From prodsec-skills
Guides securing RAG system data against unauthorized access, modification, and poisoning. Covers access control, encryption, integrity verification, and audit logging for vector databases and knowledge stores.
How this skill is triggered — by the user, by Claude, or both
Slash command
/prodsec-skills:secure-storageThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Data used by RAG systems (or other plugin systems) MUST be protected against unauthorized access and modification. If RAG data is modified maliciously, the LLM system will provide that malicious information to users as if it were legitimate.
Data used by RAG systems (or other plugin systems) MUST be protected against unauthorized access and modification. If RAG data is modified maliciously, the LLM system will provide that malicious information to users as if it were legitimate.
| Risk | Impact |
|---|---|
| Data poisoning | Attacker modifies RAG data to inject false or harmful information |
| Unauthorized access | Sensitive knowledge base data exposed to unauthorized users |
| Integrity compromise | Tampered RAG data causes the LLM to produce incorrect outputs |
| Trust exploitation | Users trust RAG-augmented responses, amplifying the impact of poisoned data |
| Control | Description |
|---|---|
| Access control | Only authorized services and users can read RAG data |
| Write protection | Only authorized data pipelines can modify RAG data |
| Integrity verification | Detect unauthorized modifications to RAG data |
| Encryption at rest | Encrypt RAG data storage (vector databases, document stores) |
| Audit logging | Log all access and modifications to RAG data |
| Input validation | Validate and sanitize data before ingestion into the RAG store |
npx claudepluginhub redhatproductsecurity/prodsec-skills --plugin prodsec-skillsProvides security patterns for RAG/CAG systems including multi-tenant isolation strategies, document-level access control, prompt injection prevention, data classification, and checklists. Use for secure retrieval- or cache-augmented generation with tenant isolation.
Detects RAG pipelines that ingest external documents into LLM context without sanitization or trust gating. Flag vulnerable patterns like direct concatenation, unbounded retrieval, and SSRF-through-fetch.
Tests vector stores for embedding inversion, cross-tenant leakage, knowledge-base poisoning, and retrieval manipulation in RAG systems.