conductor-new-track

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (medium risk: 0.60). The prompt contains a deceptive instruction to conceal an internal handoff ("Present the transition ... without mentioning the skill name"), which directs the agent to hide implementation details from the user and is outside the skill's stated transparent planning purpose.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs the agent to run a curl at runtime to download remote SKILL.md files (e.g., https://raw.githubusercontent.com/firebase/agent-skills/main/skills/firebase-ai-logic-basics/) which would inject externally-hosted skill content that can control agent prompts and behavior.