The Agent Skills Directory

[SAFE]: The skill provides instructions for a legitimate financial data processing task. No malicious patterns, such as prompt injection or data exfiltration, were identified during analysis.\n- [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for processing untrusted data. \n * Ingestion point: Receives invoice data in text, OCR, JSON, or CSV formats (SKILL.md).\n * Boundary markers: Present in the 'Safety Boundaries' section, which defines the output as analytical support and mandates human review.\n * Capability inventory: The skill is limited to extraction and reporting; it does not define network access, shell execution, or file system write capabilities.\n * Sanitization: Not explicitly defined. The risk is assessed as safe due to the narrow scope and lack of exploitable tools.\n- [SAFE]: No obfuscation, hardcoded credentials, remote code execution, or persistence mechanisms were found in the skill content or examples.

invoice-processing