feat(FN-3300): restore Claude usage tracking with Fusion Anthropic auth sto
This merge restores Claude usage tracking by reading from Fusion's Anthropic auth storage (FN-3135, FN-3289), adds fixes for response-text extraction bugs in `continueAgentConversation` and `getFirstQuestionFromAgent` with diagnostic logging for empty responses (FN-3300), and includes new test cover Fusion-Task-Id: FN-3300
This commit is contained in:
Fusion
@@ -2893,3 +2893,270 @@ describe("planning routes lock enforcement", () => {
|
||||
expect(String(streamResponse.body)).toContain("event: complete");
|
||||
});
|
||||
});
|
||||
|
||||
// ── Thinking-Block Response Extraction Tests (FN-3300) ─────────────────────
|
||||
|
||||
describe("FN-3300: thinking-block response extraction", () => {
|
||||
/**
|
||||
* Creates a mock agent that returns array content blocks (thinking + text).
|
||||
* This simulates Claude-style extended thinking responses.
|
||||
*/
|
||||
function createMockAgentWithBlocks(
|
||||
responses: Array<
|
||||
| string
|
||||
| Array<{ type: string; text?: string; thinking?: string }>
|
||||
>,
|
||||
) {
|
||||
const messages: Array<{
|
||||
role: string;
|
||||
content:
|
||||
| string
|
||||
| Array<{ type: string; text?: string; thinking?: string }>;
|
||||
}> = [];
|
||||
let callIndex = 0;
|
||||
|
||||
return {
|
||||
session: {
|
||||
state: { messages },
|
||||
prompt: vi.fn(async (msg: string) => {
|
||||
messages.push({ role: "user", content: msg });
|
||||
const response = responses[callIndex++] ?? responses[responses.length - 1];
|
||||
messages.push({ role: "assistant", content: response });
|
||||
}),
|
||||
dispose: vi.fn(),
|
||||
},
|
||||
};
|
||||
}
|
||||
|
||||
/**
|
||||
* Creates a mock streaming agent with array content blocks and callbacks.
|
||||
*/
|
||||
function setupMockStreamingAgentWithBlocks(options: {
|
||||
contentBlocks: Array<
|
||||
| string
|
||||
| Array<{ type: string; text?: string; thinking?: string }>
|
||||
>;
|
||||
thinkingOutputPerPrompt?: string[];
|
||||
}) {
|
||||
const contentBlocks = options.contentBlocks;
|
||||
const thinkingOutputPerPrompt = options.thinkingOutputPerPrompt ?? [];
|
||||
let promptIndex = 0;
|
||||
|
||||
const createFnAgentSpy = vi.fn(
|
||||
async (agentOptions?: {
|
||||
onThinking?: (delta: string) => void;
|
||||
onText?: (delta: string) => void;
|
||||
}) => {
|
||||
const messages: Array<{
|
||||
role: string;
|
||||
content:
|
||||
| string
|
||||
| Array<{ type: string; text?: string; thinking?: string }>;
|
||||
}> = [];
|
||||
|
||||
return {
|
||||
session: {
|
||||
state: { messages },
|
||||
prompt: vi.fn(async (message: string) => {
|
||||
messages.push({ role: "user", content: message });
|
||||
const thinking = thinkingOutputPerPrompt[promptIndex];
|
||||
if (thinking) {
|
||||
agentOptions?.onText?.(thinking);
|
||||
}
|
||||
const response = contentBlocks[promptIndex] ?? contentBlocks[contentBlocks.length - 1];
|
||||
messages.push({ role: "assistant", content: response });
|
||||
promptIndex += 1;
|
||||
}),
|
||||
dispose: vi.fn(),
|
||||
},
|
||||
};
|
||||
},
|
||||
);
|
||||
|
||||
__setCreateFnAgent(createFnAgentSpy as any);
|
||||
return { createFnAgentSpy };
|
||||
}
|
||||
|
||||
const questionJson = JSON.stringify({
|
||||
type: "question",
|
||||
data: {
|
||||
id: "q-scope",
|
||||
type: "single_select",
|
||||
question: "What is the scope?",
|
||||
description: "Describe the scope.",
|
||||
options: [
|
||||
{ id: "small", label: "Small", description: "Quick" },
|
||||
{ id: "medium", label: "Medium", description: "Standard" },
|
||||
{ id: "large", label: "Large", description: "Complex" },
|
||||
],
|
||||
},
|
||||
});
|
||||
|
||||
beforeEach(() => {
|
||||
__resetPlanningState();
|
||||
});
|
||||
|
||||
describe("continueAgentConversation (streaming path)", () => {
|
||||
it("falls back to thinkingOutput when message content has only thinking blocks", async () => {
|
||||
// The streaming agent accumulates text via onText callback into thinkingOutput.
|
||||
// When the message content array has only thinking-type blocks, the
|
||||
// text blocks filter yields empty string. The fix ensures we fall back
|
||||
// to the accumulated thinkingOutput instead of overwriting with "".
|
||||
setupMockStreamingAgentWithBlocks({
|
||||
contentBlocks: [
|
||||
// First prompt: only thinking blocks in message content
|
||||
[{ type: "thinking", thinking: "Let me think about this..." }],
|
||||
// Retry prompt: valid text response
|
||||
questionJson,
|
||||
],
|
||||
// The actual JSON was accumulated via onText callback during streaming
|
||||
thinkingOutputPerPrompt: [questionJson, questionJson],
|
||||
});
|
||||
|
||||
const { sessionId } = await createSessionWithAgent(
|
||||
getUniqueIp(),
|
||||
"Test plan",
|
||||
TEST_ROOT_DIR,
|
||||
);
|
||||
|
||||
// Submit response to trigger continueAgentConversation
|
||||
const result = await submitResponse(sessionId, { scope: "medium" });
|
||||
expect(result.type).toBe("question");
|
||||
if (result.type === "question") {
|
||||
expect(result.data.id).toBe("q-scope");
|
||||
}
|
||||
});
|
||||
|
||||
it("prefers text blocks over thinkingOutput when both are present", async () => {
|
||||
const differentJson = JSON.stringify({
|
||||
type: "question",
|
||||
data: {
|
||||
id: "q-from-text-block",
|
||||
type: "text",
|
||||
question: "What do you need?",
|
||||
description: "Describe.",
|
||||
},
|
||||
});
|
||||
|
||||
setupMockStreamingAgentWithBlocks({
|
||||
contentBlocks: [
|
||||
// Message has both thinking AND text blocks
|
||||
[
|
||||
{ type: "thinking", thinking: "Thinking about the response..." },
|
||||
{ type: "text", text: differentJson },
|
||||
],
|
||||
],
|
||||
// thinkingOutput has something different — should NOT be used
|
||||
thinkingOutputPerPrompt: ["old-thinking-output"],
|
||||
});
|
||||
|
||||
const { sessionId } = await createSessionWithAgent(
|
||||
getUniqueIp(),
|
||||
"Test plan",
|
||||
TEST_ROOT_DIR,
|
||||
);
|
||||
|
||||
const result = await submitResponse(sessionId, { scope: "medium" });
|
||||
expect(result.type).toBe("question");
|
||||
if (result.type === "question") {
|
||||
expect(result.data.id).toBe("q-from-text-block");
|
||||
}
|
||||
});
|
||||
});
|
||||
|
||||
describe("getFirstQuestionFromAgent (non-streaming path)", () => {
|
||||
it("extracts thinking block content when no text blocks are present", async () => {
|
||||
// Non-streaming path: createSession uses createMockAgent which returns
|
||||
// array content blocks. When only thinking blocks exist, extract their text.
|
||||
const agent = createMockAgentWithBlocks([
|
||||
// Only thinking blocks — the JSON is inside the thinking text
|
||||
[{ type: "thinking", thinking: questionJson }],
|
||||
]);
|
||||
__setCreateFnAgent(async () => agent);
|
||||
|
||||
const result = await createSession(
|
||||
getUniqueIp(),
|
||||
"Test plan",
|
||||
undefined,
|
||||
TEST_ROOT_DIR,
|
||||
);
|
||||
|
||||
expect(result.firstQuestion).toBeDefined();
|
||||
expect(result.firstQuestion.id).toBe("q-scope");
|
||||
});
|
||||
|
||||
it("prefers text blocks over thinking blocks when both present", async () => {
|
||||
const textBlockJson = JSON.stringify({
|
||||
type: "question",
|
||||
data: {
|
||||
id: "q-from-text",
|
||||
type: "text",
|
||||
question: "Text block question?",
|
||||
description: "From text block.",
|
||||
},
|
||||
});
|
||||
|
||||
const agent = createMockAgentWithBlocks([
|
||||
[
|
||||
{ type: "thinking", thinking: questionJson },
|
||||
{ type: "text", text: textBlockJson },
|
||||
],
|
||||
]);
|
||||
__setCreateFnAgent(async () => agent);
|
||||
|
||||
const result = await createSession(
|
||||
getUniqueIp(),
|
||||
"Test plan",
|
||||
undefined,
|
||||
TEST_ROOT_DIR,
|
||||
);
|
||||
|
||||
expect(result.firstQuestion).toBeDefined();
|
||||
expect(result.firstQuestion.id).toBe("q-from-text");
|
||||
});
|
||||
});
|
||||
|
||||
describe("diagnostics logging for empty response text", () => {
|
||||
it("logs warning when response text is empty after extraction", async () => {
|
||||
const { setDiagnosticsSink, resetDiagnosticsSink: resetSink } = await import(
|
||||
"../ai-session-diagnostics.js"
|
||||
);
|
||||
|
||||
const warnings: Array<{
|
||||
level: string;
|
||||
scope: string;
|
||||
message: string;
|
||||
context: Record<string, unknown>;
|
||||
}> = [];
|
||||
setDiagnosticsSink((level, scope, message, context) => {
|
||||
warnings.push({ level, scope, message, context });
|
||||
});
|
||||
|
||||
try {
|
||||
// Agent returns content array with only thinking blocks, and no
|
||||
// thinking text in them either — truly empty thinking blocks
|
||||
const agent = createMockAgentWithBlocks([
|
||||
[{ type: "thinking", thinking: "" }],
|
||||
]);
|
||||
__setCreateFnAgent(async () => agent);
|
||||
|
||||
await expect(
|
||||
createSession(getUniqueIp(), "Test plan", undefined, TEST_ROOT_DIR),
|
||||
).rejects.toThrow("Failed to get first question from AI");
|
||||
|
||||
// Should have logged a warning about empty response text
|
||||
const extractionWarning = warnings.find(
|
||||
(w) =>
|
||||
w.message === "Response text is empty or very short before parse" &&
|
||||
w.context.operation === "response-extraction",
|
||||
);
|
||||
expect(extractionWarning).toBeDefined();
|
||||
expect(extractionWarning!.context.contentBlockTypes).toEqual([
|
||||
"thinking",
|
||||
]);
|
||||
} finally {
|
||||
resetSink();
|
||||
}
|
||||
});
|
||||
});
|
||||
});
|
||||
|
||||
Reference in New Issue
Block a user