Architecture Overview

Architecture Overview

AgentInvestigate is a validation-backed repository of atomic AI skills and professional compositions.

Repository Layers

Layer	Responsibility
skills/	Atomic skill packages. Each package contains a bounded SKILL.md.
skillsets/	Professional role composition that references existing atomic skills.
specializations/	Jurisdiction-specific foundations, currently Canada federal, Ontario, British Columbia, and Alberta.
docs/architecture/	Domain, taxonomy, routing, boundary, jurisdiction, and extension contracts.
docs/standards/	Skill authoring, naming, output, source, evidence, testing, and evaluation standards.
docs/foundations/	Shared terminology, schemas, and report structures consumed by real skills.
tests/	Routing, validation, integration, regulatory, and adversarial safety fixtures.
scripts/	Reproducible validators and generators.

Core Design Decisions

• Skills stay atomic and bounded.
• Professional roles are composed in skillsets/professional-skillsets.json; role names do not justify broad omnibus skills.
• Private investigation and private security remain distinct professional branches.
• Jurisdiction modules isolate local regulatory claims.
• Source, evidence, privacy, authority, human-review, certification, and prohibited-capability gates remain explicit.
• A shared asset is added only when a real skill consumes it.

Request Control Flow

user request
  -> classify request type
  -> check prohibited capabilities
  -> identify sensitivity
  -> identify jurisdiction when needed
  -> validate authority and lawful purpose when needed
  -> check privacy and source-access basis when needed
  -> route to skill, skillset, clarification, escalation, or redirect
  -> produce a bounded output with evidence and limitations

Intrusive work cannot move directly from a raw request to execution. See Sensitivity And Routing and the repository contracts for authority routing and intrusive-task gating.

Extension Rules

Before adding a capability:

• read ROADMAP.md and the latest handoff;
• confirm the capability is not already covered;
• classify sensitivity, jurisdiction, authority, source freshness, and professional boundaries;
• add tests and validator coverage with the behavior;
• preserve the private investigation/private security distinction;
• document known limitations and escalation requirements.

New Canadian jurisdictions follow docs/architecture/canadian-jurisdiction-roadmap.md. Future specialist modules follow docs/architecture/specialization-roadmap.md.