Bolt: claude.md Project Context File Support

Persistent Project Context via claude.md

Bolt's December 15–January 23 release added support for claude.md files — a Markdown-based configuration format that provides AI agents with persistent, project-specific instructions without requiring manual re-entry each session.

The concept is borrowed from Claude Code's ecosystem, where CLAUDE.md files are placed at the root of a repository to give the model context about the project's structure, coding conventions, testing commands, and architectural decisions. Bolt adopted this standard, making claude.md a first-class citizen in its own agent workflow.

How It Works in Bolt

To use a claude.md file in Bolt, developers attach it to their project using the plus icon in the chatbox and selecting "Attach file." Once uploaded, Bolt's agent automatically detects it and uses it as the primary instruction source for the session — no additional configuration required.

The claude.md file acts as an entry point for agent instructions. It can reference other files (additional Markdown documents, text files) but the core instructions must live in the claude.md itself. The filename is not configurable — the agent specifically looks for claude.md.

Complementing Project Knowledge

Bolt already supported a Project Knowledge system where developers could manually add context through the UI. The claude.md format is designed to complement and, in many cases, replace that manual approach. Because it's a Markdown file, it's easier to include links, format structured information, and maintain multiple related files that build out a complete context picture.

For developers working across multiple AI tools — for example, using both Bolt and Claude Code in the same project — maintaining a shared claude.md file means both environments start from the same baseline understanding of the project, reducing inconsistencies between tools.

Practical Benefits

The addition of claude.md support is particularly meaningful for complex, long-running projects. Rather than spending tokens and prompts re-establishing project context at the start of every session, developers encode that context once and reference it consistently. Teams working on the same Bolt project can also align on a shared instruction set, ensuring the AI agent behaves consistently regardless of which team member is driving the session.