Skip to content

Solution for long term memory for agent coding CLIs and to facilitate handoff between different agent vendors

Rust

fix(llm): report an empty Responses answer as empty content, not a serde error

Rudy Mizrahi Celeklia577ed3a53fe
diff --git a/crates/ai-memory-llm/src/codex.rs b/crates/ai-memory-llm/src/codex.rs
index 2a975028..b9e1617e 100644
--- a/crates/ai-memory-llm/src/codex.rs
+++ b/crates/ai-memory-llm/src/codex.rs
@@ -210,8 +210,16 @@ impl LlmProvider for CodexProvider {
self.reasoning_effort,
))
.await?;
- serde_json::from_str(&extract_output_text(&response).unwrap_or_default())
- .map_err(LlmError::from)
+ // No content at all is a distinct terminal failure, not a JSON parse
+ // error, as the OpenAI path already treats it: `unwrap_or_default()` fed
+ // an empty string to serde, so the caller saw "EOF while parsing a value
+ // at line 1 column 0", which names neither the model nor the cause.
+ let Some(text) = extract_output_text(&response).filter(|t| !t.trim().is_empty()) else {
+ return Err(LlmError::EmptyContent {
+ model: self.model.clone(),
+ });
+ };
+ serde_json::from_str(&text).map_err(LlmError::from)
}
}
diff --git a/crates/ai-memory-llm/src/openai_oauth.rs b/crates/ai-memory-llm/src/openai_oauth.rs
index a9951d9d..49386404 100644
--- a/crates/ai-memory-llm/src/openai_oauth.rs
+++ b/crates/ai-memory-llm/src/openai_oauth.rs
@@ -264,7 +264,15 @@ impl LlmProvider for OpenAiOAuthProvider {
self.reasoning_effort,
))
.await?;
- let text = extract_output_text(&response).unwrap_or_default();
+ // No content at all is a distinct terminal failure, not a JSON parse
+ // error, as the OpenAI path already treats it: `unwrap_or_default()` fed
+ // an empty string to serde, so the caller saw "EOF while parsing a value
+ // at line 1 column 0", which names neither the model nor the cause.
+ let Some(text) = extract_output_text(&response).filter(|t| !t.trim().is_empty()) else {
+ return Err(LlmError::EmptyContent {
+ model: self.model.clone(),
+ });
+ };
serde_json::from_str::<serde_json::Value>(&text).map_err(LlmError::from)
}
}
diff --git a/crates/ai-memory-llm/src/opencode.rs b/crates/ai-memory-llm/src/opencode.rs
index a011a355..8482af68 100644
--- a/crates/ai-memory-llm/src/opencode.rs
+++ b/crates/ai-memory-llm/src/opencode.rs
@@ -407,7 +407,15 @@ impl OpenCodeResponsesProvider {
let response = self
.post(&self.build_request(&request, Some(text)), operation_id)
.await?;
- let text = extract_output_text(&response).unwrap_or_default();
+ // No content at all is a distinct terminal failure, not a JSON parse
+ // error, as the OpenAI path already treats it: `unwrap_or_default()` fed
+ // an empty string to serde, so the caller saw "EOF while parsing a value
+ // at line 1 column 0", which names neither the model nor the cause.
+ let Some(text) = extract_output_text(&response).filter(|t| !t.trim().is_empty()) else {
+ return Err(LlmError::EmptyContent {
+ model: self.model.clone(),
+ });
+ };
serde_json::from_str(&text).map_err(LlmError::from)
}
@@ -1194,6 +1202,69 @@ mod tests {
assert_eq!(body["model"], "mimo-v2.6-flash");
}
+ /// A 200 with no usable `output_text` is a terminal empty answer, not a
+ /// JSON parse failure. `unwrap_or_default()` fed serde an empty string, so
+ /// the caller saw `serde: EOF while parsing a value at line 1 column 0`,
+ /// which names neither the model nor the cause. The OpenAI-wire structured
+ /// path already classifies this as `empty-content`.
+ #[tokio::test]
+ async fn an_empty_responses_body_is_empty_content_not_a_serde_error() {
+ for (label, body) in [
+ (
+ "no output items",
+ json!({
+ "model": "gpt-5.6-luna",
+ "output": [],
+ "usage": { "input_tokens": 1, "output_tokens": 0 },
+ }),
+ ),
+ (
+ "an item with no content",
+ json!({
+ "model": "gpt-5.6-luna",
+ "output": [{ "type": "reasoning", "content": [] }],
+ "usage": { "input_tokens": 1, "output_tokens": 4096 },
+ }),
+ ),
+ (
+ "whitespace-only text",
+ responses_response_with_content(" \n\t "),
+ ),
+ ] {
+ let server = MockServer::start().await;
+ Mock::given(method("POST"))
+ .and(path("/v1/responses"))
+ .respond_with(ResponseTemplate::new(200).set_body_json(body))
+ .mount(&server)
+ .await;
+
+ let provider = OpenCodeProvider::new_with_base_url(
+ SecretString::from("sk-test"),
+ "gpt-5.6-luna",
+ server.uri(),
+ )
+ .unwrap();
+ let err = provider
+ .complete_structured_raw(
+ ChatRequest::user_prompt("emit JSON"),
+ json!({ "type": "object" }),
+ )
+ .await
+ .expect_err("an empty answer must surface a typed error");
+
+ assert_eq!(err.class(), "empty-content", "{label}: {err:?}");
+ assert!(!err.is_transient(), "{label} must not be retried");
+ match &err {
+ LlmError::EmptyContent { model } => {
+ assert_eq!(model, "gpt-5.6-luna", "{label}");
+ }
+ other => panic!("{label}: expected EmptyContent, got {other:?}"),
+ }
+ let requests = server.received_requests().await.unwrap();
+ assert_eq!(requests.len(), 1, "{label}: exactly one upstream call");
+ }
+ }
+
#[tokio::test]
async fn luna_structured_completion_uses_responses_api() {
let server = MockServer::start().await;