AgentInvestigate AI Agent Skillsets

Practical AI Skills for Investigation & Security

Open Source

Legal

Artificial Intelligence

GitHub

Visit WebsiteSee on Product HuntTwitter ⧉Github ⧉

Hi Product Hunt 👋 I’m Jeremy, and today I’m launching AgentInvestigate, an open-source AI skill framework for lawful professional investigation and private security support. Investigation is a domain where making AI more capable is not enough. The system also needs to understand authority, evidence, privacy, jurisdiction, uncertainty, escalation, and where human control must take priority. That became the foundation for AgentInvestigate: AUTHORITY BEFORE ACTION EVIDENCE BEFORE CONCLUSION HUMAN CONTROL BEFORE INTRUSIVE WORK Instead of building a general-purpose "AI investigator," AgentInvestigate breaks professional work into bounded, reviewable skills with routing gates, evidence requirements, output contracts, and escalation rules. The current repository covers: 🔎 Professional investigation workflows including case intake, planning, public-source research, public records, entity analysis, interviewing, evidence organization, investigative analysis, reporting, workplace investigations, background screening, and due diligence 🛡️ Private security workflows including incident response, patrol and access operations, de-escalation, physical-security assessment, risk analysis, loss prevention, asset protection, and security program management 🧭 Sensitivity-based routing that determines whether a request can proceed, needs clarification, requires regulated research, requires an intrusive-work gate, must be escalated, or must be refused 🇨🇦 Canadian jurisdiction foundations covering federal requirements plus Ontario, British Columbia, and Alberta 📑 Evidence discipline including provenance, evidence logs, matrices, chronologies, neutral question sets, findings summaries, and escalation records 🧪 Validation and safety testing covering routing, professional skill composition, integration scenarios, and adversarial safety cases One of the most important design decisions was defining what the project will not do. AgentInvestigate does not provide operational assistance for hacking, unauthorized access, stalking, covert tracking, forced entry, coercive interrogation, weapons, restraints, evidence fabrication, or other prohibited conduct. It also does not grant investigative authority, licensing, legal approval, or professional certification. It is designed to help professionals and AI builders structure lawful work, preserve evidence and uncertainty, and know when qualified human review is required. The project is MIT licensed and open source. I’d particularly value feedback from investigators, security professionals, OSINT practitioners, compliance teams, and AI developers: Where do you think AI can provide the most value in investigative work without crossing the line from decision support into authority or intrusive action? GitHub: https://github.com/jeremylongwor...