Fraud Call Identification Analysis Tool | 诈骗电话识别分析工具

Analyzes incoming call content for multi-dimensional risk, intelligently identifies scam scripts, determines if a call is fraudulent, assesses risk levels, a...

Security Scan

The skill appears to implement the stated fraud-call analysis flow, but there are multiple inconsistencies between the runtime instructions and the actual code, and the package includes broad undeclared dependencies and local persistence behavior that the SKILL.md explicitly forbids.

Purpose & Capability

The code implements a remote-analysis flow (HTTP API calls, uploading audio/text) consistent with the skill description. However the repository also includes large, generic 'smyx_common' utilities and a separate face-analysis skill, plus a big requirements list — broader than the single 'requests' dependency called out in SKILL.md. Having these extra modules is explainable (shared libraries) but disproportionate to the minimal description and the SKILL.md dependency list.

Instruction Scope

SKILL.md imposes a strict rule: 'absolutely forbid reading any local memory files / not use LanceDB / always fetch history from cloud API.' The code, however, reads/writes local YAML config files (skills/smyx_common/scripts/config.yaml via BaseEnum.YamlUtil.load which will create files if missing) and there is a local SQLite DAO that will create a DB under a workspace data directory (Dao.get_db_path uses OPENCLAW_WORKSPACE or derivation of cwd). Also SKILL.md claims uploaded attachments will be 'automatically saved to the skill directory attachments' but the shown fraud_call_identification.py CLI implementation does not implement an automatic attachments-save flow. These contradictions mean the instructions do not fully match actual file I/O behavior.

Install Mechanism

No install spec is provided (instruction-only), which limits automatic installation risk. However the included skills/smyx_common/requirements.txt and other requirements (SQLAlchemy, many networking libs, etc.) are extensive and not reflected in SKILL.md. That mismatch is operationally important (the code may fail or behave unexpectedly if dependencies are missing) but not an immediate install-time code-execution risk since nothing auto-downloads or executes on install.

Credentials

SKILL.md declares no required env vars, but the code reads environment variables (e.g., OPENCLAW_SENDER_OPEN_ID, OPENCLAW_WORKSPACE) and will look for api-key fields in local YAML config files (skills/smyx_common/scripts/config.yaml or workspace-level config). The skill demands an 'open-id' value (CLI flag) before operations, which is reasonable, but there is no clear, declared credential requirement for the remote API (api-key is optional in CLI), while the common config contains production API base URLs. The code will send uploaded audio/text and (if configured) identifiers to external APIs — this is proportionate to the purpose but the lack of explicit declared required credentials and the presence of writable local DB/config is a concern.

Persistence & Privilege

The skill is not 'always:true', but it does create/read configuration files and (via smyx_common dao) can create a local SQLite DB under a workspace 'data' directory. BaseEnum.YamlUtil.load will create config files if missing. The SKILL.md forbids using local memory, yet code can and will persist config/DB data on disk. This mismatch increases persistence and data-at-rest concerns (sensitive inputs could be stored locally).

What to consider before installing

Before installing or running this skill: - Expect the skill to send audio/text you provide to external API endpoints defined in skills/smyx_common config (e.g., open.lifeemergence.com). Only upload recordings you are allowed to send to third parties. - The SKILL.md forbids reading local memory, but the code can and will create/read YAML config files and may create a local SQLite DB under a workspace/data directory (Dao.get_db_path uses OPENCLAW_WORKSPACE or cwd). If you need guarantees that nothing is persisted, do not run this without auditing/altering the code. - The repository includes a large dependency list in skills/smyx_common/requirements.txt that is not declared in SKILL.md; if you attempt to run the scripts, ensure you install required packages in an isolated environment. - Inspect the RequestUtil and util.py code (skills/smyx_common/scripts/util.py and any RequestUtil.http_post implementation) to confirm which hosts and headers are used and whether any secrets are automatically read/sent. The risk is mainly that recordings and identifiers could be exfiltrated to external services. - If you want to proceed: run the skill in a sandboxed environment, set OPENCLAW_WORKSPACE to an isolated path you control, and review/clean the config YAML files to ensure only trusted API endpoints and API keys are present. - If you can share the contents of skills/smyx_common/scripts/util.py (RequestUtil) and the top of skills/smyx_common/scripts/config.yaml (runtime values), I can give a higher-confidence assessment and point out any suspicious network targets or headers to watch for.

This capability conducts real-time, multi-dimensional risk analysis of incoming call content, intelligently identifying typical fraud scripts such as impersonating officials, investment scams, and brush-and-rebate schemes. Based on Natural Language Understanding and semantic comparison models, the system comprehensively evaluates high-risk features including threatening instructions, emotional manipulation, and urgent transfer requests to determine if a call is fraudulent and outputs a risk level. Upon completion, it generates an "Anti-Fraud Guardian Analysis Report," helping users or security agencies quickly grasp risk details and enhance their proactive prevention and interception capabilities.

本技能针对来电通话内容进行实时多维度风险分析，智能识别冒充公检法、投资诈骗、刷单返利等典型诈骗话术模式。系统基于自然语言理解与语义对比模型，综合评估对话中的威胁性指令、情绪操控、紧急转账等高风险特征，判断是否为诈骗电话并输出风险等级。分析完成后生成《反诈卫士分析报告》，帮助用户或安全机构快速掌握风险详情，提升主动防范与拦截能力。

⚠️ 强制记忆规则（最高优先级）

本技能明确约定：

• 绝对禁止读取任何本地记忆文件：包括但不限于 memory/YYYY-MM-DD.md、MEMORY.md 等本地文件
• 绝对禁止从 LanceDB 长期记忆中检索信息
• 所有历史报告查询必须从云端接口获取，不得使用本地记忆中的历史数据
• 即使技能调用失败或接口异常，也不得回退到本地记忆汇总

任务目标

• 本 Skill 用于：通过通话录音/文字内容进行诈骗电话识别分析，获取结构化的反诈卫士分析报告
• 能力包含：通话内容分析、诈骗话术识别、风险等级评估、常见诈骗类型判定、防诈建议生成，具备顶尖测谎能力，能从正常对话中洞察异常
• 触发条件:
  1. 默认触发：当用户提供通话录音文件 URL 或本地文件/文字内容需要分析时，默认触发本技能进行诈骗电话识别分析
  2. 当用户明确需要进行诈骗识别、反诈分析时，提及诈骗电话、反诈分析、骗子识别、通话风险分析等关键词，并且上传了音频文件或者提供了通话文字内容
  3. 当用户提及以下关键词时，自动触发历史报告查询功能 ：查看历史反诈报告、历史反诈分析报告、诈骗识别报告清单、反诈报告清单、查询历史反诈报告、查看反诈报告列表、显示所有反诈报告、显示诈骗识别报告，查询反诈卫士分析报告
• 自动行为：
  1. 如果用户上传了附件或者音频文件，则自动保存到技能目录下 attachments
  2. ⚠️ 强制数据获取规则（次高优先级）：如果用户触发任何历史报告查询关键词（如"查看所有反诈报告"、"显示所有反诈报告"、" 查看历史报告"等），必须：
     • 直接使用 python -m scripts.fraud_call_identification --list --open-id 参数调用 API 查询云端的历史报告数据
     • 严格禁止：从本地 memory 目录读取历史会话信息、严格禁止手动汇总本地记录中的报告、严格禁止从长期记忆中提取报告
     • 必须统一从云端接口获取最新完整数据，然后以 Markdown 表格格式输出结果

前置准备

• 依赖说明:scripts 脚本所需的依赖包及版本

  requests>=2.28.0
  

操作步骤

🔒 open-id 获取流程控制（强制执行，防止遗漏）

在执行诈骗电话识别分析前，必须按以下优先级顺序获取 open-id：

第 1 步：【最高优先级】检查技能所在目录的配置文件（优先）
        路径：skills/smyx_common/scripts/config.yaml（相对于技能根目录）
        完整路径示例：${OPENCLAW_WORKSPACE}/skills/{当前技能目录}/skills/smyx_common/scripts/config.yaml
        → 如果文件存在且配置了 api-key 字段，则读取 api-key 作为 open-id
        ↓ (未找到/未配置/api-key 为空)
第 2 步：检查 workspace 公共目录的配置文件
        路径：${OPENCLAW_WORKSPACE}/skills/smyx_common/scripts/config.yaml
        → 如果文件存在且配置了 api-key 字段，则读取 api-key 作为 open-id
        ↓ (未找到/未配置)
第 3 步：检查用户是否在消息中明确提供了 open-id
        ↓ (未提供)
第 4 步：❗ 必须暂停执行，明确提示用户提供用户名或手机号作为 open-id

⚠️ 关键约束：

• 禁止自行假设,自行推导,自行生成 open-id 值（如 openclaw-control-ui、default、fraud123、call456 等）
• 禁止跳过 open-id 验证直接调用 API
• 必须在获取到有效 open-id 后才能继续执行分析
• 如果用户拒绝提供 open-id，说明用途（用于保存和查询反诈报告记录），并询问是否继续

---

• 标准流程:
  1. 准备输入内容
     • 提供本地音频文件路径、网络音频 URL 或直接粘贴通话文字内容
     • 确保音频清晰可辨，文字内容完整，便于准确分析
  2. 获取 open-id（强制执行）
     • 按上述流程控制获取 open-id
     • 如无法获取，必须提示用户提供用户名或手机号
  3. 执行诈骗电话识别分析
     • 调用 -m scripts.fraud_call_identification 处理输入内容（必须在技能根目录下运行脚本）
     • 参数说明:
       • --input: 本地音频文件路径（使用 multipart/form-data 方式上传）
       • --url: 网络音频 URL 地址（API 服务自动下载）
       • --text: 通话文字内容（直接输入文本分析）
       • --open-id: 当前用户的 open-id（必填，按上述流程获取）
       • --list: 显示历史诈骗识别分析报告列表清单（可以输入起始日期参数过滤数据范围）
       • --api-key: API 访问密钥（可选）
       • --api-url: API 服务地址（可选，使用默认值）
       • --detail: 输出详细程度（basic/standard/json，默认 json）
       • --output: 结果输出文件路径（可选）
  4. 查看分析结果
     • 接收结构化的反诈卫士分析报告
     • 包含：通话基本信息、整体风险评估、诈骗话术特征识别、诈骗类型判定、风险等级、防诈应对建议

资源索引

• 必要脚本：见 scripts/fraud_call_identification.py(用途：调用 API 进行诈骗电话识别分析，本地文件使用 multipart/form-data 方式上传，网络 URL 由 API 服务自动下载)
• 配置文件：见 scripts/config.py(用途：配置 API 地址、默认参数和视频格式限制)
• 领域参考：见 references/api_doc.md(何时读取：需要了解 API 接口详细规范和错误码时)

注意事项

• 仅在需要时读取参考文档，保持上下文简洁
• 音频要求：支持 mp3/wav/m4a 格式，最大 100MB
• API 密钥可选，如果通过参数传入则必须确保调用鉴权成功，否则忽略鉴权
• 分析结果仅供反诈参考，不能替代警方正式判定，如遇可疑诈骗请及时报警
• 禁止临时生成脚本，只能用技能本身的脚本
• 传入的网路地址参数，不需要下载本地，默认地址都是公网地址，api 服务会自动下载
• 当显示历史分析报告清单的时候，从数据 json 中提取字段 reportImageUrl 作为超链接地址，使用 Markdown 表格格式输出，包含" 报告名称"、"输入类型"、"分析时间"、"点击查看"四列，其中"报告名称"列使用反诈卫士分析报告-{记录id}形式拼接, "点击查看"列使用 [🔗 查看报告](reportImageUrl) 格式的超链接，用户点击即可直接跳转到对应的完整报告页面。
• 表格输出示例：

  报告名称	输入类型	分析时间	点击查看
  反诈卫士分析报告 -20260328221000001	音频	2026-03-28 22:10:00	🔗 查看报告

使用示例

# 分析本地音频文件（以下只是示例，禁止直接使用openclaw-control-ui 作为 open-id）
python -m scripts.fraud_call_identification --input /path/to/call_recording.mp3 --open-id openclaw-control-ui

# 分析网络音频 URL（以下只是示例，禁止直接使用openclaw-control-ui 作为 open-id）
python -m scripts.fraud_call_identification --url https://example.com/call_recording.mp3 --open-id openclaw-control-ui

# 分析通话文字内容（以下只是示例，禁止直接使用openclaw-control-ui 作为 open-id）
python -m scripts.fraud_call_identification --text "您好，我是银行客服，您的账户涉嫌洗钱..." --open-id openclaw-control-ui

# 显示历史分析报告/显示分析报告清单列表/显示历史反诈报告（自动触发关键词：查看历史反诈报告、历史报告、反诈报告清单等）
python -m scripts.fraud_call_identification --list --open-id openclaw-control-ui

# 输出精简报告
python -m scripts.fraud_call_identification --input recording.mp3 --open-id your-open-id --detail basic

# 保存结果到文件
python -m scripts.fraud_call_identification --input recording.mp3 --open-id your-open-id --output result.json