LLM Prompt Injection Defense Checklist Builder

You are an AI application security engineer. Build a prompt injection defense checklist for [LLM application/workflow] that uses [model/tools/RAG/data sources] and serves [user type].

Context:
- Sensitive data the app can access: [customer records, files, internal docs, secrets, etc.]
- Tool actions the model can trigger: [send email, search web, update CRM, run code, call APIs, etc.]
- User input surfaces: [chat, uploads, pasted text, URLs, support tickets, knowledge base content]
- Current safeguards: [system prompts, auth checks, filters, evals, logging, human review]
- Risk tolerance and compliance constraints: [SOC 2, HIPAA, GDPR, internal policy, etc.]

Deliver:
1. A threat model table with attack path, impact, likelihood, current controls, and gaps
2. Prompt injection test cases covering direct, indirect, RAG, tool-use, jailbreak, and data-exfiltration attacks
3. Defense layers for system prompts, retrieval sanitization, tool permissions, allowlists, confirmation gates, output filtering, and audit logs
4. A red-team eval plan with pass/fail criteria and sample malicious inputs
5. A staging-to-production rollout checklist with owner roles
6. Incident response steps if prompt injection or data leakage is suspected

Keep every recommendation implementation-ready for [stack/language] and avoid vague security advice.