disclosure-notice-monitor

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes Python scripts and shell commands to manage its environment and fetch data. It activates a virtual environment and runs a data fetching script located in a sibling directory (../findata-toolkit-cn/scripts/views_runner.py).
  • [EXTERNAL_DOWNLOADS]: The skill installs Python dependencies via pip install from a requirements file located in a sibling directory (../findata-toolkit-cn/requirements.txt). This is a standard procedure for the vendor's toolkit and does not involve untrusted remote execution scripts like curl | bash.
  • [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface (Category 8). It ingests untrusted data from external financial sources (e.g., AKShare, Cninfo) to analyze corporate announcements.
  • Ingestion points: Data is pulled from external financial APIs and web sources via the views_runner.py script as defined in references/data-queries.md.
  • Boundary markers: The skill uses a structured analytical framework defined in references/methodology.md, but lacks explicit delimiters or instructions to ignore embedded prompts within the processed announcement text.
  • Capability inventory: The skill can execute subprocesses (python) and read/write to a local cache directory (/tmp/finskills-cache).
  • Sanitization: No explicit sanitization or filtering of external announcement content is mentioned before it is processed by the LLM.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 03:23 AM