feat(#163): add Agent Self-Diagnostics API endpoint
New GET /api/agents/[id]/diagnostics endpoint enabling agents to query their own performance data for self-optimization. Sections (selectable via ?section= query param): - summary: KPIs (throughput, error rate, activity count) - tasks: completion breakdown by status/priority, throughput/day - errors: error frequency by type, recent error details - activity: activity breakdown with hourly timeline - trends: current vs previous period comparison with auto-alerts - tokens: token usage by model with cost totals Features: - Scoped to requesting agent only (no cross-agent data access) - Configurable time window via ?hours= param (1-720h) - Automatic trend alerts for error spikes, throughput drops, stalls - Works with existing activities, tasks, and token_usage tables Fixes #163
This commit is contained in:
Bhavikprit
parent
4cb86bc80b
commit
047216dbe2
|
|
@ -0,0 +1,285 @@
|
|||
import { NextRequest, NextResponse } from 'next/server';
|
||||
import { getDatabase } from '@/lib/db';
|
||||
import { requireRole } from '@/lib/auth';
|
||||
import { logger } from '@/lib/logger';
|
||||
|
||||
/**
|
||||
* GET /api/agents/[id]/diagnostics - Agent Self-Diagnostics API
|
||||
*
|
||||
* Provides an agent with its own performance metrics, error analysis,
|
||||
* and trend data so it can self-optimize.
|
||||
*
|
||||
* Query params:
|
||||
* hours - Time window in hours (default: 24, max: 720 = 30 days)
|
||||
* section - Comma-separated sections to include (default: all)
|
||||
* Options: summary, tasks, errors, activity, trends, tokens
|
||||
*
|
||||
* Response includes:
|
||||
* summary - High-level KPIs (throughput, error rate, activity count)
|
||||
* tasks - Task completion breakdown by status and priority
|
||||
* errors - Error frequency, types, and recent error details
|
||||
* activity - Activity breakdown by type with hourly timeline
|
||||
* trends - Multi-period comparison for trend detection
|
||||
* tokens - Token usage by model with cost estimates
|
||||
*/
|
||||
export async function GET(
|
||||
request: NextRequest,
|
||||
{ params }: { params: Promise<{ id: string }> }
|
||||
) {
|
||||
const auth = requireRole(request, 'viewer');
|
||||
if ('error' in auth) return NextResponse.json({ error: auth.error }, { status: auth.status });
|
||||
|
||||
try {
|
||||
const db = getDatabase();
|
||||
const resolvedParams = await params;
|
||||
const agentId = resolvedParams.id;
|
||||
const workspaceId = auth.user.workspace_id ?? 1;
|
||||
|
||||
// Resolve agent by ID or name
|
||||
let agent: any;
|
||||
if (/^\d+$/.test(agentId)) {
|
||||
agent = db.prepare('SELECT id, name, role, status, last_seen, created_at FROM agents WHERE id = ? AND workspace_id = ?').get(Number(agentId), workspaceId);
|
||||
} else {
|
||||
agent = db.prepare('SELECT id, name, role, status, last_seen, created_at FROM agents WHERE name = ? AND workspace_id = ?').get(agentId, workspaceId);
|
||||
}
|
||||
|
||||
if (!agent) {
|
||||
return NextResponse.json({ error: 'Agent not found' }, { status: 404 });
|
||||
}
|
||||
|
||||
const { searchParams } = new URL(request.url);
|
||||
const hours = Math.min(Math.max(parseInt(searchParams.get('hours') || '24', 10) || 24, 1), 720);
|
||||
const sectionParam = searchParams.get('section') || 'summary,tasks,errors,activity,trends,tokens';
|
||||
const sections = new Set(sectionParam.split(',').map(s => s.trim()));
|
||||
|
||||
const now = Math.floor(Date.now() / 1000);
|
||||
const since = now - hours * 3600;
|
||||
|
||||
const result: Record<string, any> = {
|
||||
agent: { id: agent.id, name: agent.name, role: agent.role, status: agent.status },
|
||||
timeframe: { hours, since, until: now },
|
||||
};
|
||||
|
||||
if (sections.has('summary')) {
|
||||
result.summary = buildSummary(db, agent.name, workspaceId, since);
|
||||
}
|
||||
|
||||
if (sections.has('tasks')) {
|
||||
result.tasks = buildTaskMetrics(db, agent.name, workspaceId, since);
|
||||
}
|
||||
|
||||
if (sections.has('errors')) {
|
||||
result.errors = buildErrorAnalysis(db, agent.name, workspaceId, since);
|
||||
}
|
||||
|
||||
if (sections.has('activity')) {
|
||||
result.activity = buildActivityBreakdown(db, agent.name, workspaceId, since);
|
||||
}
|
||||
|
||||
if (sections.has('trends')) {
|
||||
result.trends = buildTrends(db, agent.name, workspaceId, hours);
|
||||
}
|
||||
|
||||
if (sections.has('tokens')) {
|
||||
result.tokens = buildTokenMetrics(db, agent.name, workspaceId, since);
|
||||
}
|
||||
|
||||
return NextResponse.json(result);
|
||||
} catch (error) {
|
||||
logger.error({ err: error }, 'GET /api/agents/[id]/diagnostics error');
|
||||
return NextResponse.json({ error: 'Failed to fetch diagnostics' }, { status: 500 });
|
||||
}
|
||||
}
|
||||
|
||||
/** High-level KPIs */
|
||||
function buildSummary(db: any, agentName: string, workspaceId: number, since: number) {
|
||||
const tasksDone = (db.prepare(
|
||||
`SELECT COUNT(*) as c FROM tasks WHERE assigned_to = ? AND workspace_id = ? AND status = 'done' AND updated_at >= ?`
|
||||
).get(agentName, workspaceId, since) as any).c;
|
||||
|
||||
const tasksTotal = (db.prepare(
|
||||
`SELECT COUNT(*) as c FROM tasks WHERE assigned_to = ? AND workspace_id = ?`
|
||||
).get(agentName, workspaceId) as any).c;
|
||||
|
||||
const activityCount = (db.prepare(
|
||||
`SELECT COUNT(*) as c FROM activities WHERE actor = ? AND workspace_id = ? AND created_at >= ?`
|
||||
).get(agentName, workspaceId, since) as any).c;
|
||||
|
||||
const errorCount = (db.prepare(
|
||||
`SELECT COUNT(*) as c FROM activities WHERE actor = ? AND workspace_id = ? AND created_at >= ? AND type LIKE '%error%'`
|
||||
).get(agentName, workspaceId, since) as any).c;
|
||||
|
||||
const errorRate = activityCount > 0 ? Math.round((errorCount / activityCount) * 10000) / 100 : 0;
|
||||
|
||||
return {
|
||||
tasks_completed: tasksDone,
|
||||
tasks_total: tasksTotal,
|
||||
activity_count: activityCount,
|
||||
error_count: errorCount,
|
||||
error_rate_percent: errorRate,
|
||||
};
|
||||
}
|
||||
|
||||
/** Task completion breakdown */
|
||||
function buildTaskMetrics(db: any, agentName: string, workspaceId: number, since: number) {
|
||||
const byStatus = db.prepare(
|
||||
`SELECT status, COUNT(*) as count FROM tasks WHERE assigned_to = ? AND workspace_id = ? GROUP BY status`
|
||||
).all(agentName, workspaceId) as Array<{ status: string; count: number }>;
|
||||
|
||||
const byPriority = db.prepare(
|
||||
`SELECT priority, COUNT(*) as count FROM tasks WHERE assigned_to = ? AND workspace_id = ? GROUP BY priority`
|
||||
).all(agentName, workspaceId) as Array<{ priority: string; count: number }>;
|
||||
|
||||
const recentCompleted = db.prepare(
|
||||
`SELECT id, title, priority, updated_at FROM tasks WHERE assigned_to = ? AND workspace_id = ? AND status = 'done' AND updated_at >= ? ORDER BY updated_at DESC LIMIT 10`
|
||||
).all(agentName, workspaceId, since) as any[];
|
||||
|
||||
// Estimate throughput: tasks completed per day in the window
|
||||
const windowDays = Math.max((Math.floor(Date.now() / 1000) - since) / 86400, 1);
|
||||
const completedInWindow = recentCompleted.length;
|
||||
const throughputPerDay = Math.round((completedInWindow / windowDays) * 100) / 100;
|
||||
|
||||
return {
|
||||
by_status: Object.fromEntries(byStatus.map(r => [r.status, r.count])),
|
||||
by_priority: Object.fromEntries(byPriority.map(r => [r.priority, r.count])),
|
||||
recent_completed: recentCompleted,
|
||||
throughput_per_day: throughputPerDay,
|
||||
};
|
||||
}
|
||||
|
||||
/** Error frequency and analysis */
|
||||
function buildErrorAnalysis(db: any, agentName: string, workspaceId: number, since: number) {
|
||||
const errorActivities = db.prepare(
|
||||
`SELECT type, COUNT(*) as count FROM activities WHERE actor = ? AND workspace_id = ? AND created_at >= ? AND (type LIKE '%error%' OR type LIKE '%fail%') GROUP BY type ORDER BY count DESC`
|
||||
).all(agentName, workspaceId, since) as Array<{ type: string; count: number }>;
|
||||
|
||||
const recentErrors = db.prepare(
|
||||
`SELECT id, type, description, data, created_at FROM activities WHERE actor = ? AND workspace_id = ? AND created_at >= ? AND (type LIKE '%error%' OR type LIKE '%fail%') ORDER BY created_at DESC LIMIT 20`
|
||||
).all(agentName, workspaceId, since) as any[];
|
||||
|
||||
return {
|
||||
by_type: errorActivities,
|
||||
total: errorActivities.reduce((sum, e) => sum + e.count, 0),
|
||||
recent: recentErrors.map(e => ({
|
||||
...e,
|
||||
data: e.data ? JSON.parse(e.data) : null,
|
||||
})),
|
||||
};
|
||||
}
|
||||
|
||||
/** Activity breakdown with hourly timeline */
|
||||
function buildActivityBreakdown(db: any, agentName: string, workspaceId: number, since: number) {
|
||||
const byType = db.prepare(
|
||||
`SELECT type, COUNT(*) as count FROM activities WHERE actor = ? AND workspace_id = ? AND created_at >= ? GROUP BY type ORDER BY count DESC`
|
||||
).all(agentName, workspaceId, since) as Array<{ type: string; count: number }>;
|
||||
|
||||
const timeline = db.prepare(
|
||||
`SELECT (created_at / 3600) * 3600 as hour_bucket, COUNT(*) as count FROM activities WHERE actor = ? AND workspace_id = ? AND created_at >= ? GROUP BY hour_bucket ORDER BY hour_bucket ASC`
|
||||
).all(agentName, workspaceId, since) as Array<{ hour_bucket: number; count: number }>;
|
||||
|
||||
return {
|
||||
by_type: byType,
|
||||
timeline: timeline.map(t => ({
|
||||
timestamp: t.hour_bucket,
|
||||
hour: new Date(t.hour_bucket * 1000).toISOString(),
|
||||
count: t.count,
|
||||
})),
|
||||
};
|
||||
}
|
||||
|
||||
/** Multi-period trend comparison for anomaly/trend detection */
|
||||
function buildTrends(db: any, agentName: string, workspaceId: number, hours: number) {
|
||||
const now = Math.floor(Date.now() / 1000);
|
||||
|
||||
// Compare current period vs previous period of same length
|
||||
const currentSince = now - hours * 3600;
|
||||
const previousSince = currentSince - hours * 3600;
|
||||
|
||||
const periodMetrics = (since: number, until: number) => {
|
||||
const activities = (db.prepare(
|
||||
`SELECT COUNT(*) as c FROM activities WHERE actor = ? AND workspace_id = ? AND created_at >= ? AND created_at < ?`
|
||||
).get(agentName, workspaceId, since, until) as any).c;
|
||||
|
||||
const errors = (db.prepare(
|
||||
`SELECT COUNT(*) as c FROM activities WHERE actor = ? AND workspace_id = ? AND created_at >= ? AND created_at < ? AND (type LIKE '%error%' OR type LIKE '%fail%')`
|
||||
).get(agentName, workspaceId, since, until) as any).c;
|
||||
|
||||
const tasksCompleted = (db.prepare(
|
||||
`SELECT COUNT(*) as c FROM tasks WHERE assigned_to = ? AND workspace_id = ? AND status = 'done' AND updated_at >= ? AND updated_at < ?`
|
||||
).get(agentName, workspaceId, since, until) as any).c;
|
||||
|
||||
return { activities, errors, tasks_completed: tasksCompleted };
|
||||
};
|
||||
|
||||
const current = periodMetrics(currentSince, now);
|
||||
const previous = periodMetrics(previousSince, currentSince);
|
||||
|
||||
const pctChange = (cur: number, prev: number) => {
|
||||
if (prev === 0) return cur > 0 ? 100 : 0;
|
||||
return Math.round(((cur - prev) / prev) * 10000) / 100;
|
||||
};
|
||||
|
||||
return {
|
||||
current_period: { since: currentSince, until: now, ...current },
|
||||
previous_period: { since: previousSince, until: currentSince, ...previous },
|
||||
change: {
|
||||
activities_pct: pctChange(current.activities, previous.activities),
|
||||
errors_pct: pctChange(current.errors, previous.errors),
|
||||
tasks_completed_pct: pctChange(current.tasks_completed, previous.tasks_completed),
|
||||
},
|
||||
alerts: buildTrendAlerts(current, previous),
|
||||
};
|
||||
}
|
||||
|
||||
/** Generate automatic alerts from trend data */
|
||||
function buildTrendAlerts(current: { activities: number; errors: number; tasks_completed: number }, previous: { activities: number; errors: number; tasks_completed: number }) {
|
||||
const alerts: Array<{ level: string; message: string }> = [];
|
||||
|
||||
// Error rate spike
|
||||
if (current.errors > 0 && previous.errors > 0) {
|
||||
const errorIncrease = (current.errors - previous.errors) / previous.errors;
|
||||
if (errorIncrease > 0.5) {
|
||||
alerts.push({ level: 'warning', message: `Error count increased ${Math.round(errorIncrease * 100)}% vs previous period` });
|
||||
}
|
||||
} else if (current.errors > 3 && previous.errors === 0) {
|
||||
alerts.push({ level: 'warning', message: `New error pattern: ${current.errors} errors (none in previous period)` });
|
||||
}
|
||||
|
||||
// Throughput drop
|
||||
if (previous.tasks_completed > 0 && current.tasks_completed === 0) {
|
||||
alerts.push({ level: 'info', message: 'No tasks completed in current period (possible stall)' });
|
||||
} else if (previous.tasks_completed > 2 && current.tasks_completed < previous.tasks_completed * 0.5) {
|
||||
alerts.push({ level: 'info', message: `Task throughput dropped ${Math.round((1 - current.tasks_completed / previous.tasks_completed) * 100)}%` });
|
||||
}
|
||||
|
||||
// Activity drop (possible offline)
|
||||
if (previous.activities > 5 && current.activities < previous.activities * 0.25) {
|
||||
alerts.push({ level: 'warning', message: `Activity dropped ${Math.round((1 - current.activities / previous.activities) * 100)}% — agent may be stalled` });
|
||||
}
|
||||
|
||||
return alerts;
|
||||
}
|
||||
|
||||
/** Token usage by model */
|
||||
function buildTokenMetrics(db: any, agentName: string, workspaceId: number, since: number) {
|
||||
try {
|
||||
// session_id on token_usage may store agent name or session key
|
||||
const byModel = db.prepare(
|
||||
`SELECT model, SUM(input_tokens) as input_tokens, SUM(output_tokens) as output_tokens, COUNT(*) as request_count FROM token_usage WHERE session_id = ? AND workspace_id = ? AND created_at >= ? GROUP BY model ORDER BY (input_tokens + output_tokens) DESC`
|
||||
).all(agentName, workspaceId, since) as Array<{ model: string; input_tokens: number; output_tokens: number; request_count: number }>;
|
||||
|
||||
const total = byModel.reduce((acc, r) => ({
|
||||
input_tokens: acc.input_tokens + r.input_tokens,
|
||||
output_tokens: acc.output_tokens + r.output_tokens,
|
||||
requests: acc.requests + r.request_count,
|
||||
}), { input_tokens: 0, output_tokens: 0, requests: 0 });
|
||||
|
||||
return {
|
||||
by_model: byModel,
|
||||
total,
|
||||
};
|
||||
} catch {
|
||||
// token_usage table may not exist
|
||||
return { by_model: [], total: { input_tokens: 0, output_tokens: 0, requests: 0 } };
|
||||
}
|
||||
}
|
||||
Loading…
Reference in New Issue