ai-agent-architecturelisted

**CONSUMES**: User agent system requirements or existing agent codebase **PRODUCES**: 10 architecture decisions + design document or audit report with P0/P1/P2 findings # AI Agent Architecture A decision-by-decision navigator for building reliable agent systems. Each decision has a selection matrix derived from production systems. Each skipped decision has a documented production disaster. **Audience**: AI agents designing other agent systems. **Two modes**: - `/design` — Walk through D1–D10 for a new system - `/audit` — Check an existing system against all 10 decisions --- ## How to Use This Pack **On activation**, detect the user's intent: - Contains: "design / build / create / new agent / architect" → enter `/design` mode - Contains: "review / audit / check / existing / evaluate" → enter `/audit` mode - Ambiguous → ask: "Are you designing a new agent system or auditing an existing one?" --- ## /design Mode ## Step 0 — Scoping (Ask BEFORE reading any reference file) Ask the following 5 questions. Collect all answers before proceeding to Step 1. **Q1** — Scale: Single agent or multi-agent system? - Single agent → D2 (coordination) is likely skipped - Multi-agent → D2 is mandatory **Q2** — Session model: Stateful (long sessions, multi-turn continuity) or stateless (single-turn, ephemeral)? - Stateless → D3 (memory) may be skipped, simplified D6 (compression) sufficient - Stateful → D3 and D6 are mandatory **Q3** — Input trust boundary: Trusted inputs only (inter