prompt-builder-by-rdd
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user data (documents, rules, and examples) to generate a system prompt. This creates a surface where malicious instructions in the source documents could attempt to hijack the generated prompt's logic.
- Ingestion points: User-provided documents categorized into 'gavetas' (SKILL.md, references/BRAINSTORM_PROTOCOL.md).
- Boundary markers: The skill generates a single, structured XML block using a strict framework (references/PROMPT_FRAMEWORK.md).
- Capability inventory: The skill is meta-operational, generating text instructions rather than executing shell commands or network requests (SKILL.md).
- Sanitization: Includes manual validation steps and instructs the agent to separate rules from examples, although automated content filtering for injection is not present.
- [SAFE]: The skill implements significant safety controls, most notably a confirmation gate that prevents prompt generation until the user explicitly approves a summarized scope. It also provides comprehensive guidance on data privacy and security requirements for sensitive information (references/SOURCE_AND_SECURITY.md).
Audit Metadata