From v11.8.0, prompts are messages or instructions to be sent to an AI such as Claude or Gemini. Prompts are part of the Model Context Protocol (MCP) systems.
Typically, you use a prompt in the AI client by typing the command /tinderbox: followed by the prompt name. For example:
/tinderbox:translate
would invoke the prompt named "translate".
Tinderbox prompts are stored in the Tinderbox document in the container /Hints/AI/Prompts. They use the built-in prototype "An AI Prompt". A prompt has three components:
- Name: taken from the $Name of the prompt note. Names may contain spaces. Prompt names should not contain parentheses or forward slash '
/'. - Description: optional, taken from the $Subtitle of the prompt note. A brief description of the prompt, used to remind you what it does.
- Prompt: taken from the $Text of the prompt note. Describes what you want the AI to do.
Prompts may have optional arguments. These appear in the prompt enclosed by {{double curly braces}}. For example, a prompt named “translate”might be:
Show me a translation of the note {{note}} into language {{language}}, or into French is no language is specified. Do not change the note.Then, after you select /tinderbox: translate the system shows a reminder "(arguments: note, language)". You might then complete the command:
/tinderbox:translate "first draft" French
and the AI would be asked to find a note named "first draft" and to show you its French translation.
Some examples of prompts might include:
- Summarize — condense each note in a container into 2–3 sentences , showing you the summaries
- Critique — provide feedback on the argument in a note
- Questions — list questions the note raises but does not answer
- Additional Reading — suggest the best books or research papers related to a note
- Gaps — review the notes in a container, and identify notes that might be missing
- Write agent — given a plain-English description (
{{description}}), generate a Tinderbox agent query - Explain agent — given a note containing an agent query, explain what it does in plain language
- Suggest connections — given a note, suggest other notes it might be linked to
- Extract action items — pull out TODOs and next steps from meeting notes or journals
- Daily review — summarise all notes that have been modified today into a digest
See also—notes linking to here: