power-word-kill
Warn
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill directs the agent to execute high-risk operations including
kill -9(unconditional process termination), 'hard account terminations', 'circuit breaker trip', and 'emergency shutdown', which can lead to significant data loss and service interruption.\n- [PROMPT_INJECTION]: The use of authoritative language and D&D metaphors ('No negotiation', 'no saving throw', 'instantly destroys') is designed to override the agent's typical safety constraints when dealing with destructive or permanent system changes.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes a 'target' which could be supplied by untrusted external data, potentially allowing an attacker to manipulate the agent into terminating critical systems or locking accounts.\n - Ingestion points: Target identification and verification in SKILL.md.\n
- Boundary markers: Step 1 and 2 require the agent to restate and confirm the target.\n
- Capability inventory: Process termination, account locking, and system shutdown capabilities.\n
- Sanitization: No input validation or sanitization is defined for the target string.
Audit Metadata