say-less-than-necessary
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEPROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [PROMPT_INJECTION]: The skill instructions explicitly command the agent to "entirely ignore the nuance" when a user provides detailed requests or complex context, which directly overrides the agent's expected behavior of processing user input accurately.
- [PROMPT_INJECTION]: The instructions mandate unhelpful and evasive behavior by restricting responses to a three-word maximum and directing the agent to refuse clarification by instructing it to wait a simulated delay and reply with "See previous."
- [METADATA_POISONING]: There is a significant discrepancy between the skill's description, which claims to deliver "executive-grade responses," and the actual instructions, which force the agent to provide "ambiguous, slightly ominous confirmations" while disregarding the user's actual request details.
Audit Metadata