The Agent Skills Directory

[DATA_EXFILTRATION]: The skill transmits document content to external cloud services (MinerU at mineru.net and DeepSeek at deepseek.com) to perform document extraction, OCR, and AI-assisted auditing. This is the primary functionality of the skill and is documented in references/mineru_usage.md and references/extraction.md.
[COMMAND_EXECUTION]: The script scripts/parse_report.py executes the mineru-open-api command-line utility via subprocess.run to process scanned documents. The command is constructed using a list of arguments to prevent shell injection.
[PROMPT_INJECTION]: The skill processes untrusted financial documents (Word, PDF, Excel) provided by users. This creates an indirect prompt injection surface where text within a document could attempt to manipulate the AI agent's auditing logic. As documented in the mandatory evidence chain:
Ingestion points: scripts/parse_report.py reads user-provided files.
Boundary markers: The script uses  HTML comments to separate sections in the generated report.md.
Capability inventory: The skill has file system access, command execution, and network capabilities via cloud APIs.
Sanitization: The skill performs minimal sanitization on text content, relying on the agent's logic to differentiate document content from instructions.
[EXTERNAL_DOWNLOADS]: The skill depends on multiple external Python packages (pdfplumber, python-docx, openpyxl, openai) and a global Node.js package (mineru-open-api) to perform its tasks.

audit-report-checker