The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data logs and reports, which represents a potential attack surface if an attacker could insert malicious instructions into the source files (ROPA, DSR, or breach logs) to influence the agent when it later audits those files.
Ingestion points: The scripts agent.py and process.py load user-provided JSON and CSV files for compliance analysis and PII scanning.
Boundary markers: The skill does not implement specific delimiters or instructions to ignore embedded commands within the processed data files.
Capability inventory: The skill scripts have the capability to read local files (for PII scanning) and write results to the filesystem (JSON/CSV report generation).
Sanitization: While the scripts validate JSON and CSV structure, they do not sanitize textual content for potential natural language prompt injection patterns.
[SAFE]: The identified indirect prompt injection surface is inherent to the skill's purpose as a compliance auditing tool. The risk is minimized because the scripts focus on statistical aggregation and structure validation, and the data is not dynamically executed or used to construct complex prompts in the provided code.
[SAFE]: No signs of prompt injection, data exfiltration, obfuscation, or remote code execution were found. All dependencies are standard Python libraries, and all actions trace back to the intended purpose of GDPR data protection management.

implementing-gdpr-data-protection-controls