test(agents): dedupe agent and cron test scaffolds

This commit is contained in:
Peter Steinberger
2026-03-02 06:40:42 +00:00
parent 281494ae52
commit 7e29d604ba
38 changed files with 3114 additions and 4486 deletions

View File

@@ -13,6 +13,7 @@ vi.mock("@mariozechner/pi-coding-agent", async (importOriginal) => {
});
const mockGenerateSummary = vi.mocked(piCodingAgent.generateSummary);
type SummarizeInStagesInput = Parameters<typeof summarizeInStages>[0];
function makeMessage(index: number, size = 1200): AgentMessage {
return {
@@ -28,58 +29,63 @@ describe("compaction identifier-preservation instructions", () => {
model: "claude-3-opus",
contextWindow: 200_000,
} as unknown as NonNullable<ExtensionContext["model"]>;
const summarizeBase: Omit<SummarizeInStagesInput, "messages"> = {
model: testModel,
apiKey: "test-key",
reserveTokens: 4000,
maxChunkTokens: 8000,
contextWindow: 200_000,
signal: new AbortController().signal,
};
beforeEach(() => {
mockGenerateSummary.mockReset();
mockGenerateSummary.mockResolvedValue("summary");
});
it("injects identifier-preservation guidance even without custom instructions", async () => {
async function runSummary(
messageCount: number,
overrides: Partial<Omit<SummarizeInStagesInput, "messages">> = {},
) {
await summarizeInStages({
messages: [makeMessage(1), makeMessage(2)],
model: testModel,
apiKey: "test-key",
...summarizeBase,
...overrides,
signal: new AbortController().signal,
reserveTokens: 4000,
maxChunkTokens: 8000,
contextWindow: 200_000,
messages: Array.from({ length: messageCount }, (_unused, index) => makeMessage(index + 1)),
});
}
function firstSummaryInstructions() {
return mockGenerateSummary.mock.calls[0]?.[5];
}
it("injects identifier-preservation guidance even without custom instructions", async () => {
await runSummary(2);
expect(mockGenerateSummary).toHaveBeenCalled();
const firstCall = mockGenerateSummary.mock.calls[0];
expect(firstCall?.[5]).toContain("Preserve all opaque identifiers exactly as written");
expect(firstCall?.[5]).toContain("UUIDs");
expect(firstCall?.[5]).toContain("IPs");
expect(firstCall?.[5]).toContain("ports");
expect(firstSummaryInstructions()).toContain(
"Preserve all opaque identifiers exactly as written",
);
expect(firstSummaryInstructions()).toContain("UUIDs");
expect(firstSummaryInstructions()).toContain("IPs");
expect(firstSummaryInstructions()).toContain("ports");
});
it("keeps identifier-preservation guidance when custom instructions are provided", async () => {
await summarizeInStages({
messages: [makeMessage(1), makeMessage(2)],
model: testModel,
apiKey: "test-key",
signal: new AbortController().signal,
reserveTokens: 4000,
maxChunkTokens: 8000,
contextWindow: 200_000,
await runSummary(2, {
customInstructions: "Focus on release-impacting bugs.",
});
const firstCall = mockGenerateSummary.mock.calls[0];
expect(firstCall?.[5]).toContain("Preserve all opaque identifiers exactly as written");
expect(firstCall?.[5]).toContain("Additional focus:");
expect(firstCall?.[5]).toContain("Focus on release-impacting bugs.");
expect(firstSummaryInstructions()).toContain(
"Preserve all opaque identifiers exactly as written",
);
expect(firstSummaryInstructions()).toContain("Additional focus:");
expect(firstSummaryInstructions()).toContain("Focus on release-impacting bugs.");
});
it("applies identifier-preservation guidance on staged split + merge summarization", async () => {
await summarizeInStages({
messages: [makeMessage(1), makeMessage(2), makeMessage(3), makeMessage(4)],
model: testModel,
apiKey: "test-key",
signal: new AbortController().signal,
reserveTokens: 4000,
await runSummary(4, {
maxChunkTokens: 1000,
contextWindow: 200_000,
parts: 2,
minMessagesForSplit: 4,
});
@@ -91,14 +97,8 @@ describe("compaction identifier-preservation instructions", () => {
});
it("avoids duplicate additional-focus headers in split+merge path", async () => {
await summarizeInStages({
messages: [makeMessage(1), makeMessage(2), makeMessage(3), makeMessage(4)],
model: testModel,
apiKey: "test-key",
signal: new AbortController().signal,
reserveTokens: 4000,
await runSummary(4, {
maxChunkTokens: 1000,
contextWindow: 200_000,
parts: 2,
minMessagesForSplit: 4,
customInstructions: "Prioritize customer-visible regressions.",