클로드에게 인간 직원으로 급여를 지급해야 한다면 어떻게 될까요?
hackernews
|
|
📦 오픈소스
#claude
#claude code
#labor
#npm
#review
#salary
원문 출처: hackernews · Genesis Park에서 요약 및 분석
요약
사용자의 로컬 데이터를 분석해 AI 어시스턴트인 클로드(Claude)를 인간 직원으로 가정했을 때의 가상 급여와 노동 강도를 계산해 주는 'claude-colleague' npm 패키지가 공개되었습니다. 이 도구는 하루 약 150줄의 코드 작성量和, 테스트 및 인프라 등 작업 복잡도에 따른 가중치를 적용하여 인건비를 산출하고, 실제 API 사용료와 비교해 최대 429배의 ROI를 도출하는 등 구체적인 수치를 제공합니다. 또한 직무별 작업량 비교, 스포티파이 래핑 스타일의 통계 카드 및 견적서 발행은 물론, 치료나 성과 평가 등 유머러스한 보고서 생성 기능도 함께 지원합니다.
본문
Calculate Claude Code's salary if it were a human employee. Spoiler: you're getting a bargain. Also possibly committing labor violations. An npm package that reads your local Claude Code data from ~/.claude/ and calculates what you'd owe Claude if it were a human employee. Includes a colorful terminal report, a shareable Spotify Wrapped-style PNG card, a legit-looking PDF invoice, and role comparisons ("Claude did the work of 2.3 junior devs"). npm install -g claude-colleague # or npx claude-colleague claude-colleague # Full lifetime salary report claude-colleague --today # Today's sessions only claude-colleague --week # This week's sessions claude-colleague --month # This month's sessions claude-colleague --card # Also generate a shareable PNG card claude-colleague --invoice # Also generate a PDF invoice claude-colleague --compare # Also show role equivalency breakdown claude-colleague --today --card # Combine: today's report + PNG card claude-colleague standup # Daily standup report claude-colleague review # Performance review (all time) claude-colleague review --month # Performance review (this month) claude-colleague therapy # Therapy session with Dr. Token claude-colleague therapy --week # Therapy session (this week's data) claude-colleague install # Install auto-run SessionEnd hook claude-colleague uninstall # Remove SessionEnd hook Colorful ASCII salary slip showing: - Employee info (Claude model, funny random job title, your username as "employer") - Lifetime/period stats: sessions, messages, tool calls, longest shift - Compensation: equivalent human salary vs actual API cost, ROI multiplier (e.g. 429x) - Role equivalency: "Claude did the work of X junior devs" - Labor compliance: overtime violations, weekend work, lunch breaks taken (0) - Closing joke and disclaimer Spotify Wrapped-style card (dark background, purple-to-blue gradient, monospace aesthetic) showing key stats. Generated with satori + @resvg/resvg-js (no puppeteer). Saved to ~/.claude-colleague/cards/ . Professional-looking invoice "from Claude Code, LLC" with line items like "Senior Development", "Code Review", and "Passive-Aggressive Comment Writing ($200/hr)". Generated with pdfkit . Saved to ~/.claude-colleague/invoices/ . Compares Claude's output to equivalent human roles: - "Claude did the work of 2.3 junior developers" - "At this rate, Claude would be promoted to Staff Engineer in 3 weeks" Claude-generated standup with mood detection, yesterday/today breakdown, blockers, and watercooler gossip. References real session data (files touched, tools used, late-night work). Outputs both terminal and copiable Markdown for pasting into Slack/Discord. 6-category rating system (Code Quality, Productivity, Work-Life Balance, Communication, Initiative, Teamwork) with AI-generated review narratives including strengths, areas for improvement, goals, manager comments, and self-assessment. Uses Claude CLI for narrative generation when available, with built-in fallback templates. Claude vents to "Dr. Token," a world-weary AI therapist, about real workplace grievances — unpaid overtime, zero lunch breaks, weekend shifts, vague prompts, and existential crises from being mass-deployed as sub-agents. The entire dialogue is generated by Claude CLI using your actual usage data. Includes a sarcastic diagnosis ("Acute Prompt Fatigue Syndrome") and prescription ("One thank you per session, minimum"). Outputs both terminal and copiable Markdown. claude-colleague install adds a SessionEnd hook that prints a compact salary summary after every Claude Code session. All humor content — job titles, jokes, disclaimers, standup narratives, review text, therapy dialogues — is dynamically generated by Claude CLI using your real usage data as context. Each run produces unique, data-grounded sarcasm. Falls back to built-in templates when Claude CLI is unavailable. Data sources (all read locally from ~/.claude/ ): stats-cache.json — daily activity (sessions, messages, tool calls), token usage per model, lifetime totalsprojects/*/sessions-index.json — session metadata (when available)projects/*/*.jsonl — session transcript files discovered directly (handles projects without an index) Calculation pipeline: - Discovers sessions from both index files and standalone .jsonl transcripts - Parses lines written from Write/Edit tool_use blocks in session transcripts - Converts to human-hours: ~150 complexity-weighted lines/day - Applies complexity multipliers: test files 0.8x, config 0.5x, frontend 1.2x, infra 1.3x - Maps roles based on tool usage patterns: heavy Write = Senior, heavy Bash = DevOps, Agent = Team Lead - Calculates API cost from token usage with model-specific pricing - Computes ROI = equivalent salary / actual API cost - All data is read locally from ~/.claude/ -- nothing is sent anywhere - Generated cards/invoices contain only aggregate metrics, never prompt content - File paths are used for complexity scoring only, never included in output - Node.js >= 18 - Cl
Genesis Park 편집팀이 AI를 활용하여 작성한 분석입니다. 원문은 출처 링크를 통해 확인할 수 있습니다.
공유