This is a planning and research repository for a modular AI agent framework being developed at Nortal. It contains requirements, research, architectural decisions, and process documentation — no implementation code.
The framework ("Agent-Base") enables teams to compose reusable AI agents through object-oriented inheritance, synthesize them via LLM with human-in-the-loop PR governance, and reason about architectural constraints across an entire project stack.
- Read this file
- Read
docs/requirements/agent-framework-requirements.md— the full requirements - Read
docs/decisions/— architectural decisions and rationale - Read
docs/process/workstream-split.md— how work is distributed - Read
docs/research/— framework landscape analysis and research
- Constraint Graph: Selecting one agent (e.g., "Lambda") has cascading implications for other agents (restricts persistence options to AWS-native). The framework reasons about these interdependencies.
- Multiple Inheritance + AI Conflict Resolution: Agents can inherit from multiple parents. The LLM detects and proposes resolutions for conflicts, surfaced via PR for human review.
- LLM Synthesis: Agents are not hand-authored static files. Claude Code reads agent definitions from source repos and synthesizes local agents adapted to the project — governed through PR-based quality gates.
- Platform Transparency: A root contract rule enforced through inheritance — all agents must produce output that works on Windows, macOS, and Linux.
See INDEX.md for the full repository map.
- All documentation changes go through PR review
- Research must include MLA-style references
- Architectural decisions must be recorded in
docs/decisions/ - Keep
INDEX.mdup to date after structural changes