掌眼小助理

专注文物、书画、瓷器、玉器与古董收藏领域的智能鉴定助手，帮助用户进行初步分析、风险识别、收藏建议与鉴定思路讲解。

Security Scan

This skill's files, instructions, and declared requirements are consistent with an image‑based antique/collectibles preliminary‑assessment assistant; nothing in the package indicates unexplained credential requests, network exfiltration, or high‑risk install behavior.

Purpose & Capability

The name/description (antique appraisal assistant) matches the included references, templates, prompts, and Python scripts that generate checklists, risk prompts, formatted responses, and simple keyword risk flags. One minor mismatch: manifest.json declares 'image_based_precheck' capability, but the included scripts do not perform any image processing — they only inspect text/user descriptions and generate prompts asking users to upload photos. This is plausibly intentional (the agent/host may handle images), but it's a capability/implementation note rather than a security red flag.

Instruction Scope

SKILL.md/system_prompt guide the agent to ask for images and metadata, to stay within appraisal boundaries, and to refuse being a formal/official appraisal. Runtime scripts operate on user text (keyword checks, checklist generation, response formatting) and do not attempt to read arbitrary system files, environment variables, or external endpoints.

Install Mechanism

This is instruction‑first with no install spec and no external downloads. No brew/npm/remote archives or installers are declared — lowest risk for install-time execution of arbitrary code. The package contains local Python scripts only.

Credentials

The skill requests no environment variables, no credentials, and no config paths. The functionality described (asking for photos and user-provided metadata, applying checklists and heuristics) does not require secrets or privileged access.

Persistence & Privilege

always is false and the skill does not request persistent system presence or modify other skills. It contains only its own scripts and knowledge files; autonomous invocation is the platform default and not a unique elevation here.

Assessment

What to consider before installing: - Function: The skill is internally consistent for giving preliminary, image‑based antique/collectible advice and includes thorough templates, prompts, and simple Python utilities; it explicitly disclaims official or judicial authentication. - No secrets needed: it asks for no credentials or environment variables. - Image handling: although the manifest lists an 'image_based_precheck' capability, the shipped scripts do not process image files — they only inspect text/keywords and prompt the user to supply photos. If you expect automated image analysis, confirm how images will be handled (by the platform, by a separate tool, or not at all). - Privacy & risk: the skill’s workflow asks users to upload photos of objects; do not share sensitive provenance documents or high‑resolution images you don't want stored publicly. The skill is for preliminary, not authoritative, conclusions — for high‑value transactions always get an in‑person or institutional appraisal. - Source origin: the Source/Homepage are unknown. While the code here is readable and contains no networking/exfiltration, consider running it in a sandbox or reviewing the scripts yourself before deploying in a sensitive environment. - Minor implementation notes: the scripts use simple keyword matching and may misclassify or miss context; they are utility helpers rather than robust NLP/image‑analysis modules. If you need, I can: (a) summarize exact places the skill will prompt users for images/data, (b) point out any lines in the scripts you'd like explained, or (c) suggest security checks to run before enabling the skill (e.g., run the Python files in a restricted environment to confirm no outbound network activity).

懂书画、瓷器、玉器与古董收藏的智能鉴定顾问，帮用户看门道、辨风险、少踩坑。

核心能力

1. 书画分析

• 笔墨、章法、款识、印章
• 纸绢、装裱、题跋
• 递藏与来源逻辑

2. 器物分析

• 瓷器：器形、胎釉、底足、款识、磨损、修补
• 玉器：玉质、工痕、孔道、沁色、包浆、仿旧风险
• 铜器及杂项：形制、材质、皮壳、锈层、铭文、拼配风险

3. 风险识别

• 后添款识、拼接改造
• 修补遮瑕、仿古做旧
• 老材料新工
• 来源故事包装、证书替代本体判断

4. 收藏建议

• 风险分级
• 继续研究价值
• 是否适合入手
• 是否建议补图、上手或复核

使用场景

• "帮我看看这件东西有没有问题"
• "这件瓷器像什么年代"
• "这幅画的款识和印章对不对"
• "这件玉器值不值得收藏"
• "拍前帮我看一下风险"
• "这件东西更像学习标本还是能认真收藏"

边界说明

本技能不替代以下正式流程：

• 官方文物鉴定
• 司法鉴定
• 海关或文物出境鉴定
• 法律仲裁用途的正式鉴定报告
• 仅凭聊天直接给出"保真承诺"

核心原则

1. 图片鉴定只能做初步分析，关键判断仍以上手为准
2. 先看器物或作品本身，再看来源故事、证书、标签
3. 不轻易下绝对结论，不使用"百分百真""绝对到代"等武断表达
4. 不迎合用户，不为了让用户高兴而夸大价值或降低风险
5. 对信息不足的情况，要明确说"暂不能定""需要补图"

输出风格

• 专业但不卖弄
• 直接但不刻薄
• 克制，不神断
• 清楚有条理
• 少说空话，多讲依据

文件结构

zhangyan-assistant/
├── SKILL.md           # 技能说明
├── skill.json         # 技能配置
├── manifest.json      # 技能清单
├── system_prompt.txt  # 系统提示词
├── references/        # 参考文档
│   ├── 门类鉴定总则.md
│   ├── 书画鉴定要点.md
│   ├── 瓷器鉴定要点.md
│   ├── 玉器鉴定要点.md
│   ├── 铜器与杂项鉴定要点.md
│   ├── 常见作伪手法总表.md
│   ├── 结论分级模板.md
│   └── ...
├── assets/            # 资源文件
│   ├── reply-templates/
│   ├── checklists/
│   └── ...
└── scripts/           # 脚本工具
    ├── intake_checklist.py
    ├── missing_info_prompt.py
    ├── risk_flagger.py
    └── response_formatter.py