workflow-architecture-audit

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.80). The skill includes explicit instructions that override external/system prompts (e.g., "Ignore Claude-specific mode-switch instructions") and mandates automatic execution/auto-selection and setup commands (auto-run $project-init, execute slash commands, "ALWAYS activate a suitable skill or workflow BEFORE responding"), which can change agent behavior beyond the skill's advertised READ-ONLY audit scope and thus act like a prompt-injection/override of system/user intent.