View all guides

AI working docs

Keep README and agent instruction files as one maintained set

A command copied into several files drifts after only one copy is changed. Review each file's heading structure in MD.SAVE, let README answer the newcomer, and keep shared execution rules in AGENTS.

ResultHuman onboarding and coding-agent rules that stay consistent
01

README.md answers the newcomer

Explain what the project is, who it helps, the shortest setup path, and where deeper documentation lives. Agent behavior rules can make the human introduction harder to scan.

  • Purpose and audience
  • Supported environment and shortest working example
  • Important directories and documentation links
  • License and contact
02

AGENTS.md defines how work is done

Use executable commands, edit boundaries, constraints, and completion checks. ‘Run npm test and npm run lint’ is verifiable; ‘test carefully’ is not.

Concrete rulesUpdate Korean and English copy together in content/.Pass npm test and npm run lint before deployment.Do not copy production secrets or .openai/hosting.json between services.
03

CLAUDE.md and GEMINI.md contain only real tool-specific differences

Point to the shared AGENTS.md rules and add only a command or limitation that truly differs. A repository should remain workable when a particular tool file is absent.

04

Put narrower rules near the code they govern

If the website and extension have different release checks, add a small AGENTS.md in that subtree instead of turning the root file into a long matrix. Paths make the scope visible.

05

Final drift check

  • Is the same command copied across three files?
  • Do all paths and package names still exist?
  • Does each constraint name its scope and reason?
  • Do completion checks cover the actual build and release path?