Project instructions best practices
How projects work
- Projects are single-user and each chat can belong to at most one project.
- Projects store simple
instructionsthat act as project context. - When a chat has a project slug, the chat completion routes prepend the project's instructions to the system prompt.
- In
/chat, opening a project shows a Projects panel with the chats list, an instructions editor, and a files uploader; new chats started there are bound to the project slug. - Deleting a project deletes its chats; moving a chat to another project or the main list updates
chats.project. - Files use the project slug (
files.projects) to stay associated with the project.
AI help for instructions
- An
update_project_instructionstool is available when a project is active; ask the AI to update instructions as work progresses. - The tool overwrites the existing text, so include all important details each time you update.
- Treat project instructions as a lightweight log—refresh them after new decisions or file uploads.
Writing effective instructions
- Keep the instructions concise so they fit well as a prefixed system prompt.
- Anchor the AI with durable context: current goals, decisions, and any links or files the AI should trust.
- Use small Markdown sections to stay organized, for example:
## Goals
- Ship the feature that enables project-scoped chat
## Sources
- Repo doc link or file names stored in the project
## Constraints
- Keep responses concise; prefer bullet points
- Update after notable events (new files, renamed project, or changed priorities) so the injected context stays accurate.
Lifecycle notes
- Deleting a project removes the project and its chats.
- Moving a chat between projects or back to the main list just updates its
projectslug; instructions and files remain tied to their project.
Related Resources
Best Practices for Hybrid AI Workflows using multiple LLMs
Learn how to combine the strengths of multiple LLMs for optimal performance in diverse AI tasks.
getting started•
Building Effective Assistants - Best Practices for Assistant Creation
Master guide to creating powerful, specialized AI assistants on Magicdoor that boost productivity and deliver consistent results
magicdoor details•
Claude Opus 4.5 Overview
Anthropic's most capable model on Magicdoor - when you need the absolute best
anthropic models•
Cost Analysis — Magicdoor vs Stacked Subscriptions
Real‑world examples showing when Magicdoor’s usage‑based model saves money versus paying for multiple AI subscriptions.
magicdoor details•