context-gathering-patterns

Reduce token usage by applying a tiered reading approach that extracts frontmatter summaries first and stops reading when enough information is obtained. This prevents full-document reads and guarantees operation within token limits.

The best way to read design docs and PRDs is using a pyramid reading approach that starts with low-cost frontmatter, moves to summaries, and only reads detailed sections if necessary. This ensures you stay within token limits.

Frontmatter-driven context gathering works by extracting the agent_context from the BREAKDOWN frontmatter to quickly establish semantic context. It then reads only the minimal relevant sections needed to complete the task.

Yes, you can use tiered reading for planning and implementation across SDDs and ADRs. The pattern provides clear guidance for when to consult detailed docs, retros, and architecture decision records.

You should not use a minimal context read approach when a task requires understanding the entirety of a document. This pattern is designed for situations where only relevant sections are required to preserve performance and accuracy.

No, you do not need to parse full documents to get accurate semantic context. Frontmatter-driven summaries provide quick semantic context, allowing you to avoid full-document reads and operate within token limits.