klingai-usage-analytics — скилл для Claude Code: как установить
Скилл klingai-usage-analytics настраивает учёт генераций видео в Kling AI, считая потраченные кредиты и ошибки по логам JSONL. Помогает следить за расходами на видео.
MIT 1 установок ★ 2 935 обновлён 2026-11-01
npx skills add jeremylongshore/tons-of-skills-marketplace --skill klingai-usage-analytics -a claude-codeЧто делает klingai-usage-analytics
Скилл добавляет в проект систему учёта операций видеогенерации Kling AI. Основа работы — журнал событий формата JSONL, куда класс KlingEventLogger записывает отправку задач, завершение, длительность и списанные кредиты. В лог намеренно запрещено писать промпты, лица, голоса, ссылки с подписью и персональные данные.
Класс UsageAnalytics собирает метрики за день: процент успешных генераций, среднее время рендеринга и статистику по моделям. Функция cost_analysis считает недельные траты, умножая сумму кредитов на фиксированные 0.14 доллара. Также доступна выгрузка событий в файл kling_usage.csv.
Какие задачи решает
- считать списанные кредиты и оценивать расходы в долларах
- отслеживать долю успешных генераций видео и типы ошибок
- группировать выполненные задачи по моделям Kling AI
- выгружать журнал генераций в файл CSV
Примеры запросов агенту
- покажи kling ai usage report за сегодня
- запусти cost analysis генераций klingai за последние 7 дней
- сделай экспорт klingai metrics в CSV
Как установить klingai-usage-analytics
1Через Skills CLI (Vercel)
npx skills add jeremylongshore/tons-of-skills-marketplace --skill klingai-usage-analytics -a claude-code2Вручную: положить файлы
~/.claude/skills/klingai-usage-analytics/SKILL.mdПапка скилла целиком; для одного проекта — .claude/skills/klingai-usage-analytics/
1Через Skills CLI (Vercel)
npx skills add jeremylongshore/tons-of-skills-marketplace --skill klingai-usage-analytics -a codex2Вручную: положить файлы
.agents/skills/klingai-usage-analytics/SKILL.mdДля всех проектов — ~/.agents/skills/
1Через Skills CLI (Vercel)
npx skills add jeremylongshore/tons-of-skills-marketplace --skill klingai-usage-analytics -a cursor2Вручную: положить файлы
.cursor/skills/klingai-usage-analytics/SKILL.mdCursor читает и .agents/skills/, и .claude/skills/
1GitHub CLI
gh skill install jeremylongshore/tons-of-skills-marketplace skills/.curated/klingai-usage-analytics2Вручную: положить файлы
.github/skills/klingai-usage-analytics/SKILL.md1Через Skills CLI (Vercel)
npx skills add jeremylongshore/tons-of-skills-marketplace --skill klingai-usage-analytics -a gemini-cli2Командой платформы
gemini skills install https://github.com/jeremylongshore/tons-of-skills-marketplace --path skills/.curated/klingai-usage-analyticsВручную: положить файлы
.devin/skills/klingai-usage-analytics/SKILL.mdСтарый путь .windsurf/skills/ тоже читается
Через Skills CLI (Vercel)
npx skills add jeremylongshore/tons-of-skills-marketplace --skill klingai-usage-analytics -a opencodeClaude.ai
Упакуйте папку в zip и загрузите в Settings → Capabilities → Skills
По шагам, с командами для каждой платформы: Как установить скилл в Claude Code: пять способов
Вопросы о klingai-usage-analytics
Как установить klingai-usage-analytics в Claude Code?
Выполните в терминале команду npx skills add jeremylongshore/tons-of-skills-marketplace --skill klingai-usage-analytics -a claude-code.
Что делает klingai-usage-analytics?
Он логирует вызовы Kling AI в файлы JSONL, строит отчёты по кредитам и экспортирует метрики в CSV.
Какие данные запрещено сохранять в логах по требованиям скилла?
В журнал нельзя записывать сами промпты, исходные медиафайлы, биометрию лиц и голосов, персональные данные и временные подписанные URL.
Исходный текст
SKILL.md · MIT
--- name: klingai-usage-analytics description: 'Build usage analytics and reporting for Kling AI video generation. Use when tracking patterns, analyzing costs, or building dashboards. Trigger with phrases like ''klingai analytics'', ''kling ai usage report'', ''klingai metrics'', ''video generation stats''. ' allowed-tools: Read, Write, Edit, Bash(npm:*), Grep version: 1.18.0 license: MIT author: Jeremy Longshore <[email protected]> tags: - saas - kling-ai - analytics - reporting compatibility: Designed for Claude Code --- # Kling AI Usage Analytics ## Overview Track video generation usage with structured logging, aggregate metrics, daily reports, and cost analysis. Built on JSONL event logs that can feed into any analytics platform. ## Event Logger ```python import json import time from datetime import datetime from pathlib import Path class KlingEventLogger: """Append-only JSONL event log for Kling AI operations.""" def __init__(self, log_dir: str = "logs"): self.log_dir = Path(log_dir) self.log_dir.mkdir(exist_ok=True) def _write(self, event: dict): date = datetime.utcnow().strftime("%Y-%m-%d") filepath = self.log_dir / f"kling-{date}.jsonl" event["timestamp"] = datetime.utcnow().isoformat() with open(filepath, "a") as f: f.write(json.dumps(event) + "\n") def log_submission(self, task_id, prompt, model, duration, mode): self._write({ "event": "task_submitted", "task_id": task_id, "model": model, "duration": int(duration), "mode": mode, "prompt_len": len(prompt), }) def log_completion(self, task_id, status, elapsed_sec, credits_used): self._write({ "event": "task_completed", "task_id": task_id, "status": status, "elapsed_sec": elapsed_sec, "credits_used": credits_used, }) def log_error(self, task_id, error_type, message): self._write({ "event": "task_error", "task_id": task_id, "error_type": error_type, "message": message[:200], }) ``` ## Analytics Aggregator ```python from collections import defaultdict class UsageAnalytics: """Aggregate metrics from JSONL event logs.""" def __init__(self, log_dir: str = "logs"): self.log_dir = Path(log_dir) def _read_events(self, date: str = None): pattern = f"kling-{date}.jsonl" if date else "kling-*.jsonl" events = [] for filepath in sorted(self.log_dir.glob(pattern)): with open(filepath) as f: for line in f: events.append(json.loads(line)) return events def daily_summary(self, date: str = None) -> dict: date = date or datetime.utcnow().strftime("%Y-%m-%d") events = self._read_events(date) submitted = [e for e in events if e["event"] == "task_submitted"] completed = [e for e in events if e["event"] == "task_completed"] errors = [e for e in events if e["event"] == "task_error"] succeeded = [e for e in completed if e["status"] == "succeed"] failed = [e for e in completed if e["status"] == "failed"] total_credits = sum(e.get("credits_used", 0) for e in completed) avg_elapsed = (sum(e["elapsed_sec"] for e in succeeded) / len(succeeded) if succeeded else 0) by_model = defaultdict(int) for e in submitted: by_model[e["model"]] += 1 return { "date": date, "total_submitted": len(submitted), "succeeded": len(succeeded), "failed": len(failed), "errors": len(errors), "success_rate": f"{len(succeeded) / max(len(completed), 1) * 100:.1f}%", "total_credits": total_credits, "avg_generation_sec": round(avg_elapsed), "by_model": dict(by_model), } def print_report(self, date: str = None): s = self.daily_summary(date) print(f"\n=== Kling AI Usage Report: {s['date']} ===") print(f"Submitted: {s['total_submitted']}") print(f"Succeeded: {s['succeeded']}") print(f"Failed: {s['failed']}") print(f"Success rate: {s['success_rate']}") print(f"Credits used: {s['total_credits']}") print(f"Avg time: {s['avg_generation_sec']}s") print(f"By model:") for model, count in s["by_model"].items(): print(f" {model}: {count}") ``` ## Cost Analysis ```python def cost_analysis(analytics: UsageAnalytics, days: int = 7): """Analyze cost trends over recent days.""" from datetime import timedelta daily_costs = [] for i in range(days): date = (datetime.utcnow() - timedelta(days=i)).strftime("%Y-%m-%d") summary = analytics.daily_summary(date) daily_costs.append({ "date": date, "credits": summary["total_credits"], "videos": summary["total_submitted"], "estimated_usd": summary["total_credits"] * 0.14, }) total_credits = sum(d["credits"] for d in daily_costs) total_videos = sum(d["videos"] for d in daily_costs) total_cost = sum(d["estimated_usd"] for d in daily_costs) print(f"\n=== {days}-Day Cost Summary ===") print(f"Total credits: {total_credits}") print(f"Total videos: {total_videos}") print(f"Est. cost: ${total_cost:.2f}") print(f"Avg/day: ${total_cost / days:.2f}") for d in daily_costs: print(f" {d['date']}: {d['credits']} credits, {d['videos']} videos, ${d['estimated_usd']:.2f}") ``` ## Export to CSV ```python import csv def export_usage_csv(analytics: UsageAnalytics, output: str = "kling_usage.csv"): events = analytics._read_events() with open(output, "w", newline="") as f: writer = csv.DictWriter(f, fieldnames=["timestamp", "event", "task_id", "model", "status", "credits_used", "elapsed_sec"]) writer.writeheader() for e in events: writer.writerow({k: e.get(k, "") for k in writer.fieldnames}) print(f"Exported {len(events)} events to {output}") ``` ## Prerequisites - An append-only event sink with restricted access, an explicit retention/deletion schedule, a timezone, and an approved pricing source for aggregate cost calculations. - Define an opaque run/task identifier and a schema that records operational facts only. Prompts, source media, faces/voices, signed URLs, personal data, credentials, and free-form provider messages are prohibited in logs. - Seed the parser and dashboard with synthetic events, and define anomaly thresholds, export destinations, and an owner who can pause generation when spend or policy signals drift. ## Instructions 1. Validate each event against the allowlisted schema before writing it. Normalize timestamps to the documented timezone and reject records with raw prompts, media, identities, or secrets. 2. Use opaque IDs to correlate submission, completion, error, and cleanup events. Deduplicate retries and preserve the original event time so totals cannot be inflated by a replay. 3. Aggregate by date, model, mode, status, and credit bucket. Apply the current approved rate only to totals, and keep dashboards and CSV exports at aggregate level. 4. Alert on failure-rate, latency, credit, retention, policy, and unexpected-destination thresholds. Pause new work for a confirmed anomaly, run a synthetic canary after remediation, and require owner approval before resuming. 5. Enforce retention deletion, verify the deletion receipt, and retain only a redacted aggregate report and incident/rollback reference. ## Output Produce an aggregate report containing date range, submitted/succeeded/failed counts, success rate, latency summary, credits, estimated cost range, model distribution, anomaly state, retention/deletion status, and owner approval or rollback state. It must not expose prompts, media, likenesses, audio, signed URLs, contact details, billing identifiers, or credentials. ## Error Handling - Quarantine malformed, duplicate, out-of-order, or schema-breaking events without counting them twice; report the aggregate gap and repair from an approved source. - If a log contains prohibited content, stop exports, restrict access, remove the offending record under the retention policy, and record only the redacted cleanup receipt. - If pricing is missing or stale, show credits without currency conversion and mark cost as unknown. If storage, deletion, or anomaly checks fail, pause dependent reporting and generation until an owner approves recovery. ## Examples A safe fixture can contain `run_id=synthetic-run-01`, `event=task_completed`, `model=kling-v2-5-turbo`, `status=succeed`, `credits_used=10`, `elapsed_sec=42`, and no prompt or URL. A seven-day report may publish counts and cost ranges to an internal dashboard only after `schema=pass`, `pii_scan=pass`, `retention=verified`, and `export=aggregate-only`. ## Resources - [Developer Portal](https://app.klingai.com/global/dev) - [Pricing Reference](https://app.klingai.com/global/dev/document-api/productBilling/prePaidResourcePackage)
© Jeremy Longshore · MIT · jeremylongshore/tons-of-skills-marketplace