kleosr/cursorkleosr

What it solves

It solves the problem of "agent amnesia" in AI-powered IDEs like Cursor, where agents often start every session from zero or require the user to manually provide context and state files.

How it works

The project uses a specific Cursor rule (.cursor/rules/workflow-memory.mdc) set to alwaysApply: true, which forces the AI agent to follow a strict operating loop in every session. The agent automatically reads two root markdown files—workflow_state.md (for current progress, plans, and blockers) and project_config.md (for goals and tech stack)—before taking action. After completing work, the agent writes the updated state back to the markdown files, ensuring a persistent, diffable memory that exists within the repository itself.

Who it’s for

Developers using the Cursor IDE who want their AI agents to maintain a consistent, long-term memory of project goals and current progress without manual prompting.

Highlights

• Always-on memory: Uses Cursor rules to automate state retrieval and updates.
• Zero dependencies: Relies entirely on plain markdown files, making it editor-agnostic and easy to version control.
• Structured workflow: Implements a clear ANALYZE → PREPARE → IMPLEMENT → VALIDATE loop.
• Transparent state: Project progress and configurations are stored in human-readable, greppable text files.