Files
oh-my-pi/packages/coding-agent/test/advisor-context-maintenance.test.ts
T
Paolo Mazzitti 9d240ea0ad feat(coding-agent): show advisor cost separately in the status line
Render the Advisor spend next to the primary-model cost as `$2.67 (sub) + $0.41 (adv)`, leaving the status line unchanged until an Advisor cost exists.

Record the cost from finalized advisor `message_end` events in a per-session ledger instead of deriving it from the live advisor transcript, so an in-session compaction or any other history rewrite no longer resets the reported spend. The ledger is cleared for a new session and once a different-session switch commits, and survives a switch that rolls back.
2026-07-27 13:52:22 +00:00

275 lines
12 KiB
TypeScript

import { afterEach, beforeEach, describe, expect, it, vi } from "bun:test";
import { Agent, type AgentMessage, type CompactionSummaryMessage, countTokens } from "@oh-my-pi/pi-agent-core";
import { calculateContextTokens, estimateTokens, resolveThresholdTokens } from "@oh-my-pi/pi-agent-core/compaction";
import type { AssistantMessage } from "@oh-my-pi/pi-ai";
import { createMockModel, type MockModel, registerMockApi } from "@oh-my-pi/pi-ai/providers/mock";
import { ModelRegistry } from "@oh-my-pi/pi-coding-agent/config/model-registry";
import { Settings } from "@oh-my-pi/pi-coding-agent/config/settings";
import { estimateToolSchemaTokens } from "@oh-my-pi/pi-coding-agent/modes/utils/context-usage";
import { AgentSession } from "@oh-my-pi/pi-coding-agent/session/agent-session";
import { AuthStorage } from "@oh-my-pi/pi-coding-agent/session/auth-storage";
import { SessionManager } from "@oh-my-pi/pi-coding-agent/session/session-manager";
import { TempDir } from "@oh-my-pi/pi-utils";
const CONTEXT_WINDOW = 372_000;
const CACHE_READ_TOKENS = 371_200;
const INPUT_TOKENS = 200;
const OUTPUT_TOKENS = 150;
interface MaintenanceHarness {
advisor: Agent;
advisorMock: MockModel;
settings: Settings;
}
interface AdvisorCompactionSummaryFixture extends CompactionSummaryMessage {
advisorUsageAnchorStartIndex?: number;
}
describe("AgentSession advisor context maintenance", () => {
let tempDir: TempDir;
let authStorage: AuthStorage;
let session: AgentSession;
beforeEach(async () => {
tempDir = TempDir.createSync("@pi-advisor-context-maintenance-");
authStorage = await AuthStorage.create(tempDir.join("auth.db"));
authStorage.setRuntimeApiKey("anthropic", "test-key");
});
afterEach(async () => {
vi.restoreAllMocks();
await session?.dispose();
authStorage.close();
await tempDir.remove();
});
function createHarness(): MaintenanceHarness {
const primaryMock = createMockModel({
provider: "anthropic",
responses: [{ content: ["primary complete"] }],
});
const advisorMock = createMockModel({
provider: "anthropic",
contextWindow: CONTEXT_WINDOW,
responses: [{ content: ["advisor reviewed current update"] }],
});
const modelRegistry = new ModelRegistry(authStorage, tempDir.join("models.yml"));
const settings = Settings.isolated({
"advisor.syncBacklog": "1",
"compaction.enabled": true,
"compaction.strategy": "context-full",
"contextPromotion.enabled": false,
});
const agent = new Agent({
getApiKey: () => "test-key",
initialState: { model: primaryMock, systemPrompt: [], tools: [] },
streamFn: primaryMock.stream,
});
session = new AgentSession({
agent,
sessionManager: SessionManager.inMemory(),
settings,
modelRegistry,
advisorTools: [],
advisorStreamFn: advisorMock.stream,
});
settings.setModelRole("advisor", "anthropic/claude-sonnet-4-5");
expect(session.setAdvisorEnabled(true)).toBe(true);
const advisor = session.getAdvisorAgent();
if (!advisor) throw new Error("Expected advisor agent to be active");
advisor.setModel(advisorMock);
// Keep maintenance on the no-summary recovery branch without blocking the
// primary prompt's own credential preflight.
vi.spyOn(modelRegistry, "getApiKey").mockImplementation(async model =>
model === primaryMock ? "test-key" : undefined,
);
return { advisor, advisorMock, settings };
}
function usageAnchor(advisorMock: MockModel, timestamp: number, cost = 0): AssistantMessage {
return {
role: "assistant",
content: [{ type: "text", text: "prior advisor output" }],
api: advisorMock.api,
provider: advisorMock.provider,
model: advisorMock.id,
usage: {
input: INPUT_TOKENS,
output: OUTPUT_TOKENS,
cacheRead: CACHE_READ_TOKENS,
cacheWrite: 0,
totalTokens: CACHE_READ_TOKENS + INPUT_TOKENS + OUTPUT_TOKENS,
cost: { input: 0, output: cost, cacheRead: 0, cacheWrite: 0, total: cost },
},
stopReason: "stop",
timestamp,
};
}
function compactionSummary(timestamp: number): AdvisorCompactionSummaryFixture {
return {
role: "compactionSummary",
summary: "bounded advisor summary",
tokensBefore: CACHE_READ_TOKENS + INPUT_TOKENS + OUTPUT_TOKENS,
timestamp,
// `[summary, retained]` is the compacted array; index 2 is the first
// position eligible for a newly appended provider-usage anchor.
advisorUsageAnchorStartIndex: 2,
};
}
it("maintains a 371,200-token cached advisor context before the 372,000-token window", async () => {
const { advisor, advisorMock, settings } = createHarness();
const anchor = usageAnchor(advisorMock, Date.now() - 1_000, 0.5);
advisor.emitExternalEvent({ type: "message_end", message: anchor });
expect(session.getAdvisorCost()).toBeCloseTo(0.5, 8);
await session.prompt("small current update");
expect(advisorMock.calls).toHaveLength(1);
const advisorCall = advisorMock.calls[0];
const update = advisorCall.context.messages.find(message => message.role === "user");
if (!update) throw new Error("Expected the advisor's incremental update");
const threshold = resolveThresholdTokens(CONTEXT_WINDOW, settings.getGroup("compaction"));
const providerAndUpdateTokens = calculateContextTokens(anchor.usage) + estimateTokens(update as AgentMessage);
expect(calculateContextTokens(anchor.usage)).toBe(CACHE_READ_TOKENS + INPUT_TOKENS + OUTPUT_TOKENS);
expect(providerAndUpdateTokens).toBeGreaterThan(threshold);
// Provider usage triggers maintenance, but recovery sends only the bounded
// current update into the reset advisor context.
expect(JSON.stringify(advisorCall.context.messages)).toContain("small current update");
expect(JSON.stringify(advisor.state.messages)).not.toContain("prior advisor output");
expect(session.getAdvisorCost()).toBeCloseTo(0.5, 8);
});
it("includes advisor system prompt and tool schemas in the local maintenance floor", async () => {
const { advisor, advisorMock, settings } = createHarness();
const seed: AgentMessage = { role: "user", content: "small stored advisor message", timestamp: 1 };
advisor.state.messages.push(seed);
const storedTokens = estimateTokens(seed, { excludeEncryptedReasoning: true });
const fixedPrefixTokens = countTokens(advisor.state.systemPrompt) + estimateToolSchemaTokens(advisor.state.tools);
const threshold = storedTokens + Math.floor(fixedPrefixTokens / 2);
settings.set("compaction.thresholdTokens", threshold);
await session.prompt("tiny local-floor update");
const advisorCall = advisorMock.calls[0];
const update = advisorCall.context.messages.find(message => message.role === "user");
if (!update) throw new Error("Expected the advisor's incremental update");
const messagesOnlyTokens = storedTokens + estimateTokens(update as AgentMessage);
expect(messagesOnlyTokens).toBeLessThan(threshold);
expect(messagesOnlyTokens + fixedPrefixTokens).toBeGreaterThan(threshold);
expect(JSON.stringify(advisor.state.messages)).not.toContain("small stored advisor message");
});
it("ignores retained provider usage that predates the latest advisor compaction", async () => {
const { advisor, advisorMock } = createHarness();
const compactedAt = Date.now();
const summary = compactionSummary(compactedAt);
const retained = usageAnchor(advisorMock, compactedAt);
retained.content = [{ type: "text", text: "retained pre-compaction output" }];
advisor.state.messages.push(summary, retained);
await session.prompt("post-compaction update");
expect(advisorMock.calls).toHaveLength(1);
const sentContext = JSON.stringify(advisorMock.calls[0].context.messages);
expect(sentContext).toContain("retained pre-compaction output");
expect(sentContext).toContain("post-compaction update");
});
it("accepts equal-timestamp usage appended after the explicit compaction boundary", async () => {
const { advisor, advisorMock } = createHarness();
const compactedAt = Date.now();
const summary = compactionSummary(compactedAt);
const retained = usageAnchor(advisorMock, compactedAt);
retained.content = [{ type: "text", text: "retained pre-compaction output" }];
const fresh = usageAnchor(advisorMock, compactedAt);
fresh.content = [{ type: "text", text: "fresh post-compaction output" }];
advisor.state.messages.push(summary, retained, fresh);
await session.prompt("equal-timestamp post-compaction update");
expect(advisorMock.calls).toHaveLength(1);
const sentContext = JSON.stringify(advisorMock.calls[0].context.messages);
expect(sentContext).toContain("equal-timestamp post-compaction update");
expect(sentContext).not.toContain("retained pre-compaction output");
expect(sentContext).not.toContain("fresh post-compaction output");
});
it("forwards the advisor session metadata to overflow compaction requests", async () => {
// Regression for #6625 review: advisor overflow compaction issues a direct
// `compact(...)` request that bypasses the advisor `Agent`, so the metadata
// resolver installed on the agent never runs for it. The direct call must
// still emit the advisor's `metadata.user_id` session identity.
// The advisor model is the first compaction candidate; registering the mock
// API lets the compaction one-shot's `completeSimple` route to it so the
// summarization request actually reaches the mock (and its recorded calls).
registerMockApi();
const primaryMock = createMockModel({
provider: "anthropic",
responses: [{ content: ["primary complete"] }],
});
const advisorMock = createMockModel({
provider: "anthropic",
contextWindow: CONTEXT_WINDOW,
handler: () => ({ content: ["bounded advisor summary"] }),
});
const modelRegistry = new ModelRegistry(authStorage, tempDir.join("models.yml"));
const settings = Settings.isolated({
"advisor.syncBacklog": "1",
"compaction.enabled": true,
"compaction.strategy": "context-full",
"contextPromotion.enabled": false,
});
const agent = new Agent({
getApiKey: () => "test-key",
initialState: { model: primaryMock, systemPrompt: [], tools: [] },
streamFn: primaryMock.stream,
});
session = new AgentSession({
agent,
sessionManager: SessionManager.inMemory(),
settings,
modelRegistry,
advisorTools: [],
advisorStreamFn: advisorMock.stream,
});
settings.setModelRole("advisor", "anthropic/claude-sonnet-4-5");
expect(session.setAdvisorEnabled(true)).toBe(true);
const advisor = session.getAdvisorAgent();
if (!advisor?.sessionId) throw new Error("Expected advisor agent with a provider session id");
advisor.setModel(advisorMock);
// Unlike the recovery-branch harness, the advisor holds usable credentials
// so maintenance runs the LLM summarization compaction path.
vi.spyOn(modelRegistry, "getApiKey").mockResolvedValue("test-key");
// Two accumulated turns so compaction has older history to summarize while
// retaining the most recent one (a single message would be fully retained,
// making compaction a no-op).
advisor.state.messages.push(
usageAnchor(advisorMock, Date.now() - 2_000),
usageAnchor(advisorMock, Date.now() - 1_000),
);
await session.prompt("small current update");
// A summarization compaction one-shot actually ran (its prompt wraps the
// conversation in <conversation> tags).
const compactionCalls = advisorMock.calls.filter(call =>
JSON.stringify(call.context.messages).includes("<conversation>"),
);
expect(compactionCalls.length).toBeGreaterThan(0);
// Every advisor request — the compaction one-shot and the advisor turn —
// carries the advisor's own provider session id via metadata.user_id.
for (const call of advisorMock.calls) {
const userId = call.options?.metadata?.user_id;
if (typeof userId !== "string") throw new Error("Expected advisor metadata.user_id");
expect((JSON.parse(userId) as { session_id?: string }).session_id).toBe(advisor.sessionId);
}
});
});