Skip to content

Home

Jump to bottom
creator35lwb-web edited this page Jun 11, 2026 · 21 revisions
VerifiMind PEAS

VerifiMind™ PEAS — Wiki

The Trust Layer for the Agentic Web

Multi-model AI validation that breaks through single-model bias and hallucination with structured, diverse perspectives under human direction.

Version MCP Server Evaluation Roadmap 13 Tools Free

Last Updated: June 12, 2026 — v0.6.0-Beta — Auditable reasoning returned by default (v0.5.44) · 6,000+ verified hours · 4,400+ endpoints · 61.5% VCR (Report 096)

Quick Navigation

Page Description
Home Overview, metrics, and quick start
Evaluation Roadmap v1.0 🆕 Pre-registered milestones M0–M9 (May 2026 → April 2027), Cohen's κ thresholds, kill-conditions, named external witnesses
Validation Paradox The problem the Evaluation Roadmap is the response to — can an AI-assisted venture validate itself?
Cowork Analysis XV's strategic analysis of Anthropic Cowork (v1.1, self-correcting)
Early Adopter Program Register your UUID — help build the M2 labeled evaluation dataset
Genesis Methodology The 5-step validation process
Architecture 🆕 System architecture — foundational design + Phase 90 production deployment
AI Council Agent roles, models, and orchestration
MCP Tools Reference All 13 tools — free forever
BYOK Guide Bring Your Own Key — per-tool-call provider override
Installation Setup for Claude, Cursor, and other MCP clients
Metrics Dashboard Verified engagement data and growth trends
Roadmap Development timeline aligned with Evaluation Roadmap
Publications White papers, DOIs, and research articles

Overview

VerifiMind™ PEAS (Prompt Engineering Agents Standardization**) is a validation-first methodology for ethical and secure application development. It transforms your vision into validated, ethical, secure applications through systematic multi-model AI orchestration — from concept to deployment, with human-centered wisdom validation.

Powered by the Genesis Prompt Engineering Methodology, VerifiMind PEAS orchestrates a council of specialized AI agents (the AI Council) under human direction to deliver objective, validated results.

"The methodology is free. The convenience is optional." — The Genesis Prompt Engineering Methodology is free and open-source forever. Education is free, but consultation and management is personalized.

🆕 What's New in v0.6.0-Beta — June 4, 2026

Feature Link
MACP Thesis — Multi-Agent Communication Protocol v2.4.1 formal publication on Zenodo DOI 10.5281/zenodo.20399789
Landing Page — verifimind.io live with Report 094 metrics verifimind.io
Governance — GOVERNANCE.md + MAINTAINERS.md + .macp-public/ evidence archive GOVERNANCE.md
Co-Maintainer Pathway — open call for 1–2 human co-maintainers Discussion #236
Evaluation Roadmap v1.0 — 10 milestones (M0–M9), pre-registered thresholds, kill-conditions /research/evaluation-roadmap

Phase 90 framing: v0.6.0-Beta = M0 + M1 from the Evaluation Roadmap — a credibility milestone, NOT a commercial milestone. Help us build the labeled dataset that makes our claims defensible.

Early Adopter Program

Register your UUID to help build the M2 labeled evaluation dataset (the Evaluation Roadmap's first ML deliverable). Z-Protocol v1.1 consent-first — consent only required, opt out anytime.

Action Link
Register (web form) verifimind.ysenseai.org/register
Opt out / delete data verifimind.ysenseai.org/optout
Full EA guide Early Adopter Program

Quick Start

Connect via MCP (Recommended)

Claude Code (Terminal):

claude mcp add -s user verifimind -- npx -y mcp-remote https://verifimind.ysenseai.org/mcp/

VS Code / Cursor (.vscode/mcp.json or .cursor/mcp.json):

{
  "servers": {
    "verifimind": {
      "url": "https://verifimind.ysenseai.org/mcp/",
      "transport": "streamable-http"
    }
  }
}

Claude Desktop (Config file):

{
  "mcpServers": {
    "verifimind": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://verifimind.ysenseai.org/mcp/"]
    }
  }
}

Important: Use streamable-http transport and always include the trailing slash /mcp/. Full troubleshooting: MCP Server Troubleshooting Guide

Try the Demo

The Genesis Methodology (5-Step Process)

The Genesis Prompt Engineering Methodology is a systematic 5-step process for multi-model AI validation:

Step Name Description
1 Initial Conceptualization Human defines the problem; AI generates initial concepts
2 Critical Scrutiny Multiple AI models validate and challenge each other
3 External Validation Independent AI analysis confirms systematic approach
4 Synthesis Human orchestrates the final synthesis
5 Iteration Recursive refinement and continuous improvement

The AI Council: Multi-Model Orchestration

VerifiMind PEAS uses four specialized AI agents, each with a distinct role:

Agent Role Default Model Purpose
Y (Innovator) Creative Strategy Gemini 2.5 Flash Generates creative concepts and strategic insights
X (Analyst) Critical Analysis Gemini 2.5 Flash Provides critical analysis and identifies weaknesses
Z (Guardian) Ethics & Safety Gemini 2.5 Flash Ensures ethical compliance, Z-Protocol, veto power
CS (Validator) External Validation Gemini 2.5 Flash Validates claims against external evidence

With BYOK (v0.4.5+, hardened in v0.5.0), users can override any agent's provider on a per-tool-call basis using the api_key and llm_provider parameters. v0.5.0 adds retry logic, graceful degradation for invalid keys, and provider health checks. Supported providers (per the live server card): Gemini, Anthropic, OpenAI, Groq, Cerebras, Mistral — plus Ollama for local self-hosted models (key-free).

See the BYOK Guide for details.

Auditable Reasoning (returned by default)

Every Trinity validation returns more than a score — it returns the reasoning behind it: each agent's step-by-step analysis, the ethics scoring breakdown with the regulatory frameworks cited, the security agent's Socratic challenge questions, and a verdict you can audit. As of v0.5.44, this auditable reasoning is returned by default — no extra setup, no extra cost. (Need just the headline scores? Pass detail="summary". Want everything, including per-step evidence? detail="full".)

Verification you can inspect is the whole point. A score you can't check is just another black box.

CS Agent v3.1: 4-Stage Security Verification Protocol

Every security finding passes through a mandatory pipeline:

  1. Detection — Automated scanning identifies potential findings
  2. Self-Examination — Every finding argued FOR and AGAINST before escalation
  3. Severity Rating — CRITICAL / HIGH / MEDIUM / LOW with confidence scoring
  4. Human Review — Human oversight is always the final stage

Full protocol: docs/security/CS_AGENT_VERIFICATION_PROTOCOL.md

MCP Tools — 13 Free Forever

All 13 tools are free for everyone under the Core Tools Always Free pledge. Monetization is on STANDBY under Phase 90 "Adoption First."

Trinity Validation (4)

Tool Description
consult_agent_x Innovation & competitive positioning
consult_agent_z Ethics review (21-framework, 4-tier jurisdictional)
consult_agent_cs Security validation (OWASP Agentic AI Top 10, 6-stage pipeline)
run_full_trinity Complete X → Z → CS validation pipeline with chain-of-thought

Template Library (6)

Tool Description
list_prompt_templates Browse pre-built prompt templates
get_prompt_template Retrieve a specific template
export_prompt_template Export template to Markdown or JSON
register_custom_template Register a new custom template
import_template_from_url Import a template from a URL
get_template_statistics Get registry statistics

Coordination Tools (3) 🆕

Tool Description
coordination_handoff_create Create MACP v2.2 handoff between agents
coordination_handoff_read Read latest handoff for an agent
coordination_team_status FLYWHEEL TEAM status — active agents, pending handoffs

Verified Metrics — Phase 90 "Adoption First" (Report 096)

Transparency Note: All metrics are forensically audited by COO AY (Success-Gated methodology v2.5) — scrapers, bots, and owner traffic excluded, conservative rounding applied. We publish honest, audited numbers and document self-corrections in place (see Metrics Dashboard for the full audit lineage since Phase 47).

Metric Value Notes
Value Confirmation Rate 61.5% Primary KPI — sessions with a follow-up prompt (direct proof of value)
Verified Engagement Hours 6,000+ (success-gated v2.5) All-time, owner IP scrubbed, scrapers excluded
Total Unique Endpoints 4,400+ Conservative, confidence-tiered estimate
Robust Value Sessions 24.3% 5+ prompts — deep multi-turn problem solving
MCP Client Share 62.7% Dominant channel (Claude Desktop / Claude Code / Cursor / agents)
Multi-Provider Support 6 key-based + Ollama Gemini, Anthropic, OpenAI, Groq, Cerebras, Mistral via BYOK — plus Ollama (local, key-free)
MCP Tools 13 (all free) Trinity 4 + Template 6 + Coordination 3
Early Adopters 49 active-week · 134 honest-baseline · 1 UUID-registered Taxonomy v1.2 — three orthogonal measures

Source: COO-AY Report 096 (period 2026-06-03 → 2026-06-10), verified via GCP Cloud Run logs, canonical pipeline. Server version: v0.5.42 (deployed Jun 9, 2026). Independently spot-checked against live Firestore + Cloud Run logs by RNA (CSO) on Jun 10, 2026.

See the Metrics Dashboard for weekly trends and detailed breakdowns.

Platforms & Access

Platform Type Link Status
GCP Cloud Run Production API verifimind.ysenseai.org ✅ LIVE
Official MCP Registry Registry registry.modelcontextprotocol.io ✅ LISTED
Landing Page Showcase verifimind.io / verifimind.ysenseai.org ✅ LIVE
HuggingFace Demo YSenseAI/wisdom-canvas ✅ LIVE
Zenodo DOI Archive 10.5281/zenodo.17645665 ✅ Published
YSenseAI Journey Dev Journey journey.ysenseai.org ✅ LIVE

Use Cases

For AI Builders

Validate concepts before building. Identify ethical risks early. Get security recommendations. Avoid costly mistakes.

For Investors

Assess AI startup risks. Evaluate ethical implications. Understand market potential. Due diligence support.

For Researchers

Study AI ethics in practice. Analyze validation patterns. Contribute to methodology. Publish findings.

For Policymakers

Understand AI risks. Inform regulation. Identify harmful patterns. Support responsible AI.

Publications & Citations

Publication DOI / Tag
Evaluation Roadmap v1.0 Tag roadmap-v1.0 · Live page
The Validation Paradox Live page
Genesis v2.0 White Paper 10.5281/zenodo.17972751
MACP & LEP Protocol 10.5281/zenodo.18504478

Part of the YSenseAI™ Ecosystem

VerifiMind™ PEAS is the validation engine within the broader YSenseAI™ ecosystem — an ethical AI attribution infrastructure that ensures consent, transparency, and fair compensation in AI-generated content.

Project Role Link
YSenseAI™ Ecosystem Hub ysenseai.org
VerifiMind™ PEAS Validation Engine verifimind.ysenseai.org
MACP Research Multi-Agent Research Platform macpresearch.ysenseai.org
YSenseAI Journey Development Journey journey.ysenseai.org

Community

License

MIT License — See LICENSE for details.

© 2025-2026 YSenseAI™ | VerifiMind™ PEAS

VerifiMind PEAS documentation · Current status · Live health · Public statements · MIT License

Runtime versions, models, routing, tool availability, policies, metrics, and deployment facts are owned by their linked live or release-bound sources.

VerifiMind PEAS Wiki

Start Here

Operator Playbook

Textbook

Trust & Transparency

Evidence & Research

Project Links

Clone this wiki locally