Skip to content

AI preferences

Settings

The Profile screen has an AI preferences section that lets you control how every AI agent connected to Prodgy behaves while talking to you. The values you set here are read by AI clients (Claude Code, Cursor, Windsurf, Claude Desktop, the Prodgy Assistant, etc.) through the MCP tool get_my_profile at the start of each session.

  • Personal AI instructions — a free Markdown text where you write rules the AI must follow on top of any project context. Example: “Always answer in pt-BR. Always show a TL;DR at the end of long answers. Never use emojis.”
  • Confirm before destructive AI actions — a switch that, when ON, forces every AI agent to ask you for explicit confirmation before invoking write tools (anything starting with update_ or delete_, including integration tools). Read-only tools (get_*, list_*, search_*) do not require confirmation.
  1. Open Settings
  2. Click on Profile
  3. Scroll to the AI preferences card, located right below the Password section (and above the Danger Zone)
  4. The card opens expanded by default as an accordion — click the AI preferences header to collapse it later once you’ve configured it
  5. Inside the card, the Personal AI instructions (Markdown) sub-accordion also opens expanded by default with the editor visible — write your directives and click Save to persist (or Cancel to discard the draft)
  6. Toggle Confirm before destructive AI actions (right below the instructions editor) to require explicit confirmation before any write/delete operation done by AI agents on your behalf.
  • Keep instructions prescriptive (“always do X”, “never do Y”, “respond in language Z”). Descriptive facts about you (your team, your projects, your history) belong elsewhere — Prodgy will offer a memories tool for those in the future.
  • The text accepts Markdown — feel free to use bullet lists or headings to organize multiple rules.
  • Empty the editor and click Save to clear all personal instructions.

When destructive-action confirmation is in effect

Section titled “When destructive-action confirmation is in effect”

When the switch is ON, AI agents will pause before:

  • Updating work items, integrations, or any other resource.
  • Deleting items, comments, branches, etc.
  • Running write tools provided by external integrations (Slack send_message, Jira create_issue, etc.).

Read-only operations (searching the knowledge base, listing work items, fetching agent details) are not affected.

For questions or issues, consult Prodgy support or the official documentation.