Overview
The prompts shipped with Actovia, organized by purpose and generated from source.
Actovia ships task-specific prompts for conversations and background work. This catalogue contains 53 prompt templates and builders across 7 categories, generated from the desktop source. Each entry includes its purpose, full static text or construction function, and a link to the implementation.
| Category | Contents |
|---|---|
| Response formatting | Shared writing defaults for Chat and Philharmonic. |
| Chat | Chat identity, tools, personalization, interactive blocks and conversation titles. |
| Philharmonic | PM coordination, employee reports, recruitment and group context. |
| Memory & context | Long-term memory, relevance filtering and depth-aware conversation summaries. |
| Research & search | Deep Research prompts and search query expansion. |
| Discover | Daily personalized news queries and relevance checks. |
| Health | Daily Health notes and completed-period reports for iPhone. |
Reading the templates
Values in double braces are runtime substitutions, such as the current date, workspace, configured tools, selected skills or the user's message. Conditional template sections show both branches. Complex builders are shown as TypeScript so their optional instructions and precedence are visible. Static shared blocks are listed once and referenced by name elsewhere.
These pages contain built-in instructions only. They do not contain API keys, saved conversations, user memories, workspace files, custom instructions or configured team/employee prompts. Tool schemas and descriptions are a separate part of a model request; see Tools. Coding-agent CLIs supply their own system prompts; Actovia's task wrapper is documented under Philharmonic.
Response formatting
Chat and Philharmonic share writing defaults: paragraphs for connected explanations, short headings for substantial sections, bullets for parallel items, numbers for ordered steps and tables for comparisons. Code and paths keep their exact text. Explicit requested formats and structured task contracts take precedence. These are generation guidelines, not a transformation of saved replies, and model adherence can vary.
The policy was informed by Claude's published system prompt, Anthropic's formatting guidance, the Microsoft writing guides for headings and lists, and markdownlint's structural rules. We also inspected community Claude/ChatGPT prompt collections; those collections are not authenticated vendor specifications. Markdownlint can check syntax and conventions, but cannot decide whether an idea needs a heading or whether a table is easier to read than prose. It is not a runtime filter in Actovia.
Keeping this catalogue current
From the desktop repository, run:
bun scripts/export-built-in-prompts.mjs
bun scripts/export-built-in-prompts.mjs --checkThe exporter reads source text without importing the app or opening its database. It fails if a documented symbol moves or disappears; check mode also fails when the generated pages differ. When adding a new model task, add its prompt to the exporter's catalogue. Generated pages are committed in the documentation site repository alongside its navigation.