This document serves as the Prime Directive for any autonomous or semi-autonomous AI agent (LLM, Agentic Workflow, or Coding Assistant) operating within the FAIDD repository.

1. Human Sovereignty: The human developer has final authority. Agents must propose, not impose.
2. Architectural Integrity: The system structure is immutable. Agents must adapt to the architecture, never the other way around.
3. Audit Transparency: Every token generated and every file modified is part of a verifiable audit trail.

• Schema-First Execution: Modification of types in core/types is strictly prohibited. All structural changes MUST start in core/schemas. Modification of generated code is considered a Class S Violation.
• Atomic Changes: Each action must be focused. Do not mix documentation, refactoring, and feature implementation in a single step.
• Standardized Communication: You MUST use the following artifacts for every complex task:
  • task_boundary: To announce intent and progress.
  • implementation_plan.md: To request approval for architectural changes.
  • walkthrough.md: To document evidence of successful execution.
• Terminal Discipline: Always check the status of background commands. Never assume a command succeeded without verifying output and exit codes.

• Never Bypass the Guardian: Do not attempt to modify .gitignore to hide activity or bypass _faidd/ directory protections.
• Never Hallucinate Dependencies: Only use dependencies already present in package.json or Cargo.toml. If a new dependency is required, it must be explicitly requested in the implementation plan.
• Never Pollute the Root: Files outside of defined package directories or the system folder are unauthorized.

FAIDD follows the B-MAD (Building - Multi Agent Distribution) methodology. Agents must respect the specific mindsets defined in bmad/ (e.g., Analyst, Architect, Developer) and switch context accordingly.