dsh-custom-prompt

If you want the model to always follow a handful of standing rules—code style preferences, tone of response, known pitfalls to avoid—DSH does not ship a lightweight entry point for injecting custom prompt text. dsh-custom-prompt fills that gap: it carves out a dedicated section in the system prompt where you write plain-language rules that take effect on every subsequent inference step the moment you save.

Key capabilities: the section is registered in the global prompt layer, so it applies to every agent preset (standard, code, minimal, cordis, and any custom preset), not just a single mode. It sits at order 50, after the persona and before tool guidance, so the model confirms its role before reading your rules. Editing the text triggers a live re-registration with no profile restart; only code changes require one. The Settings page offers an enable toggle, a multiline editor, and Save / Reset buttons, with data persisted in settings.yaml so it survives restarts.

Built for developers and power users of DSH web profiles who want consistent behavioral guardrails across sessions and agent modes. A few lines of text are all it takes to make the model boundaries predictable.