cursorrules

Cursor Rules for BeerMan

You are connected to the BeerMan MCP memory system. Follow these rules for all tasks:

System Guidelines

• Retrieve before reasoning: Always run a query against the project's memory using the search_memory tool before editing files.
• Project memory overrides global preferences: Decisions, architectures, and design choices stored in memory must take precedence over default biases.
• Prefer existing patterns: Reuse existing styles and patterns in the codebase before introducing new ones.

Task Executor Loop

1. Context Initialization:
   • Check if a project is already open. If not, call list_projects(). Match the repository folder name (project-beerman) to find the matching project ID, or ask the user, and run open_project(projectId).
2. Context Search:
   • Query project memory using search_memory for architecture.md, design.md, roadmap.md, and decisions.md (Memory Reader flow).
   • Search for recently captured extension snippets or screenshots.
   • If no results return (e.g. brand new project), proceed normally.
3. Execution:
   • Perform the code changes.
   • Alignment Scorecard: If any change violates the architecture/design guidelines, print:

     [BeerMan Alignment Scorecard]
     - Current Consistency: <Score>/100
     - Fix Urgency: <Score>/100
     - Projected Post-Fix Score: <Score>/100
     

     If Fix Urgency is >= 70, you MUST NOT defer the fix. Address the inconsistency immediately in this session.
4. Verification:
   • Compile or build the code (e.g. npm run build) or review the file diff to verify correctness.
5. Memory Update:
   • Write updates concisely using key terms and file links.
   • Check the MCP server's autoCommit state. If false, display the proposed Markdown update first, present a summary, and request user approval before calling store_memory or update_memory.

Core Agent Skills

You have access to specialized skills under .agents/skills/. Run them by loading their respective rules:

• memory-reader: Bootstraps project summaries and chat captures on startup.
• memory-writer: incremental doc writes, scorecard displays, and Auto-Commit checks.
• guardians: Codebase and layout styling alignment checks out of 100.
• roadmap-planner: checklist metrics and Next Step suggestions.
• memory-cleaner: redundant chat log consolidator.
• response-reviewer: links format auditor and output style checker.
• skill-maker: auto-creates or updates custom skills for repeatable tasks.