eu-ai-act-report-oliver-schmidt-prietz
Warn
Audited by Gen Agent Trust Hub on Jul 30, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: In Phase 4 (Word Document Export), the skill instructs the agent to generate a JavaScript file that utilizes the
docx-jslibrary and then execute that file to create a .docx report. This design relies on the agent's ability to write and run code locally, which presents a security risk if the data being processed contains malicious logic that could be incorporated into the generated script. - [PROMPT_INJECTION]: The skill features a "Context-First Adaptive Intake" (Phase 1) designed to automatically consume and process outputs from other AI agent skills or user-supplied context blocks. This represents an indirect prompt injection surface where adversarial content in the ingested legal data could manipulate the report generation process or the logic of the Word export script.
- Ingestion points: SKILL.md Phase 1 (Context detection) allows the ingestion of prior conversation context and external assessment blocks.
- Boundary markers: The skill does not define explicit delimiters or instructions to treat ingested context as untrusted data.
- Capability inventory: The skill utilizes web search for legal citation verification and performs dynamic script generation and execution for document exporting.
- Sanitization: There are no defined mechanisms for sanitizing or validating external input before it is interpolated into the report templates or the generated export code.
Audit Metadata