The AI Coding Workflow That Finally Stopped Making Me Repeat myself

Описание и применение

Внедрение четырёх Markdown‑файлов (AGENTS.md, OVERVIEW.md, MEMORY.md, ERRORS.md) и навыков управления контекстом позволяет AI‑агенту сохранять знания о проекте, избегая повторения работы.

Задача

AI coding agents lacked persistent context across sessions, causing repetitive work and confusion about project history, decisions, and errors.

Результат

Agents now consistently reference project documentation, reducing repetitive work and ensuring decisions and errors are preserved for future sessions.

Рабочий процесс

• Read AGENTS.md for agent behavior rules
• Read OVERVIEW.md to understand current system state
• Check MEMORY.md for relevant decisions
• Check ERRORS.md for past painful issues
• Inspect code implementation
• Perform the task
• Update OVERVIEW.md if changes affect current state
• Add to MEMORY.md only if a new decision was made
• Add to ERRORS.md only if a failure is worth remembering
• Leave files unchanged otherwise

Доказательства

I created a project-context skill that teaches the agent how to use these files instead of relying on the agent to guess.

Источник

Afterward: Fix OVERVIEW.md if anything in it became untrue.

Источник