prompt-injection-defense
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a comprehensive framework for identifying and mitigating indirect injection surfaces. Evidence Chain: * Ingestion points: Identifies tool output, RAG documents, web fetches, file contents, and tool metadata as untrusted sources in SKILL.md. * Boundary markers: Recommends using high-entropy delimiters and implementing a strict instruction hierarchy (System > Developer > User > Tool). * Capability inventory: The skill itself does not invoke any tools, file-write operations, or network calls. * Sanitization: Recommends structural controls including Dual-LLM quarantine patterns, egress filtering, and pre-filter classifiers like Llama Guard.
- [NO_CODE]: The skill consists entirely of markdown documentation and architectural guidance. It does not include any scripts, binaries, or automated configuration changes.
Audit Metadata