dsh-prompt-library
DSH (DeepSeek Harness) prompt library plugin: provides prompt management, AI polish and persona customization in the chat bar, helping you accumulate, reuse and continuously improve prompts.
The plugin treats conversation-level prompts as its core. Every feature is designed around this capability, while other auxiliary features are just a bonus. We warmly welcome users to share ideas and suggestions, and will keep iterating based on them.
Key Features
Prompt Library
- Manage frequently-used prompts (title + body + tags), with search, sorting, tag grouping and usage-count statistics
- The right panel shares the same style as the host's left sidebar, automatically squeezing and shrinking the chat area, freely toggling between expand/collapse
- Insert: appends to the existing content in the input box
- Overwrite: directly replaces the whole draft with this prompt
- Insert & Send: fills in template variables and sends with one click (uses the variable dialog when it contains
{{}}); outside # scenarios it is only available when the draft is empty, avoiding accidentally carrying existing content; when triggered by #, the trigger word is filtered out while the preceding text is kept and sent along
- Type
# to quickly trigger selection with real-time filtering
- The bottom shows the total number of tags and prompts in real time
Generate Skills
- Select prompts under "Prompt Library → Import/Export" to generate DSH Skills in batch
- AI generates an English skill name and description from the prompt content; the skill is written to
~/.dsh/skills/<name>/SKILL.md, and can be triggered by typing /skill-name in the chat box, or auto-matched by the model from the description
- A link between the prompt and the skill is created automatically; regenerating the same prompt overwrites the original skill directory instead of adding endlessly
- If the body contains
{{variables}}, the "placeholder auto-fill" capability is marked at generation time — when using the skill, AI infers and fills them automatically from the current semantic context, no manual input needed