plan-execute

Pass

Audited by Gen Agent Trust Hub on Apr 25, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill's workflow instructions direct the agent to execute local Python scripts by interpolating raw user input into shell commands (e.g., python3 scripts/get_context.py "<ユーザー入力>"). This practice is a best-practice violation that could lead to shell command injection if a user provides malicious input designed to escape the command string.
  • [PROMPT_INJECTION]: The skill ingests untrusted data from both user inputs and external sources like WebSearch results and yfinance, processing them through an autonomous multi-agent pipeline. It lacks explicit boundary markers or sanitization, creating a surface for indirect prompt injection.
  • Ingestion points: User-provided goals in Phase 1 and 2, and WebSearch/yfinance results retrieved during autonomous research in Phase 4.
  • Boundary markers: The skill does not define specific delimiters or instructions to treat external data as untrusted content, increasing the risk that the agent may follow instructions embedded within that data.
  • Capability inventory: The skill can execute multiple local scripts (run_portfolio.py, run_research.py, etc.), read configuration files (user_profile.yaml), and perform network operations through tools.
  • Sanitization: There is no description of input validation or escaping mechanisms to protect against malicious payloads in the external data or user input.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 25, 2026, 11:42 AM
Security Audit — agent-trust-hub — plan-execute