What problem does it solve? Starting a new AI-assisted session on a large codebase wastes tokens and time when the assistant immediately greps and reads hundreds of source files. This Skill enforces a cache-first workflow so the assistant loads curated project documentation, TODO lists, and memory indexes before touching application source code. ## Core Features & Use Cases - Cache-First Session Start: Loads a fixed spine of files (codebase index, freshness scan, latest TODO, memory index) before any semantic search or grep. - Token Budget Enforcement: Honors a configurable cap on additional cache files via token_policy.max_cache_files_default in the project manifest. - Grep-Before-Read Discipline: Requires grepping headings of large cache files and reading only targeted sections instead of full files. - Use Case: A developer returning to the ndestates-io Laravel/Filament project starts a session; the assistant loads ARCHITECTURE.md and CONCERNS.md, confirms context, and only reads source files after the user confirms direction. ## Quick Start Ask the assistant to load the project cache first and summarize the current architecture and open concerns before making any code changes.
