feat: add hook based context injection for ai agents

[dev-ahmadbilal]

Implements hook-based context injection to keep AI coding agents aligned with our hexagonal architecture across sessions and models.

What's included:

• AGENTS.md — session-level hot memory: layer map, dependency rules, NEVER list, routing table
• docs/layers/ + docs/cross-cutting/ — 9 leaf docs injected on demand per file type
• scripts/inject-context.mjs — PreToolUse hook that blocks wrong file placements and injects relevant docs before every edit
• scripts/arch-validate.sh — PostToolUse hook with 8 blocking checks:
  • domain importing TypeORM
  • service using @InjectRepository
  • raw NestJS exceptions in services
  • unmapped repo returns
  • console.* usage
  • export default
  • wrong directories
• .claude/settings.json — wires both hooks

Why:

Without enforcement, agents bypass the hexagonal pattern silently — injecting TypeORM directly into services, skipping Mapper.toDomain(), and placing files in incorrect directories. Each violation becomes precedent for the next agent.

The hooks fire unconditionally on every edit; agents cannot proceed past a violation.

See docs/agent-hooks-guide.md for before/after examples and token cost analysis.

Based on: Hook-Based Context Injection for AI Coding Agents by Andrew Patterson.