File size: 9,223 Bytes
bc575bc | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 | import type { AgentMessage } from "@earendil-works/pi-agent-core";
import type { AssistantMessage, Context, Model } from "@earendil-works/pi-ai";
import { beforeEach, describe, expect, it, vi } from "vitest";
import {
type CompactionPreparation,
compact,
completeSummarization,
generateSummary,
generateSummaryWithUsage,
} from "../src/core/compaction/index.ts";
const { completeSimpleMock } = vi.hoisted(() => ({
completeSimpleMock: vi.fn(),
}));
vi.mock("@earendil-works/pi-ai/compat", async (importOriginal) => {
const actual = await importOriginal<typeof import("@earendil-works/pi-ai/compat")>();
return {
...actual,
completeSimple: completeSimpleMock,
};
});
function createModel(
reasoning: boolean,
maxTokens = 8192,
compat?: Model<"anthropic-messages">["compat"],
): Model<"anthropic-messages"> {
return {
id: reasoning ? "reasoning-model" : "non-reasoning-model",
name: reasoning ? "Reasoning Model" : "Non-reasoning Model",
api: "anthropic-messages",
provider: "anthropic",
baseUrl: "https://api.anthropic.com",
reasoning,
input: ["text"],
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
contextWindow: 200000,
maxTokens,
...(compat ? { compat } : {}),
};
}
const mockSummaryResponse: AssistantMessage = {
role: "assistant",
content: [{ type: "text", text: "## Goal\nTest summary" }],
api: "anthropic-messages",
provider: "anthropic",
model: "claude-sonnet-4-5",
usage: {
input: 10,
output: 10,
cacheRead: 0,
cacheWrite: 0,
totalTokens: 20,
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
},
stopReason: "stop",
timestamp: Date.now(),
};
const mockToolCallResponse: AssistantMessage = {
...mockSummaryResponse,
content: [{ type: "toolCall", id: "tool-call-1", name: "read", arguments: { path: "README.md" } }],
stopReason: "toolUse",
};
const messages: AgentMessage[] = [{ role: "user", content: "Summarize this.", timestamp: Date.now() }];
describe("generateSummary reasoning options", () => {
beforeEach(() => {
completeSimpleMock.mockReset();
completeSimpleMock.mockResolvedValue(mockSummaryResponse);
});
it("uses the provided thinking level for reasoning-capable models", async () => {
const result = await generateSummaryWithUsage(
messages,
createModel(true),
2000,
"test-key",
undefined,
undefined,
undefined,
undefined,
"medium",
);
expect(result.text).toBe("## Goal\nTest summary");
expect(result.usage).toEqual(mockSummaryResponse.usage);
expect(completeSimpleMock).toHaveBeenCalledTimes(1);
expect(completeSimpleMock.mock.calls[0][2]).toMatchObject({
reasoning: "medium",
apiKey: "test-key",
});
});
it("preserves the string result from generateSummary", async () => {
await expect(generateSummary(messages, createModel(false), 2000, "test-key")).resolves.toBe(
"## Goal\nTest summary",
);
});
it("uses fresh routing sessions without prompt caching", async () => {
await generateSummary(messages, createModel(false), 2000, "test-key");
await generateSummary(messages, createModel(false), 2000, "test-key");
const requestOptions = completeSimpleMock.mock.calls.map((call) => call[2]);
expect(requestOptions).toHaveLength(2);
expect(requestOptions.every((options) => options?.cacheRetention === "none")).toBe(true);
const sessionIds = requestOptions.map((options) => options?.sessionId);
expect(sessionIds[0]).not.toBe(sessionIds[1]);
});
it("honors caller-supplied routing session and tool choice without prompt caching", async () => {
await completeSummarization(
createModel(false),
{ systemPrompt: "Summarize", messages: [] },
{ sessionId: "current-routing-session", cacheRetention: "long", toolChoice: "auto" },
);
expect(completeSimpleMock.mock.calls[0][2]).toMatchObject({
sessionId: "current-routing-session",
cacheRetention: "none",
toolChoice: "auto",
});
});
it("preserves the standalone split-turn summary prompt", async () => {
const preparation: CompactionPreparation = {
firstKeptEntryId: "entry-keep",
messagesToSummarize: [],
turnPrefixMessages: messages,
isSplitTurn: true,
tokensBefore: 100,
fileOps: { read: new Set(), written: new Set(), edited: new Set() },
settings: { enabled: true, reserveTokens: 2000, keepRecentTokens: 20 },
};
await compact(preparation, createModel(false), "test-key");
const requestContext = completeSimpleMock.mock.calls[0][1] as Context;
const prompt = JSON.stringify(requestContext.messages);
expect(prompt).toContain("This is the PREFIX of a turn that was too large to keep");
expect(prompt).toContain("<conversation>");
});
it("rejects tool calls from conversation summaries", async () => {
completeSimpleMock.mockResolvedValueOnce(mockToolCallResponse);
await expect(generateSummaryWithUsage(messages, createModel(false), 2000, "test-key")).rejects.toThrow(
"Summarization attempted to call a tool",
);
});
it("rejects tool calls from split-turn summaries", async () => {
completeSimpleMock.mockResolvedValueOnce(mockToolCallResponse);
const preparation: CompactionPreparation = {
firstKeptEntryId: "entry-keep",
messagesToSummarize: [],
turnPrefixMessages: messages,
isSplitTurn: true,
tokensBefore: 100,
fileOps: { read: new Set(), written: new Set(), edited: new Set() },
settings: { enabled: true, reserveTokens: 2000, keepRecentTokens: 20 },
};
await expect(compact(preparation, createModel(false), "test-key")).rejects.toThrow(
"Turn prefix summarization attempted to call a tool",
);
});
it("rejects a length-limited history summary", async () => {
completeSimpleMock.mockResolvedValueOnce({
...mockSummaryResponse,
stopReason: "length",
content: [{ type: "text", text: "partial" }],
});
await expect(generateSummaryWithUsage(messages, createModel(false), 2000, "test-key")).rejects.toThrow(
"generation hit the token cap",
);
});
it("rejects a length-limited split-turn summary", async () => {
completeSimpleMock.mockResolvedValueOnce({
...mockSummaryResponse,
stopReason: "length",
content: [{ type: "text", text: "partial" }],
});
const preparation: CompactionPreparation = {
firstKeptEntryId: "entry-keep",
messagesToSummarize: [],
turnPrefixMessages: messages,
isSplitTurn: true,
tokensBefore: 100,
fileOps: { read: new Set(), written: new Set(), edited: new Set() },
settings: { enabled: true, reserveTokens: 2000, keepRecentTokens: 20 },
};
await expect(compact(preparation, createModel(false), "test-key")).rejects.toThrow(
"generation hit the token cap",
);
});
it("does not set reasoning when thinking is off", async () => {
await generateSummary(
messages,
createModel(true),
2000,
"test-key",
undefined,
undefined,
undefined,
undefined,
"off",
);
expect(completeSimpleMock).toHaveBeenCalledTimes(1);
expect(completeSimpleMock.mock.calls[0][2]).toMatchObject({
apiKey: "test-key",
});
expect(completeSimpleMock.mock.calls[0][2]).not.toHaveProperty("reasoning");
});
it("does not set reasoning for non-reasoning models", async () => {
await generateSummary(
messages,
createModel(false),
2000,
"test-key",
undefined,
undefined,
undefined,
undefined,
"medium",
);
expect(completeSimpleMock).toHaveBeenCalledTimes(1);
expect(completeSimpleMock.mock.calls[0][2]).toMatchObject({
apiKey: "test-key",
});
expect(completeSimpleMock.mock.calls[0][2]).not.toHaveProperty("reasoning");
});
it("leaves Anthropic refusal fallback handling to pi-ai model metadata", async () => {
await generateSummary(
messages,
createModel(true, 8192, {
allowedFallbackModels: [
{
provider: "anthropic",
model: "claude-opus-4-8",
cost: { input: 5, output: 25, cacheRead: 0.5, cacheWrite: 6.25 },
},
],
}),
2000,
"test-key",
);
expect(completeSimpleMock).toHaveBeenCalledTimes(1);
expect(completeSimpleMock.mock.calls[0][2]).not.toHaveProperty("refusalFallbacks");
});
it("does not set Anthropic refusal fallback for models without allowed fallback targets", async () => {
await generateSummary(messages, createModel(true), 2000, "test-key");
expect(completeSimpleMock).toHaveBeenCalledTimes(1);
expect(completeSimpleMock.mock.calls[0][2]).not.toHaveProperty("refusalFallbacks");
});
it("clamps compaction summary maxTokens to the model output cap", async () => {
const preparation: CompactionPreparation = {
firstKeptEntryId: "entry-keep",
messagesToSummarize: messages,
turnPrefixMessages: messages,
isSplitTurn: true,
tokensBefore: 600000,
fileOps: { read: new Set(), written: new Set(), edited: new Set() },
settings: { enabled: true, reserveTokens: 500000, keepRecentTokens: 20000 },
};
const result = await compact(preparation, createModel(false, 128000), "test-key");
expect(result.usage).toEqual({
...mockSummaryResponse.usage,
input: 20,
output: 20,
totalTokens: 40,
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
});
expect(completeSimpleMock.mock.calls.map((call) => call[2]?.maxTokens)).toEqual([128000, 128000]);
});
});
|