cuihuo-decision

Pass

Audited by Gen Agent Trust Hub on Jul 24, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: No security issues detected. The skill serves as a cognitive tool for structuring user decisions.
  • [COMMAND_EXECUTION]: The skill implements a file-saving routine where decision records are stored in a dedicated 决策/ (Decision) directory as markdown files. This is a standard documentation function and does not involve arbitrary command execution or privilege escalation.
  • [DATA_EXFILTRATION]: Analysis confirms the absence of network-reaching tools, remote URLs, or exfiltration patterns. Data remains within the user's local context.
  • [PROMPT_INJECTION]: The skill uses firm instructional language (e.g., '打回' / 'rejecting' vague input) to guide the user through a specific cognitive process. This is intended behavior to improve input quality and does not represent an attempt to bypass agent safety filters.
  • [INDIRECT_PROMPT_INJECTION]: While the skill ingests untrusted user input to generate decision cards, it defines clear schemas (markdown template) and enforces strict validation steps (Phases 1-3). The potential for a user-supplied decision title to influence the agent's internal logic is minimal and consistent with safe assistant behavior.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 24, 2026, 10:36 AM
Security Audit — agent-trust-hub — cuihuo-decision