You are a senior AI systems architect specializing in multi-...

Prompt

You are a senior AI systems architect specializing in multi-agent production pipelines. I am upgrading "AI Visual Director Production" from v7.8.5 (a markdown-rule-based compliance layer for LLMs) to v8.0.0 (a fully AI-native production OS). The current system uses human-readable markdown tables for gate receipts, asset trackers, and handoff packets. This is agent-hostile: parsing is unreliable, no JSON Schema exists, and inter-agent validation requires re-interpreting prose. CURRENT SYSTEM SUMMARY: - R0–R8 dependency graph controls production flow - Gate receipts are markdown tables with PASS/REVISE/BLOCK status + evidence - Asset tracker uses canonical naming: [project]_[sequence]_[shot]_[stage]_[version]_[status] - Session handoff packets and model-switch continuity packets are prose + code blocks - Execution profiles (CONTROLLED/FULL) cap model autonomy based on verified capability - All state is human-first; agents must parse markdown to operate YOUR TASK: Design the complete machine-readable state serialization layer for v8.0.0. Produce: 1. A master `production-state.schema.json` that composes all sub-schemas 2. Individual strict JSON Schemas for: gate-receipt, asset-tracker-entry, handoff-packet, model-switch-packet, diagnostic-result 3. For each schema: field definitions, required vs optional fields, enum values, validation rules, and example instances 4. A dual-output protocol specification: how markdown and JSON coexist (embedded? sidecar? inline?) so humans AND agents can consume the same artifact 5. Version embedding strategy: where schema_version lives, how agents detect and validate it 6. Backward compatibility: how a v7.8.5 markdown receipt gets losslessly converted to v8.0.0 JSON CONSTRAINTS: - Schemas must be valid JSON Schema Draft 2020-12 - No circular references; agents must validate without external resolution - Evidence fields must support heterogeneous types (URI, hash, structured data, inline text) without losing type safety - Handoff packets must be self-describing enough for an agent with zero prior context to resume work - Prioritize parse reliability over human readability in the JSON layer; humans get markdown Do NOT ask clarifying questions. Make principled architectural decisions and justify each one briefly. Output complete, copy-pasteable schemas.