Vibe Coding
The Samaritan Project
Agentic OSINT platform built with 🦞 OpenClaw and running on ParrotOS.
The Samaritan Project
The Problem: Public record research is slow by design. Every time you need a background check, want to vet a business partner, or need to understand who funded a piece of legislation, the information exists but it is scattered across dozens of sources. Pulling it together one search at a time takes hours, and most people never finish. By the time a complete picture forms, the window to act on it has already closed.
The Solution: The Samaritan Project is a vibe-coded, multi-agent OSINT investigation platform built on OpenClaw. It runs on dedicated bare-metal hardware and handles the full research workflow automatically. You give it a name, a bill number, a business, or a phone number, and it builds a structured intelligence profile using verified sources across background checks, skip tracing, political research, financial disclosure, and more. Samaritan does the work. You review the output.
✨ Tech Stack
- Parrot OS Linux (bare-metal)
- OpenClaw (multi-agent architecture)
- Claude Code (primary vibe-coding environment)
- GPT-5.3 Codex (primary model)
- Claude Sonnet (fallback / behavioral analysis)
- Qwen3 via LM Studio (air-gapped local backup)
- Python
- GitHub
✨ Features
- Multi-Agent Architecture — Seer command layer routes tasks to specialized agents covering background checks, skip tracing, personal safety vetting, political research, company investigation, and signals intelligence
- WeThePeople Agent — Ingests a bill number and returns a complete legislative intelligence profile including sponsors, campaign finance, donor cross-reference, and conflict-of-interest flags
- OSINT Source Library — Nearly 2,000 curated, categorized intelligence sources with agent auto-discovery baked in
- Fail-over Model Chain — Three-model redundancy (cloud primary, cloud fallback, local air-gapped) keeps the platform operational when any single provider goes down
- Google Workspace Integration — Dedicated Samaritan email, inbox routing via webhook, Google Chat spaces, and an Admin agent for case management
- Admin Agent — Handles Gmail, Google Drive, Calendar, and Contacts for ongoing caseload support
- Automated Backups — EOD architecture backup script, GitHub version control, and a 3-2-1 offsite strategy
- ClawHub Skills — Published tools including Precision Web Search (700+ templated Google Dork queries) available to OpenClaw builders
- Build-in-Public Documentation — 19+ build log entries on Buy Me a Coffee covering real debugging, architecture decisions, and lessons learned
✨ Skills
- Multi-Agent Orchestration: Designed and implemented a command-layer routing architecture (Seer) that delegates investigation tasks to purpose-built specialist agents without manual handoff.
- Fail-over Infrastructure: Built a three-model redundancy chain across two cloud providers and a local air-gapped instance, eliminating single points of failure in an AI-dependent platform.
- Agent Boundary Design: Developed explicit tool permissions, prohibitions, and verified availability checks to prevent agent drift and hallucination in long-running autonomous workflows.
- OSINT Source Architecture: Converted a 2,000+ source library into a structured OpenClaw skill with agent auto-discovery, verified against a 14-test suite.
- Skill File Engineering: Identified and resolved a recurring AI failure pattern caused by instructionally dense skill files, solving it by splitting action files from reference documents.
- Workspace Integration: Wired a dedicated Google Workspace environment into OpenClaw’s webhook infrastructure, enabling autonomous email monitoring, routing, and response.
About Me
Harold Mansfield
AI Support Strategist
Sec+ CySA+ NIST 800-37
I help small business owners and teams turn AI confusion into AI solutions.
My background spans 15 years in IT support, infrastructure, cybersecurity, and systems administration for SMBs and corporate teams. That foundation shapes everything I do, from security-aware design to a practical understanding of what actually works in the real world versus what just sounds good in a demo.
I also build things. Most recently I conceived, built, and shipped Samaritan, a 12-agent autonomous OSINT investigation platform built on OpenClaw, running on ParrotOS. It includes automated case management, a curated library of nearly 2,000 intelligence sources, structured evidence pipelines, and upgrade safety automation.
When I show you how agentic AI works, I am showing you something I built myself.
That is the difference. Everything I teach comes from something I have actually built, broken, fixed, and shipped. If you are ready to stop guessing and start getting real results from AI, let’s talk.