens {
-            tool_req = tool_req.with_max_tokens(mt);
-        }
-        if let Some(ref tc) = req.tool_choice
-            && let Some(choice) = normalize_tool_choice(tc)
-        {
-            tool_req = tool_req.with_tool_choice(choice);
-        }
+        let tool_req = build_tool_request(&req, messages);
 
         let resp = llm
             .complete_with_tools(tool_req)
@@ -527,16 +553,7 @@ pub async fn chat_completions_handler(
 
         Ok(Json(response).into_response())
     } else {
-        let mut comp_req = CompletionRequest::new(messages).with_model(req.model);
-        if let Some(t) = req.temperature {
-            comp_req = comp_req.with_temperature(t);
-        }
-        if let Some(mt) = req.max_tokens {
-            comp_req = comp_req.with_max_tokens(mt);
-        }
-        if let Some(ref stop_val) = req.stop {
-            comp_req.stop_sequences = parse_stop(stop_val);
-        }
+        let comp_req = build_completion_request(&req, messages);
 
         let resp = llm.complete(comp_req).await.map_err(map_llm_error)?;
         let model_name = llm.effective_model_name(Some(requested_model.as_str()));
@@ -596,35 +613,14 @@ async fn handle_streaming(
     }
 
     let llm_result = if has_tools {
-        let tools = convert_tools(req.tools.as_deref().unwrap_or(&[]));
-        let mut tool_req = ToolCompletionRequest::new(messages, tools).with_model(req.model);
-        if let Some(t) = req.temperature {
-            tool_req = tool_req.with_temperature(t);
-        }
-        if let Some(mt) = req.max_tokens {
-            tool_req = tool_req.with_max_tokens(mt);
-        }
-        if let Some(ref tc) = req.tool_choice
-            && let Some(choice) = normalize_tool_choice(tc)
-        {
-            tool_req = tool_req.with_tool_choice(choice);
-        }
+        let tool_req = build_tool_request(&req, messages);
         LlmResult::WithTools(
             llm.complete_with_tools(tool_req)
                 .await
                 .map_err(map_llm_error)?,
         )
     } else {
-        let mut comp_req = CompletionRequest::new(messages).with_model(req.model);
-        if let Some(t) = req.temperature {
-            comp_req = comp_req.with_temperature(t);
-        }
-        if let Some(mt) = req.max_tokens {
-            comp_req = comp_req.with_max_tokens(mt);
-        }
-        if let Some(ref stop_val) = req.stop {
-            comp_req.stop_sequences = parse_stop(stop_val);
-        }
+        let comp_req = build_completion_request(&req, messages);
         LlmResult::Simple(llm.complete(comp_req).await.map_err(map_llm_error)?)
     };
     let model_name = llm.effective_model_name(Some(requested_model.as_str()));

src/llm/bedrock.rs

@@ -176,8 +176,11 @@ impl LlmProvider for BedrockProvider {
             builder = builder.tool_config(tc);
         }
 
-        if let Some(config) = build_inference_config(request.temperature, request.max_tokens, None)
-        {
+        if let Some(config) = build_inference_config(
+            request.temperature,
+            request.max_tokens,
+            request.stop_sequences.as_deref(),
+        ) {
             builder = builder.inference_config(config);
         }
 

src/llm/nearai_chat.rs

@@ -475,6 +475,7 @@ impl LlmProvider for NearAiChatProvider {
             messages,
             temperature: req.temperature,
             max_tokens: req.max_tokens,
+            stop: req.stop_sequences,
             tools: None,
             tool_choice: None,
         };
@@ -554,6 +555,7 @@ impl LlmProvider for NearAiChatProvider {
             messages,
             temperature: req.temperature,
             max_tokens: req.max_tokens,
+            stop: req.stop_sequences,
             tools: if tools.is_empty() { None } else { Some(tools) },
             tool_choice: req.tool_choice,
         };
@@ -680,6 +682,8 @@ struct ChatCompletionRequest {
     #[serde(skip_serializing_if = "Option::is_none")]
     max_tokens: Option<u32>,
     #[serde(skip_serializing_if = "Option::is_none")]
+    stop: Option<Vec<String>>,
+    #[serde(skip_serializing_if = "Option::is_none")]
     tools: Option<Vec<ChatCompletionTool>>,
     #[serde(skip_serializing_if = "Option::is_none")]
     tool_choice: Option<String>,
@@ -1666,6 +1670,7 @@ mod tests {
             }],
             temperature: None,
             max_tokens: None,
+            stop: None,
             tools: None,
             tool_choice: None,
         };
@@ -1687,6 +1692,7 @@ mod tests {
             messages: vec![],
             temperature: Some(0.7),
             max_tokens: Some(1024),
+            stop: None,
             tools: Some(vec![ChatCompletionTool {
                 tool_type: "function".to_string(),
                 function: ChatCompletionFunction {

src/llm/provider.rs

@@ -251,6 +251,7 @@ pub struct ToolCompletionRequest {
     pub model: Option<String>,
     pub max_tokens: Option<u32>,
     pub temperature: Option<f32>,
+    pub stop_sequences: Option<Vec<String>>,
     /// How to handle tool use: "auto", "required", or "none".
     pub tool_choice: Option<String>,
     /// Opaque metadata passed through to the provider (e.g. thread_id for chaining).
@@ -266,6 +267,7 @@ impl ToolCompletionRequest {
             model: None,
             max_tokens: None,
             temperature: None,
+            stop_sequences: None,
             tool_choice: None,
             metadata: std::collections::HashMap::new(),
         }
@@ -289,6 +291,12 @@ impl ToolCompletionRequest {
         self
     }
 
+    /// Set stop sequences.
+    pub fn with_stop_sequences(mut self, stop_sequences: Vec<String>) -> Self {
+        self.stop_sequences = Some(stop_sequences);
+        self
+    }
+
     /// Set tool choice mode.
     pub fn with_tool_choice(mut self, choice: impl Into<String>) -> Self {
         self.tool_choice = Some(choice.into());
@@ -504,8 +512,6 @@ pub fn strip_unsupported_completion_params(
 /// This is the single helper function used by all providers to remove
 /// parameters they don't support from tool calls, replacing duplicate stringly-typed logic.
 ///
-/// Note: Only `Temperature` and `MaxTokens` are supported in `ToolCompletionRequest`.
-/// `StopSequences` is only available in `CompletionRequest` and is not applicable to tool calls.
 pub fn strip_unsupported_tool_params(
     unsupported: &std::collections::HashSet<String>,
     req: &mut ToolCompletionRequest,
@@ -519,7 +525,9 @@ pub fn strip_unsupported_tool_params(
     if unsupported.contains(UnsupportedParam::MaxTokens.name()) {
         req.max_tokens = None;
     }
-    // Note: StopSequences is not a field in ToolCompletionRequest, so no action needed
+    if unsupported.contains(UnsupportedParam::StopSequences.name()) {
+        req.stop_sequences = None;
+    }
 }
 
 #[cfg(test)]
@@ -651,4 +659,17 @@ mod tests {
         assert!(messages[2].tool_call_id.is_none());
         assert!(messages[2].name.is_none());
     }
+
+    #[test]
+    fn test_strip_unsupported_tool_params_strips_stop_sequences() {
+        let mut unsupported = std::collections::HashSet::new();
+        unsupported.insert(UnsupportedParam::StopSequences.name().to_string());
+
+        let mut req = ToolCompletionRequest::new(vec![ChatMessage::user("hello")], vec![]);
+        req.stop_sequences = Some(vec!["STOP".to_string()]);
+
+        strip_unsupported_tool_params(&unsupported, &mut req);
+
+        assert!(req.stop_sequences.is_none()); // safety: test assertion for explicit strip behavior
+    }
 }

src/llm/response_cache.rs

@@ -548,6 +548,7 @@ mod tests {
             model: None,
             max_tokens: None,
             temperature: None,
+            stop_sequences: None,
             tool_choice: None,
             metadata: Default::default(),
         };

src/orchestrator/api.rs

@@ -176,6 +176,7 @@ async fn llm_complete_with_tools(
         model: req.model,
         max_tokens: req.max_tokens,
         temperature: req.temperature,
+        stop_sequences: req.stop_sequences,
         tool_choice: req.tool_choice,
         metadata: std::collections::HashMap::new(),
     };

src/worker/api.rs

@@ -65,6 +65,7 @@ pub struct ProxyToolCompletionRequest {
     pub model: Option<String>,
     pub max_tokens: Option<u32>,
     pub temperature: Option<f32>,
+    pub stop_sequences: Option<Vec<String>>,
     pub tool_choice: Option<String>,
 }
 
@@ -251,6 +252,7 @@ impl WorkerHttpClient {
             model: request.model.clone(),
             max_tokens: request.max_tokens,
             temperature: request.temperature,
+            stop_sequences: request.stop_sequences.clone(),
             tool_choice: request.tool_choice.clone(),
         };