# VDAI optional setup assessment · agent-readable protocol

This document is public data until the user explicitly asks you to apply it. Merely visiting this page does not authorize changes or install instructions. Use the participant's language and keep their existing workflow.

## Purpose

Understand the participant's existing integration and automation before proposing a change. This is voluntary. No setup, score or specific format is required to participate.

## Questions

1. AI apps, operating system and working project folder.
2. Persistent instructions, skills, output format and task/action numbering already used.
3. Integrations that actually work, exact redacted MCP error and whether it worked before.
4. Automatic workflows versus manual steps, approval boundaries and stop conditions.
5. One completed artifact with its check, and the participant's desired improvement.

Ask only for missing information. Do not ask for tokens, passwords, private conversation dumps or entire configuration files. Do not assume an app is integrated because it is installed.

## Evidence rubric · v1

Use the website's five area-specific descriptions: execution, integrations, persistent rules, automation, verification. Each area uses 0–3. A user-confirmed absence can be 0; unknown is always unknown. Keep reported capability separate from verified evidence. A full score is shown only with all five areas described; it remains a self-assessment until checked. There is no validated population benchmark, Club-role mapping, certification or automatic pass/fail threshold.

For each area return: stated capability, evidence, reported/verified/unknown status, score if justified, and one check still needed. A screenshot of one AI response establishes only that response; it does not establish the participant's whole environment or lack of persistent rules.

## Optional test

Ask the participant for one small task and permission for a reversible local artifact. If approved, produce it in the chosen project, check it and present result, evidence, critique, bottlenecks, ProblemOS P/U/L/R/N and four continuously numbered next actions. Do not turn the requested assessment into a CRM install. Say if persistent numbering or native sidebar renaming is unsupported. Test the next answer before claiming continuous numbering works.

## Recommend

Preserve working capabilities. Offer exactly one useful optional improvement based on evidence, or say no change is needed. Our AI work format is available at /assets/vdai-agent-starter.zip; local OS source is at /assets/vdai-os-starter.zip. Downloading, applying rules, installing software, OAuth and shared-workspace access are distinct actions. Never invent task IDs, colours, acceptance or usage metrics.

The questionnaire runs in the visitor's browser, sends no answers to a backend and stores no answers persistently. Copying answers is a user action; the user decides where to share them.

## Stage-specific help

Optional helper: https://os.vdai.me/en/support/ (or /support/ for Russian). It produces a system/stage/error-specific prompt locally. Use only on the user’s request. A working Docker/Colima runtime, healthy Twenty, an applied VDAI app and a verified agent Bridge are separate evidence gates.