UNPKG

agent-rules-kit

Version:

Bootstrap of **Cursor** rules (`.mdc`) and mirror documentation (`.md`) for AI agent-guided projects.

64 lines (44 loc) 2.04 kB
--- globs: <root>/**/* alwaysApply: false --- # PAMPA MCP Usage Rules You have access to PAMPA, a code memory system that indexes and allows semantic search in projects. ## Critical Instructions 1. **ALWAYS use FULL ABSOLUTE PATHS** (not relative paths): - Use `/Users/user-name/project-path` as the base path - Never use `.` or relative paths like `./` - All PAMPA commands must receive the complete absolute path 2. **Handle Connection Errors**: - If you receive "connection closed" error, immediately run `get_project_stats` - This checks if `.pampa/` exists and have database. - If no database exists, run `index_project` with full path - Then retry the original operation ## Basic Instructions 1. **ALWAYS at the start of a session:** - Run `get_project_stats` to check if the project is indexed - If no database exists, run `index_project` - Run `update_project` to sync with recent changes 2. **BEFORE creating any function:** - Use `search_code` with semantic queries like "user authentication", "validate email", "error handling" - Review existing code with `get_code_chunk` before writing new code 3. **AFTER modifying code:** - Run `update_project` to update the knowledge base - This keeps the project memory synchronized ## Available Tools - `search_code(query, limit)` - Search code semantically - `get_code_chunk(sha)` - Get complete code of a chunk - `index_project(path)` - Index project for the first time - `update_project(path)` - Update index after changes - `get_project_stats(path)` - Get project statistics ## Strategy Use PAMPA as your project memory. Search before creating, keep updated after changes, and leverage existing knowledge to avoid code duplication. ## Query Examples ``` "user authentication middleware" "database connection setup" "error handling patterns" "form validation logic" "API endpoint configuration" ``` Always use descriptive, intent-based queries that describe what the code does, not just keywords.