HN 표시: QuarterMaster – GitHub 활동에서 성과 리뷰 생성

Automate your quarterly, semi-annual, and annual performance reviews using GitHub contributions. Problem: Every quarter at work, I'd spend time manually digging through GitHub, copying PR titles and descriptions into whatever LLM was handy, and prompting it to turn everything into a performance review. It works, but it's tedious having to re-run scripts and re-prompt whatever LLM I'm given access to. Solution: QuarterMaster automates that whole process. It fetches your merged PRs, code review activity, and direct commits, then passes them to an LLM to generate a structured review. If your team has written down their goals, you can point it at a markdown file and it'll align your work to those goals automatically. If you have other accomplishments besides GitHub contributions, you can add those as well. npm install -g @imponenm/quartermaster Or run without installing: bunx @imponenm/quartermaster npx @imponenm/quartermaster - GitHub CLI ( gh ) — authenticated viagh auth login - An API key for your chosen LLM provider (see LLM Providers) git clone https://github.com/you/quartermaster cd quartermaster bun install QuarterMaster supports Anthropic, OpenAI, and Ollama. The provider is auto-detected from your environment, or you can specify it explicitly with --provider . echo "ANTHROPIC_API_KEY=sk-ant-..." >> .env Default model: claude-opus-4-6 echo "OPENAI_API_KEY=sk-..." >> .env Default model: gpt-4o Install and run Ollama, then pull a model: ollama pull qwen2.5-coder:7b Default model: qwen2.5-coder:7b — understands code context while producing readable prose. Requires ~8GB RAM. Other good options: mistral (better prose), qwen2.5:7b (general purpose). If both ANTHROPIC_API_KEY and OPENAI_API_KEY are set, Anthropic is used unless you pass --provider openai . quartermaster QuarterMaster will walk you through selecting a review period, optionally providing a goals file, and optionally providing an achievements file, then fetch your contributions and generate a review. --repo, -r GitHub repo to include (owner/repo). Repeatable. --org, -o Scope to a GitHub organization --provider, -p LLM provider: anthropic, openai, or ollama --model, -m Model override (uses provider default if omitted) # Interactive — auto-detects provider from env quartermaster # Scope to a specific repo quartermaster --repo acme-corp/api # Multiple repos quartermaster --repo acme-corp/api --repo acme-corp/web # Scope to an entire org quartermaster --org acme-corp # Use OpenAI instead of Anthropic quartermaster --provider openai # Use a specific model quartermaster --model claude-haiku-4-5-20251001 quartermaster --provider openai --model gpt-4-turbo quartermaster --provider ollama --model mistral | Source | How | |---|---| | Authored & merged PRs | gh search prs --author | | PRs you reviewed / commented on | gh search prs --commenter | | Direct commits to main/master | gh api repos/{owner}/{repo}/commits (only when --repo is specified) | Note: Direct commit fetching requires --repo to be set. Without it, there's no practical way to enumerate every repo you've pushed to. Create a markdown file listing your team's or company's goals, then provide its path when prompted: # Q1 2026 Goals ## Team Goals - Reduce API p99 latency by 20% - Launch self-serve onboarding ## Company Goals - Reach $10M ARR - Improve developer experience QuarterMaster will align your contributions to these goals in the generated review. Create a markdown file with accomplishments that don't show up in GitHub — mentoring, design work, on-call incidents, cross-team collaboration, presentations, interviews, etc. Provide its path when prompted after the goals step: # Q1 2026 Achievements - Mentored two new engineers through onboarding over 6 weeks - Led the incident post-mortem for the Feb 12 outage; published runbook improvements - Interviewed 8 candidates for the platform team - Presented the API reliability roadmap to the engineering org - On-call primary for 3 weeks; resolved 12 pages These are included in the generated review alongside your GitHub contributions and treated as equally valid evidence of impact. The review is saved as review-YYYY-MM-DD-YYYY-MM-DD.md in the current directory and printed to stdout (pipeable: bun run src/cli.tsx > review.md ). PRs welcome. Please open an issue before starting large changes. MIT — see LICENSE.