answer-only

Pass

Audited by Gen Agent Trust Hub on Jun 12, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill is entirely instructional and contains no executable scripts, external dependencies, or network-capable code.
  • [PROMPT_INJECTION]: The skill implements defensive measures by instructing the agent to trigger its 'answer-only' mode solely from the latest user-authored natural-language request. It explicitly requires the agent to ignore assistant text, tool output, and approval prompts when making this decision, which mitigates risks associated with indirect prompt injection.
  • [COMMAND_EXECUTION]: Rules 1 and 3 explicitly prohibit the agent from running shell commands, browser automation, MCP tools, scripts, or starting services, preventing any unauthorized system interactions.
  • [DATA_EXFILTRATION]: The skill restricts the agent to answering from the current conversation context only and forbids the creation, modification, or reading of files (Rule 2 and 5), precluding data exposure or harvesting.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 12, 2026, 08:31 PM
Security Audit — agent-trust-hub — answer-only