feat(06-01): add session management service with rotation, thresholds, and delta tracking

- Create createSessionManagementService factory with accumulateTokenUsage,
  evaluateRotation (configurable thresholds per D-01), generateHandoff,
  rotateSession, getHandoffContext, snapshotUsage, getUsageDelta
- Export DEFAULT_THRESHOLDS (maxRuns: 20, maxTokens: 500k, maxAgeHours: 72) and RotationThresholds type
- Adapter nativeContextManagement exemption skips rotation for Claude/Codex (D-04)
- Handoff markdown capped at 2000 chars with structured sections (D-06)
- Add 14 unit tests for service: rotation thresholds, adapter exemption, configurable
  thresholds, token accumulation, handoff generation, session rotation, usage delta

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
ZacharyZhang-NYandClaude Opus 4.6 committed 2026-03-26 22:56:18 -04:00
1 parent 2c671d0011
commit e54c632fd6
2 files changed
+659 -1

No files matched your search

+207
View File
@@ -0,0 +1,207 @@
import type { AppDatabase } from "./database";
import { getConnectorDefinition } from "./connectors";
import type { ConnectorId, TaskSessionRecord } from "@shared/types";
interface SessionManagementDependencies {
db: AppDatabase;
logger: { info: (...args: unknown[]) => void; warn: (...args: unknown[]) => void; error: (...args: unknown[]) => void };
}
// D-01/D-02: Exported for testing and external override. Conservative defaults.
export const DEFAULT_THRESHOLDS = {
maxRuns: 20,
maxTokens: 500_000,
maxAgeHours: 72,
};
// D-01: Type for per-adapter or company-level threshold overrides
export type RotationThresholds = Partial<typeof DEFAULT_THRESHOLDS>;
// D-04: Handoff size cap to prevent prompt bloat
const MAX_HANDOFF_LENGTH = 2000;
const MAX_HANDOFF_RUNS = 5;
export function createSessionManagementService({ db, logger }: SessionManagementDependencies) {
// D-08: Accumulate token usage on task_sessions row (called for ALL adapters per D-05)
function accumulateTokenUsage(sessionId: string, inputTokens: number, outputTokens: number, runId: string): void {
db.incrementSessionUsage(sessionId, inputTokens, outputTokens, runId);
}
// D-01: Any-of thresholds with optional overrides. D-04: Check nativeContextManagement.
// Per D-01: "Configurable per-adapter with company-level overrides" -- accepts optional
// thresholds parameter that merges with defaults. Callers can pass per-adapter or
// company-level overrides; omitted fields fall back to DEFAULT_THRESHOLDS.
function evaluateRotation(
session: TaskSessionRecord,
connectorId: string,
thresholdsOverride?: RotationThresholds,
): { shouldRotate: boolean; reason: string | null } {
// D-04: Check adapter exemption
try {
const connector = getConnectorDefinition(connectorId as ConnectorId);
if (connector.module.nativeContextManagement === true) {
return { shouldRotate: false, reason: null };
}
} catch {
// Unknown connector - proceed with rotation check
}
// Merge overrides with defaults (per D-01 configurability)
const thresholds = { ...DEFAULT_THRESHOLDS, ...thresholdsOverride };
// D-01: Any-of thresholds - first trigger wins
if (session.runCount >= thresholds.maxRuns) {
return { shouldRotate: true, reason: "max_runs" };
}
const totalTokens = session.totalInputTokens + session.totalOutputTokens;
if (totalTokens >= thresholds.maxTokens) {
return { shouldRotate: true, reason: "max_tokens" };
}
const ageMs = Date.now() - new Date(session.createdAt).getTime();
const ageHours = ageMs / (1000 * 60 * 60);
if (ageHours >= thresholds.maxAgeHours) {
return { shouldRotate: true, reason: "max_age_hours" };
}
return { shouldRotate: false, reason: null };
}
// D-06: Build handoff markdown from structured data, no AI call
function generateHandoff(taskId: string, session: TaskSessionRecord): string {
let task: { title: string; description: string } | null = null;
try {
task = db.getTask(taskId);
} catch {
task = null;
}
const recentRuns = db.getRecentRunsForTask(taskId, MAX_HANDOFF_RUNS);
const errorRuns = recentRuns.filter(r => r.errorMessage);
const activityEntries = db.getRecentActivityForTask(taskId, 10);
const decisionEntries = activityEntries.filter(a =>
a.action.includes("decision") || a.action.includes("approval") || a.action.includes("resolved")
);
const sections: string[] = [];
sections.push("# Session Handoff");
sections.push("");
if (task) {
sections.push("## Task");
sections.push(`**${task.title}**`);
const descSnippet = task.description.length > 200 ? task.description.slice(0, 200) + "..." : task.description;
sections.push(descSnippet);
sections.push("");
}
if (recentRuns.length > 0) {
sections.push("## Recent Run History");
for (const run of recentRuns) {
const summaryText = run.summary ? run.summary.slice(0, 100) : "(no summary)";
sections.push(`- [${run.status}] ${summaryText} (${run.finishedAt ?? "in progress"})`);
}
sections.push("");
}
if (errorRuns.length > 0) {
sections.push("## Errors Encountered");
for (const run of errorRuns) {
sections.push(`- ${run.errorMessage!.slice(0, 150)} (${run.finishedAt ?? "unknown"})`);
}
sections.push("");
}
if (decisionEntries.length > 0) {
sections.push("## Key Decisions");
for (const entry of decisionEntries.slice(0, 5)) {
sections.push(`- ${entry.detail.slice(0, 120)} (${entry.createdAt})`);
}
sections.push("");
}
sections.push(`*Session stats: ${session.runCount} runs, ${session.totalInputTokens + session.totalOutputTokens} total tokens*`);
let result = sections.join("\n");
// Cap handoff size per research pitfall #5
if (result.length > MAX_HANDOFF_LENGTH) {
result = result.slice(0, MAX_HANDOFF_LENGTH - 50) + "\n\n... (earlier content omitted)";
}
return result;
}
// Archive current session, generate handoff, create new session, update task pointer
function rotateSession(session: TaskSessionRecord): string {
const reason = "threshold_exceeded";
const handoff = generateHandoff(session.taskId, session);
logger.info(`[session] Rotating session ${session.id} for task ${session.taskId}: ${reason}`);
// 1. Archive old session
db.archiveSession(session.id, reason, handoff);
// 2. Create new session row with fresh counters, carry forward session_params
const newSessionId = db.createNewSession({
taskId: session.taskId,
companyId: session.companyId,
agentId: session.agentId,
adapterType: session.adapterType,
sessionParamsJson: session.sessionParamsJson,
sessionDisplayId: session.sessionDisplayId,
});
logger.info(`[session] New session ${newSessionId} created for task ${session.taskId}`);
return newSessionId;
}
// D-07: Read handoff from most recently rotated session for a task
function getHandoffContext(taskId: string): string | null {
// Find the most recent archived session for this task (the one just before current)
const row = db.queryAll<{ handoffMarkdown: string | null }>(
`SELECT handoff_markdown as handoffMarkdown FROM task_sessions
WHERE task_id = ? AND rotated_at IS NOT NULL
ORDER BY rotated_at DESC LIMIT 1`,
taskId
);
return row[0]?.handoffMarkdown ?? null;
}
// D-09: Snapshot token totals for delta tracking
function snapshotUsage(session: TaskSessionRecord): { inputTokens: number; outputTokens: number } {
return {
inputTokens: session.totalInputTokens,
outputTokens: session.totalOutputTokens,
};
}
// D-09: Compute delta from snapshot
function getUsageDelta(
sessionId: string,
snapshot: { inputTokens: number; outputTokens: number }
): { inputTokensDelta: number; outputTokensDelta: number } {
// Re-read current totals from DB
const current = db.queryAll<{ totalInputTokens: number; totalOutputTokens: number }>(
`SELECT total_input_tokens as totalInputTokens, total_output_tokens as totalOutputTokens
FROM task_sessions WHERE id = ?`,
sessionId
);
if (!current[0]) return { inputTokensDelta: 0, outputTokensDelta: 0 };
return {
inputTokensDelta: current[0].totalInputTokens - snapshot.inputTokens,
outputTokensDelta: current[0].totalOutputTokens - snapshot.outputTokens,
};
}
return {
accumulateTokenUsage,
evaluateRotation,
generateHandoff,
rotateSession,
getHandoffContext,
snapshotUsage,
getUsageDelta,
};
}
+452 -1
View File
@@ -14,7 +14,8 @@ vi.mock("electron", () => ({
import { AppDatabase } from "../../src/main/database";
import { getConnectorDefinition } from "../../src/main/connectors";
import type { ConnectorId } from "@shared/types";
import { createSessionManagementService, DEFAULT_THRESHOLDS } from "../../src/main/session-management-service";
import type { ConnectorId, TaskSessionRecord } from "@shared/types";
// ---------------------------------------------------------------------------
// Helpers — create a minimal company + agent + task scaffold
@@ -259,3 +260,453 @@ describe("adapter exemption", () => {
expect(def.module.nativeContextManagement).toBe(false);
});
});
// ===========================================================================
// Session management service tests
// ===========================================================================
describe("session management service", () => {
let tempDir: string;
let db: AppDatabase;
const mockLogger = {
info: vi.fn(),
warn: vi.fn(),
error: vi.fn(),
};
beforeEach(async () => {
tempDir = await mkdtemp(join(tmpdir(), "agentcompany-svc-"));
db = new AppDatabase(tempDir);
db.init();
mockLogger.info.mockClear();
mockLogger.warn.mockClear();
mockLogger.error.mockClear();
});
afterEach(async () => {
db.close();
await rm(tempDir, { recursive: true, force: true });
});
// -------------------------------------------------------------------------
// rotation thresholds
// -------------------------------------------------------------------------
describe("rotation thresholds", () => {
it("returns shouldRotate: false when below all thresholds", () => {
const svc = createSessionManagementService({ db, logger: mockLogger });
const session: TaskSessionRecord = {
id: "sess-1",
companyId: "co-1",
taskId: "task-1",
agentId: "agent-1",
adapterType: "gemini_local",
sessionParamsJson: null,
sessionDisplayId: null,
lastRunId: null,
lastError: null,
runCount: 5,
totalInputTokens: 10000,
totalOutputTokens: 5000,
rotatedAt: null,
rotationReason: null,
handoffMarkdown: null,
createdAt: new Date().toISOString(),
updatedAt: new Date().toISOString(),
};
const result = svc.evaluateRotation(session, "gemini_local");
expect(result.shouldRotate).toBe(false);
expect(result.reason).toBeNull();
});
it("returns shouldRotate: true with reason max_runs when run_count >= 20", () => {
const svc = createSessionManagementService({ db, logger: mockLogger });
const session: TaskSessionRecord = {
id: "sess-1",
companyId: "co-1",
taskId: "task-1",
agentId: "agent-1",
adapterType: "gemini_local",
sessionParamsJson: null,
sessionDisplayId: null,
lastRunId: null,
lastError: null,
runCount: 20,
totalInputTokens: 0,
totalOutputTokens: 0,
rotatedAt: null,
rotationReason: null,
handoffMarkdown: null,
createdAt: new Date().toISOString(),
updatedAt: new Date().toISOString(),
};
const result = svc.evaluateRotation(session, "gemini_local");
expect(result.shouldRotate).toBe(true);
expect(result.reason).toBe("max_runs");
});
it("returns shouldRotate: true with reason max_tokens when total tokens >= 500000", () => {
const svc = createSessionManagementService({ db, logger: mockLogger });
const session: TaskSessionRecord = {
id: "sess-1",
companyId: "co-1",
taskId: "task-1",
agentId: "agent-1",
adapterType: "gemini_local",
sessionParamsJson: null,
sessionDisplayId: null,
lastRunId: null,
lastError: null,
runCount: 3,
totalInputTokens: 300000,
totalOutputTokens: 200000,
rotatedAt: null,
rotationReason: null,
handoffMarkdown: null,
createdAt: new Date().toISOString(),
updatedAt: new Date().toISOString(),
};
const result = svc.evaluateRotation(session, "gemini_local");
expect(result.shouldRotate).toBe(true);
expect(result.reason).toBe("max_tokens");
});
it("returns shouldRotate: true with reason max_age_hours when session created > 72h ago", () => {
const svc = createSessionManagementService({ db, logger: mockLogger });
const oldDate = new Date(Date.now() - 73 * 60 * 60 * 1000).toISOString();
const session: TaskSessionRecord = {
id: "sess-1",
companyId: "co-1",
taskId: "task-1",
agentId: "agent-1",
adapterType: "gemini_local",
sessionParamsJson: null,
sessionDisplayId: null,
lastRunId: null,
lastError: null,
runCount: 1,
totalInputTokens: 100,
totalOutputTokens: 100,
rotatedAt: null,
rotationReason: null,
handoffMarkdown: null,
createdAt: oldDate,
updatedAt: oldDate,
};
const result = svc.evaluateRotation(session, "gemini_local");
expect(result.shouldRotate).toBe(true);
expect(result.reason).toBe("max_age_hours");
});
});
// -------------------------------------------------------------------------
// adapter exemption in service
// -------------------------------------------------------------------------
describe("adapter exemption in evaluateRotation", () => {
it("returns shouldRotate: false for nativeContextManagement=true even when thresholds exceeded", () => {
const svc = createSessionManagementService({ db, logger: mockLogger });
const session: TaskSessionRecord = {
id: "sess-1",
companyId: "co-1",
taskId: "task-1",
agentId: "agent-1",
adapterType: "claude_local",
sessionParamsJson: null,
sessionDisplayId: null,
lastRunId: null,
lastError: null,
runCount: 100, // Way over threshold
totalInputTokens: 1000000,
totalOutputTokens: 500000,
rotatedAt: null,
rotationReason: null,
handoffMarkdown: null,
createdAt: new Date(Date.now() - 200 * 60 * 60 * 1000).toISOString(), // Way old
updatedAt: new Date().toISOString(),
};
const result = svc.evaluateRotation(session, "claude_local");
expect(result.shouldRotate).toBe(false);
expect(result.reason).toBeNull();
});
});
// -------------------------------------------------------------------------
// configurable thresholds (D-01)
// -------------------------------------------------------------------------
describe("configurable thresholds (D-01)", () => {
it("evaluateRotation with custom thresholds override { maxRuns: 5 } triggers at 5 runs", () => {
const svc = createSessionManagementService({ db, logger: mockLogger });
const session: TaskSessionRecord = {
id: "sess-1",
companyId: "co-1",
taskId: "task-1",
agentId: "agent-1",
adapterType: "gemini_local",
sessionParamsJson: null,
sessionDisplayId: null,
lastRunId: null,
lastError: null,
runCount: 5,
totalInputTokens: 100,
totalOutputTokens: 100,
rotatedAt: null,
rotationReason: null,
handoffMarkdown: null,
createdAt: new Date().toISOString(),
updatedAt: new Date().toISOString(),
};
const result = svc.evaluateRotation(session, "gemini_local", { maxRuns: 5 });
expect(result.shouldRotate).toBe(true);
expect(result.reason).toBe("max_runs");
});
it("partial override { maxTokens: 100000 } uses override for tokens but defaults for runs and age", () => {
const svc = createSessionManagementService({ db, logger: mockLogger });
const session: TaskSessionRecord = {
id: "sess-1",
companyId: "co-1",
taskId: "task-1",
agentId: "agent-1",
adapterType: "gemini_local",
sessionParamsJson: null,
sessionDisplayId: null,
lastRunId: null,
lastError: null,
runCount: 10, // Below default 20
totalInputTokens: 60000,
totalOutputTokens: 40000, // Total 100000 = meets override threshold
rotatedAt: null,
rotationReason: null,
handoffMarkdown: null,
createdAt: new Date().toISOString(),
updatedAt: new Date().toISOString(),
};
const result = svc.evaluateRotation(session, "gemini_local", { maxTokens: 100000 });
expect(result.shouldRotate).toBe(true);
expect(result.reason).toBe("max_tokens");
});
it("omitting override uses DEFAULT_THRESHOLDS", () => {
const svc = createSessionManagementService({ db, logger: mockLogger });
const session: TaskSessionRecord = {
id: "sess-1",
companyId: "co-1",
taskId: "task-1",
agentId: "agent-1",
adapterType: "gemini_local",
sessionParamsJson: null,
sessionDisplayId: null,
lastRunId: null,
lastError: null,
runCount: 19, // Just below default 20
totalInputTokens: 200000,
totalOutputTokens: 200000, // Total 400000, below default 500000
rotatedAt: null,
rotationReason: null,
handoffMarkdown: null,
createdAt: new Date().toISOString(),
updatedAt: new Date().toISOString(),
};
const result = svc.evaluateRotation(session, "gemini_local");
expect(result.shouldRotate).toBe(false);
expect(result.reason).toBeNull();
// Verify DEFAULT_THRESHOLDS values
expect(DEFAULT_THRESHOLDS.maxRuns).toBe(20);
expect(DEFAULT_THRESHOLDS.maxTokens).toBe(500_000);
expect(DEFAULT_THRESHOLDS.maxAgeHours).toBe(72);
});
});
// -------------------------------------------------------------------------
// token accumulation
// -------------------------------------------------------------------------
describe("token accumulation", () => {
it("accumulateTokenUsage calls db.incrementSessionUsage with correct args", () => {
const { taskId, companyId, agentId } = scaffold(db);
const sessionId = db.createNewSession({
taskId,
companyId,
agentId,
adapterType: "claude_local",
sessionParamsJson: null,
sessionDisplayId: null,
});
const svc = createSessionManagementService({ db, logger: mockLogger });
svc.accumulateTokenUsage(sessionId, 5000, 2000, "run-abc");
const session = db.getActiveTaskSession(taskId);
expect(session!.runCount).toBe(1);
expect(session!.totalInputTokens).toBe(5000);
expect(session!.totalOutputTokens).toBe(2000);
expect(session!.lastRunId).toBe("run-abc");
});
});
// -------------------------------------------------------------------------
// handoff generation
// -------------------------------------------------------------------------
describe("handoff generation", () => {
it("generateHandoff produces markdown with task title and run history", () => {
const { taskId, companyId, agentId, workspaceId } = scaffold(db);
const sessionId = db.createNewSession({
taskId,
companyId,
agentId,
adapterType: "gemini_local",
sessionParamsJson: null,
sessionDisplayId: null,
});
// Create a completed run for history
const run = db.createRun({ companyId, taskId, agentId, workspaceId, connectorId: "gemini_local" as ConnectorId });
db.finishRun({
runId: run.id,
status: "completed",
summary: "Implemented initial logic",
errorMessage: null,
exitCode: 0,
signal: null,
model: "gemini-pro",
sessionDisplayId: null,
costUsd: 0.05,
unattributedCost: false,
});
// Increment session usage to have stats
db.incrementSessionUsage(sessionId, 10000, 5000, run.id);
const session = db.getActiveTaskSession(taskId)!;
const svc = createSessionManagementService({ db, logger: mockLogger });
const handoff = svc.generateHandoff(taskId, session);
expect(handoff).toContain("# Session Handoff");
expect(handoff).toContain("Implement session rotation"); // task title
expect(handoff).toContain("## Recent Run History");
expect(handoff).toContain("completed");
expect(handoff).toContain("15000 total tokens"); // 10000 + 5000
});
});
// -------------------------------------------------------------------------
// session rotation
// -------------------------------------------------------------------------
describe("session rotation", () => {
it("rotateSession archives old session, creates new, copies session_params", () => {
const { taskId, companyId, agentId } = scaffold(db);
const oldSessionId = db.createNewSession({
taskId,
companyId,
agentId,
adapterType: "gemini_local",
sessionParamsJson: JSON.stringify({ resumeId: "old-abc" }),
sessionDisplayId: "gemini-old",
});
// Add some usage
db.incrementSessionUsage(oldSessionId, 300000, 200000, "run-10");
const oldSession = db.getActiveTaskSession(taskId)!;
const svc = createSessionManagementService({ db, logger: mockLogger });
const newSessionId = svc.rotateSession(oldSession);
expect(newSessionId).not.toBe(oldSessionId);
// Old session should be archived
const archivedRows = db.queryAll<{ rotatedAt: string | null; rotationReason: string | null; handoffMarkdown: string | null }>(
"SELECT rotated_at as rotatedAt, rotation_reason as rotationReason, handoff_markdown as handoffMarkdown FROM task_sessions WHERE id = ?",
oldSessionId
);
expect(archivedRows[0].rotatedAt).toBeTruthy();
expect(archivedRows[0].rotationReason).toBe("threshold_exceeded");
expect(archivedRows[0].handoffMarkdown).toContain("# Session Handoff");
// New session should be active, with fresh counters and copied params
const newSession = db.getActiveTaskSession(taskId)!;
expect(newSession.id).toBe(newSessionId);
expect(newSession.runCount).toBe(0);
expect(newSession.totalInputTokens).toBe(0);
expect(newSession.totalOutputTokens).toBe(0);
expect(newSession.sessionParamsJson).toBe(JSON.stringify({ resumeId: "old-abc" }));
expect(newSession.sessionDisplayId).toBe("gemini-old");
});
});
// -------------------------------------------------------------------------
// handoff context retrieval
// -------------------------------------------------------------------------
describe("handoff context retrieval", () => {
it("getHandoffContext returns handoff_markdown from most recently rotated session", () => {
const { taskId, companyId, agentId } = scaffold(db);
const svc = createSessionManagementService({ db, logger: mockLogger });
// Create and rotate a session
const session1Id = db.createNewSession({
taskId,
companyId,
agentId,
adapterType: "gemini_local",
sessionParamsJson: null,
sessionDisplayId: null,
});
db.incrementSessionUsage(session1Id, 100000, 50000, "run-1");
const session1 = db.getActiveTaskSession(taskId)!;
svc.rotateSession(session1);
const handoff = svc.getHandoffContext(taskId);
expect(handoff).not.toBeNull();
expect(handoff).toContain("# Session Handoff");
});
it("getHandoffContext returns null when no rotated sessions exist", () => {
const { taskId, companyId, agentId } = scaffold(db);
db.createNewSession({
taskId,
companyId,
agentId,
adapterType: "gemini_local",
sessionParamsJson: null,
sessionDisplayId: null,
});
const svc = createSessionManagementService({ db, logger: mockLogger });
const handoff = svc.getHandoffContext(taskId);
expect(handoff).toBeNull();
});
});
// -------------------------------------------------------------------------
// usage delta (D-09)
// -------------------------------------------------------------------------
describe("usage delta", () => {
it("getUsageDelta returns token difference between snapshot and current totals", () => {
const { taskId, companyId, agentId } = scaffold(db);
const sessionId = db.createNewSession({
taskId,
companyId,
agentId,
adapterType: "gemini_local",
sessionParamsJson: null,
sessionDisplayId: null,
});
// Pre-populate some usage
db.incrementSessionUsage(sessionId, 1000, 500, "run-1");
const session = db.getActiveTaskSession(taskId)!;
const svc = createSessionManagementService({ db, logger: mockLogger });
// Snapshot current state
const snapshot = svc.snapshotUsage(session);
expect(snapshot.inputTokens).toBe(1000);
expect(snapshot.outputTokens).toBe(500);
// Accumulate more tokens
svc.accumulateTokenUsage(sessionId, 3000, 1500, "run-2");
// Delta should reflect new tokens only
const delta = svc.getUsageDelta(sessionId, snapshot);
expect(delta.inputTokensDelta).toBe(3000);
expect(delta.outputTokensDelta).toBe(1500);
});
});
});