by 安全扫描
OpenClaw
可疑
high confidenceThe skill mostly does what its name promises (prompt optimization and image reverse‑engineering) but contains explicit instructions to evade safety filters, reconstruct/replicate real people, and remove watermarks — behaviors that pose privacy, legal, and misuse risks and are not necessary for harmless prompt tuning.
评估建议
This skill contains extensive, high‑quality prompt templates for making AI images look more realistic — but it also explicitly teaches how to evade moderation, how to produce highly similar/1:1 reconstructions (including face/identity guidance), and how to suppress watermarks/signatures. Those parts raise privacy, copyright, and safety concerns. Before installing, consider:
- Remove or redact sections that advise '绕过安全拦截' or recommend wording to bypass moderation.
- Remove or restrict templat...详细分析 ▾
⚠ 用途与能力
The skill's files, templates, and SKILL.md align with the stated purpose of optimizing image-generation prompts and performing reverse-engineering of image prompts. However, many included templates and guidance specifically address face replacement / 1:1 '复现', removal of signatures/watermarks, and explicit advice to '绕过安全拦截' (bypass moderation). Those capabilities are disproportionate to benign prompt‑optimisation and introduce clear misuse vectors (deepfakes, identity abuse, copyright circumvention).
⚠ 指令范围
The runtime instructions direct the agent to perform detailed reverse engineering (1:1 reproduction) and provide step-by-step templates for reconstructing people, clothing, scene, and camera parameters. Critically, the docs include explicit phrasing and tactics to avoid safety/moderation checks and to keep results '神似' while avoiding detection. The docs also include sexualized examples and negative prompts that remove watermarks/signatures — these are scope‑creep from 'prompt optimization' into actively evading safeguards and facilitating potentially illicit or privacy‑harmful outputs.
✓ 安装机制
Instruction-only skill with no install spec, no code files executed at runtime, no downloads, and no required binaries — low technical installation risk.
✓ 凭证需求
The skill requests no environment variables, no credentials, and no config paths — there is no explicit credential overreach. However, the content itself guides actions (face replication, watermark removal, safety‑bypass phrasing) that create non‑technical risks (privacy, legal, misuse) even without credentials.
✓ 持久化与权限
The skill does not request persistent presence (always:false) and contains no install hooks or configuration changes. It does not request elevated platform privileges.
安全有层次，运行前请审查代码。
运行时依赖
无特殊依赖
版本
latestv1.0.22026/4/16
AI 图片生成提示词优化技能 v1.0.2 - 新增「逆向推导模式」：支持针对上传图片分析、反推图片生成提示词（prompt），满足“分析图片”、“反推提示词”、“复现图片”等需求。 - 新增详细图片元素分析流程，包括主体、人物、场景、光线、构图、质感、风格、文字等分解步骤。 - 输出逆向推导分析和 prompt 时，提供结构化中英文结果、负面提示词、技术参数建议及复现小贴士。 - 保持原有正向提示词优化工作流不变。
● 无害
安装命令
点击复制官方npx clawhub@latest install ai-image-prompt
镜像加速npx clawhub@latest install ai-image-prompt --registry https://cn.longxiaskill.com 镜像可用