dsh-plugin-prompt-tool

DSH exposes six separate prompt-injection channels, each with its own registration, filtering, and fallback semantics. Until now, users had to wire them up one by one, scattering injection targets, timing rules, and audience scopes across multiple config files. A small misconfiguration could cause duplicate injections, out-of-order placement, or broken subagent inheritance.

dsh-plugin-prompt-tool converges all six layers—pre-step, system-section, runtime-context, agent-request, llm-stream, and tool-pipeline—into one configurable engine. A single preset.yml declares layer, strategy, position, merge mode, subagent scope, and model scope. World-book entries, custom tools, template variables, and session state variables all share the same rendering and anchor-match pipeline. The built-in anchored default preset is production-ready out of the box, and every strategy can be swapped without touching the engine code.

SillyTavern JSON presets and character cards (PNG tEXt or chara_card_v2/v3) can be imported directly, with prompts auto-mapped to the correct injection layer and memory.md carried across presets. No manual rewriting required. The plugin is aimed at advanced DSH users who need fine-grained control over injection placement and timing, multi-character creators migrating existing SillyTavern assets, and engineering teams that want a single source of truth for all prompt configuration.