tracking-crypto-portfolio

Warn

Audited by Gen Agent Trust Hub on Mar 13, 2026

Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTIONSAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill processes external, user-provided JSON files to load portfolio holdings, which creates an indirect prompt injection surface. \n
  • Ingestion points: scripts/portfolio_loader.py reads JSON data from the file system. \n
  • Capability inventory: The skill uses Bash(crypto:portfolio-*), Read, and Write tools across its scripts. \n
  • Boundary markers: There are no specific boundary markers or instructions to the model to ignore potential injection content within the JSON data. \n
  • Sanitization: Data is validated and cast to numeric types where appropriate, but string fields like coin symbols and notes are reflected in the output. \n- [COMMAND_EXECUTION]: The skill defines and uses a scoped Bash tool permission pattern (crypto:portfolio-*) to execute its Python scripts, which minimizes the risk of arbitrary command execution. \n- [SAFE]: Network requests are made to the CoinGecko API (api.coingecko.com) to retrieve market prices. This is a well-known service, and the data exchanged is restricted to public asset symbols without involving sensitive user information or system credentials. \n- [SAFE]: There is a discrepancy between the reported author (aAAaqwq) and the author listed in the metadata (Jeremy Longshore), which is flagged for potential metadata poisoning or misleading information.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 13, 2026, 06:58 AM