context-optimizer

To reduce token waste in LLM context windows, audit your SKILL.md structures to identify redundantly loaded references and refactor outputs using a lean, pointer-based framing. This minimizes loaded content while maximizing actionable results.

A pointer pattern for context-efficient LLM skill design outputs a compact summary and a stable pointer to fetch detailed content on demand. This mechanism prevents token bloat by loading data only when explicitly required.

You audit an existing skill for token inefficiencies by running a REVIEW mode analysis that identifies bloated sections, redundantly loaded references, and token-waste patterns. This produces a lean routing description and a pointer-based data view.

Yes, you can design pointer-based SKILL.md structures from scratch using DESIGN mode. This creates lean, context-efficient designs that minimize loaded content and validate token budgets before implementation.

You should not use a pointer-based context optimization approach when your SKILL.md content is strictly sequential or cannot be partitioned into on-demand summaries. Pointer patterns require the ability to defer detailed content loading.