llm-prompt-template-override

What problem does it solve?

It prevents prompt-template injection from causing the LLM to wrongly treat provided documents as non-dependency reference material, which can lead to missing or hallucinated answers.

Core Features & Use Cases

• Dependency-aligned document injection: Wraps injected content in XML tags that align with the host template’s notion of dependencies/artifacts.
• Template-aware placeholder replacement: Uses StringReplace-style substitution to replace the host template’s empty-dependency literal (including Japanese/English variants) with the injected attached-documents section.
• Two-step fallback behavior: Falls back to safe prepend injection when placeholder replacement is unavailable or when dependency artifacts are already present.
• Instruction/intent signaling: Embeds explicit intent via HTML comments and structured document indexing to improve citation-like referencing and reduce response distortion.

Quick Start

Ask your system to inject external notes into the worker prompt by replacing the host template’s dependency placeholder with an <attached-documents> block, using a prepend-only fallback if the placeholder literal cannot be found.