getcompanion-ai/co-mono
1
0
Fork 0
mirror of https://github.com/getcompanion-ai/co-mono.git synced 2026-04-19 14:01:15 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Fix orphaned tool calls by inserting synthetic empty results

Browse source 
When a user interrupts a tool call flow (sends a message without providing
tool results), APIs like OpenAI Responses and Anthropic fail because:
- OpenAI requires tool outputs for function calls
- OpenAI requires reasoning items to have their following items
- Anthropic requires non-empty content for error tool results

Instead of filtering out orphaned tool calls (which breaks thinking signatures),
we now insert synthetic empty tool results with isError: true and content
'No result provided'. This preserves the conversation structure and satisfies
all API requirements.
This commit is contained in:
Mario Zechner 2025-12-20 20:30:57 +01:00
parent 95fcda5887
commit fb1fdb6006
2 changed files with 133 additions and 106 deletions
Download patch file Download diff file Expand all files Collapse all files

30
packages/ai/test/tokens.test.ts
View file

@ -17,7 +17,7 @@ async function testTokensOnAbort<TApi extends Api>(llm: Model<TApi>, options: Op
messages: [
{
role: "user",
content: "Write a long poem with 10 stanzas about the beauty of nature.",
content: "Write a long poem with 20 stanzas about the beauty of nature.",
timestamp: Date.now(),
},
],
@ -27,10 +27,14 @@ async function testTokensOnAbort<TApi extends Api>(llm: Model<TApi>, options: Op
const response = stream(llm, context, { ...options, signal: controller.signal });
let abortFired = false;
let text = "";
for await (const event of response) {
if (!abortFired && (event.type === "text_delta" || event.type === "thinking_delta")) {
abortFired = true;
setTimeout(() => controller.abort(), 3000);
text += event.delta;
if (text.length >= 1000) {
abortFired = true;
controller.abort();
}
}
}
@ -38,16 +42,26 @@ async function testTokensOnAbort<TApi extends Api>(llm: Model<TApi>, options: Op
expect(msg.stopReason).toBe("aborted");
// OpenAI providers only send usage in the final chunk, so when aborted they have no token stats
// Anthropic and Google send usage information early in the stream
if (llm.api === "openai-completions" || llm.api === "openai-responses") {
// OpenAI providers, Gemini CLI, zai, and the GPT-OSS model on Antigravity only send usage in the final chunk,
// so when aborted they have no token stats Anthropic and Google send usage information early in the stream
if (
llm.api === "openai-completions" ||
llm.api === "openai-responses" ||
llm.provider === "google-gemini-cli" ||
llm.provider === "zai" ||
(llm.provider === "google-antigravity" && llm.id.includes("gpt-oss"))
) {
expect(msg.usage.input).toBe(0);
expect(msg.usage.output).toBe(0);
} else {
expect(msg.usage.input).toBeGreaterThan(0);
expect(msg.usage.output).toBeGreaterThan(0);
expect(msg.usage.cost.input).toBeGreaterThan(0);
expect(msg.usage.cost.total).toBeGreaterThan(0);
// Antigravity Gemini and Claude models report token usage, but no cost
if (llm.provider !== "google-antigravity") {
expect(msg.usage.cost.input).toBeGreaterThan(0);
expect(msg.usage.cost.total).toBeGreaterThan(0);
}
}
}