This package prepares a Research Square-oriented method article grounded in the context-forge repository.

Context Forge: A Lightweight Method for Diversity-Aware Context Packing and Prompt-Injection-Aware Retrieval

Research Square accepts method articles and data-oriented technical submissions. This package therefore frames context-forge as a concrete method paper with validation through repository tests and controlled examples rather than a broad opinion piece about prompting.

• Source repo: context-forge
• Package surface: JavaScript library plus CLI
• Evidence in repo:
  • document chunking
  • relevance scoring
  • diversity-aware ranking
  • edge-pin ordering
  • prompt injection and secret-like pattern scanning
  • unit tests for selection, risk detection, and heading preservation

• paper.md - full manuscript draft
• abstract.txt - platform-ready abstract
• keywords.txt - suggested keywords
• author-bio.txt - short author biography
• cover-note.txt - platform note
• paper.bib - lightweight bibliography
• submission-metadata.json - structured metadata
• assets/workflow-figure.svg - figure asset

• Primary platform target: Research Square
• Preferred article type: Method Article
• Keep the claims tied to implemented functions and test coverage