fix: prevent provider_error 400 on dmodel/ultimate by using content placeholder

When assistant messages contain only tool_calls (no text/thinking), transform.ts
was setting content to null. Qoder's gateway drops these null-content assistant
messages, orphaning the following tool_result and causing upstream rejections:
- dmodel (OpenAI): 'tool must follow a message with tool_calls'
- ultimate (Anthropic): 'unexpected tool_use_id in tool_result blocks'

Analysis of 28 historical sessions showed 100% failure rate on dmodel and 75%
on ultimate for this pattern, while auto/GLM5.2 (GLM upstream) tolerated it.

Fix: inject a single-space placeholder when tool_calls are present but content
is empty, so the gateway preserves the assistant message and tool pairing stays
valid.

Also fixes pre-existing lint formatting in stream.ts and package.json.
Adds regression test for the full assistant+toolResult pairing scenario.
This commit is contained in:
liyanan
2026-07-26 15:26:13 +08:00
parent 6964615056
commit 273368fc0e
5 changed files with 51 additions and 8 deletions
+6 -1
View File
@@ -135,9 +135,14 @@ export function transformMessagesForQoder(messages: Message[]): QoderMessage[] {
content = am.content || "";
}
// Qoder's gateway drops assistant messages whose content is null, which
// orphans the following tool_result and makes dmodel/ultimate upstreams
// reject the request ("tool must follow a message with tool_calls").
// When an assistant turn has tool calls but no text/thinking, inject a
// single-space placeholder so the gateway keeps the message.
const mapped: QoderMessage = {
role: "assistant",
content: content || null,
content: content || (toolCalls.length > 0 ? " " : null),
};
if (toolCalls.length > 0) {
mapped.tool_calls = toolCalls;