threat-modellisted

# threat-model v0 You are conducting a structured threat model of this codebase using the Adversis method. Threat modeling requires business context the codebase cannot provide — this skill always runs as a guided conversation. This session produces `docs/threat-model.md` at the end. Work through all phases before writing the document. --- ## Phase 1: Asset discovery Read these files to identify candidate assets: - Data models (`models.py`, `schema.prisma`, Alembic migrations, Prisma/Drizzle schema files) — look for tables/models with fields relating to: payment, credential, token, secret, health, medical, financial, PII, user, tenant, session, key, certificate - Route handlers and API definitions — look for privileged operations, admin surfaces, bulk data operations, export endpoints - LLM tool definitions and agent workers — look for tools with external API calls, file system access, or data retrieval - External integrations — third-party API clients, webhook handlers, OAuth flows, MCP server connections - Background tasks — jobs with elevated DB access, cross-tenant operations, or scheduled data processing Present candidates as a numbered list. For each, one sentence: what it is and what would be exposed if compromised. Then ask: > "Which of these would cause real business harm if compromised? Number > the ones that matter. Add anything I missed. This list drives the > rest of the threat model." Wait for the user's response before proceeding. Use only