progressive-disclosure

Long prompt documents exceeding 500 lines trigger the Lost-in-the-Middle phenomenon, where the LLM misses over 30% of middle-section instructions. This Skill audits oversized SKILL.md files and proposes splitting content into metadata, body, and references layers to mitigate the issue.

Progressive disclosure is a three-layer model that segments long prompt content into metadata, body, and references. Metadata is always loaded, the body loads under 500 lines, and references load on-demand, preventing the LLM from dropping middle instructions.

Run an audit to identify files exceeding the 500-line threshold. The Skill automatically proposes separating references into a dedicated directory and implements pointer-based loading strategies to ensure the main SKILL.md stays slim.

You need to separate references when your SKILL.md or long prompt management document exceeds 500 lines. The Skill enforces loading rules that keep metadata always loaded and push references to on-demand loading to avoid Lost-in-the-Middle issues.

Yes, the Skill applies a three-layer model with configurable thresholds for prompt safety. While 500 lines is the default limit for loading the body, you can adjust the threshold to segment oversized content and trigger references separation.

The Skill specifically targets Lost-in-the-Middle issues caused by document length rather than semantic complexity. It relies on strict loading rules and threshold configurations, so improperly segmenting references without pointer-based loading may still disrupt LLM context handling.