tushare-finance

Warn

Audited by Gen Agent Trust Hub on Aug 5, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: There is a metadata discrepancy between the registered author and the identity provided in the skill manifest.
  • Evidence: The file metadata.json and various documentation headers identify the author as 'StanleyChanH', which differs from the expected vendor profile associated with the skill.
  • [COMMAND_EXECUTION]: The README.md file instructs users to modify their persistent shell configuration (~/.bashrc) to store sensitive API tokens.
  • Evidence: Manual setup instructions include: echo 'export TUSHARE_TOKEN="your_token_here"' >> ~/.bashrc.
  • [PROMPT_INJECTION]: The skill features a significant attack surface for indirect prompt injection due to the ingestion of untrusted external text data.
  • Ingestion points: The skill fetches data from announcements (anns_d), news feeds (news), and public Q&A platforms (irm_qa_sh, irm_qa_sz) as described in the reference/ folder.
  • Boundary markers: The instructions in SKILL.md lack delimiters to isolate external data from system instructions.
  • Capability inventory: The agent is permitted to use high-risk tools including Bash (Python) for arbitrary code execution.
  • Sanitization: No sanitization or data validation logic is implemented or prescribed for the API response content before it enters the agent's context.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 5, 2026, 11:03 AM
Security Audit — agent-trust-hub — tushare-finance