What problem does it solve? Long AI coding sessions waste tokens by re-reading large files and producing verbose answers. This Skill enforces a cache-first workflow that loads only an index plus a small number of targeted cache or memory files, keeping context small and responses short. ## Core Features & Use Cases - Capped context loading: Reads the manifest token policy, README index, codebase scan, and TODO first, then loads at most 2 codebase docs and 1 memory file, with grep-before-read rules for large files. - Token-budgeted responses: Caps replies at roughly 120 words in bullet form with path citations, and appends a token meter footer generated by the token_monitor.py script each turn. - Session refresh and chaining: Recommends starting a fresh thread at 100k context tokens and composes with chain, standup, and orchestrator skills for deeper work. - Use Case: At the start of every session in the ndestates-io repo, invoke this Skill so the assistant loads only the INDEX and relevant cache sections, then answers in short cited bullets instead of re-reading source files. ## Quick Start Start the session in cache-efficient mode and answer my questions using only the index and targeted cache files with short bulleted responses.
