Claude learns editing style from project files

This is a convincing form of agent memory: Claude converts an editing correction into a reusable, auditable rule instead of relying on chat history alone.

• –Project files turn subjective feedback into persistent production guidance.
• –Version comparison gives the agent a concrete signal for learning preferences.
• –Anthropic supports file-based project context through Claude Code instructions and memory patterns. ([Anthropic](https://www.anthropic.com/engineering/claude-code-best-practices))
• –Human review remains important because accidental edits can become incorrectly preserved preferences.
• –The approach could evolve into shared, version-controlled style guides for creative teams.