linkfox-aigc-imagegen-cloth

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.90). One URL (the agent-files link to a release.zip) is a direct archive download used to distribute/install a skill (executable content) outside standard package channels, which is a common vector for malware and matches the HIGH‑RISK indicators.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). 在套图编排的 plan 阶段，运行时会把 agent 外部输入（用户消息形成的 point/pointHint/imageDesc 等）拼进 scripts/collection_pipeline/plan.py 的 S1 推理模板并调用 textgen（scripts/collection_pipeline/utils.py: run_textgen），从而让后续 LLM 出图提示词承载这类自由文本；但该 textgen 只消费模板化字段而非读取外部任意文本源或队列/Feed。

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

• Potentially malicious external URL detected (high risk: 0.90). SKILL.md instructs at runtime to download and install remote code from https://agent-files.linkfox.com/skills/linkfox-onboarding/release.zip (a release zip of a skill), which is a runtime fetch of external code that would be installed/executed to control onboarding prompts—this is a high-confidence runtime external dependency.