Show GN: ALIGNS.md - 초정렬이 오기 전 내 판단 기준과 취향을 에이전트와 자가 정렬해보세요

• ALIGNS.md is an open-source skill that enables AI agents to read conversation logs, commits, and PR comments to learn and retain user preferences.
• In a pilot test dated 2026-10-15, using ALIGNS.md with the align skill resulted in 0 additional replies across Workshop, Guide, and Demo tasks, compared to 4, 7, and 9 respectively for generic instructions.
• The tool separates project execution rules (AGENTS.md) from user expectations (ALIGNS.md), allowing agents to check their work against agreed-upon criteria and self-correct drift.

How it works

• Users install the skill via npx skills add hmmhmmhm/aligns.md --skill align and define priorities, decision boundaries, and things to avoid in a markdown file.
• The agent references ALIGNS.md during tasks to make decisions consistent with user intent, asking for clarification only when a choice falls outside agreed boundaries.
• The skill supports automatic updates for the method while preserving user-defined criteria, ensuring that tool updates do not overwrite personal alignment settings.

Team features

• An optional path (option 8) allows teams to collect approved decision examples from code reviews and agent conversations.
• These examples can be exported as training datasets for fine-tuning coding agents, with a workflow that includes team review and manual submission.