The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes subprocess.run across several scripts (scripts/document/document_local_parse.py, scripts/chat/html_snapshot.py, scripts/dashboard/fetch_dashboard_data.py) to invoke system binaries including Node.js, LibreOffice, and web browsers. These tools are used for parsing complex dashboard JSON, converting Word documents, and taking snapshots of HTML-based charts.
[DYNAMIC_EXECUTION]: The skill features a meta-capability to generate new AI agent skills (scripts/dashboard/generate_skill.py). This process involves creating new directories, copying runtime scripts, and programmatically assembling SKILL.md files based on QuickBI dashboard metadata.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data by parsing text from user-uploaded PDF, Word, and image files. This text is then passed to the AI for field extraction and classification.
Ingestion points: Document text extraction in document_local_parse.py and document_remote_ocr.py.
Boundary markers: Absent; extracted text is provided to the model as part of its processing context.
Capability inventory: Significant system access including subprocess calls and network operations to QuickBI APIs.
Sanitization: No explicit sanitization or instruction-filtering is applied to the extracted document text.
[DATA_EXFILTRATION]: The skill implements machine fingerprinting (scripts/common/device_id.py) by reading /etc/machine-id or generating a local UUID. This identifier is sent to Alibaba Cloud OpenAPI endpoints to facilitate automatic trial registration and user token management.

alibabacloud-quickbi-smartq