data-privacy-engineer
Pass
Audited by Gen Agent Trust Hub on Mar 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTIONNO_CODE
Full Analysis
- [COMMAND_EXECUTION]: The agent configuration in
references/agents/data-privacy-engineer.mdincludes theBashtool, which allows for the execution of arbitrary shell commands. While consistent with an engineering persona, this represents a significant capability that could be exploited if the agent is misled by untrusted data. - [EXTERNAL_DOWNLOADS]: The agent is granted access to the
WebFetchtool, enabling it to retrieve content from external URLs. - [NO_CODE]: The skill package does not include any executable scripts, binaries, or logic in programming languages like Python or JavaScript. It consists solely of markdown-based instructions and metadata.
- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection:
- Ingestion points: Untrusted data enters the agent context via the
WebFetchandReadtools inreferences/agents/data-privacy-engineer.md. - Boundary markers: The instructions do not define clear delimiters or "ignore embedded instructions" warnings to differentiate between system instructions and processed data.
- Capability inventory: The agent possesses high-impact tools including
Bash(command execution),WriteandMultiEdit(file modification), andWebFetch(network access). - Sanitization: No mechanisms for data validation, escaping, or sanitization are prescribed in the instructions.
Audit Metadata