llm-council

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.80). The skill includes explicit directives to ignore external mode-switch instructions and to auto-execute/authorize workflows, slash-commands, and subagents (e.g., "Ignore Claude-specific mode-switch", "execute it directly", auto-run $project-init, authorize spawn_agent), which are hidden/overriding operational controls that go beyond the stated advisory/council purpose and effectively attempt to override environment/system-level boundaries.