skills/rade-nugroho/agentic-enteprises-skill/prompt-engineer-agent-prompts-evals/Gen Agent Trust Hub
prompt-engineer-agent-prompts-evals
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides guidance on building agents that process untrusted user data and documents. While this introduces an attack surface, the instructions specifically mandate robustness testing (Scenario S7) and the use of clear delimiters to separate developer instructions from user-provided content.
- Ingestion points: The skill describes processing user messages and external documents as described in
references/agent_system_prompts.mdandreferences/agent_eval_scenarios.md. - Boundary markers: The skill explicitly recommends using clear delimiters for volatile context in
references/agent_system_prompts.md. - Capability inventory: None. The skill consists of documentation and guidelines without executable code.
- Sanitization: The skill recommends testing for PII redaction and refusal policies in
references/agent_eval_scenarios.md. - [DATA_EXPOSURE]: The documentation includes safety scenarios (S8) and explicit guidance to verify that agents do not leak credentials, internal URLs, or PII during evaluation traces.
- [PROMPT_INJECTION]: The skill advocates for defensive design by including specific evaluation scenarios (S7) to ensure agents ignore instructions embedded within user documents or external inputs.
Audit Metadata