抖音违禁词检测

抖音/短视频违禁词和敏感词检测（本地词库版，无需 API Key）。每天首次使用自动从 GitHub 开源词库更新本地缓存，离线检测文案合规性。支持多词库合并（广告极限词、平台限流词、暴恐、色情、涉枪涉爆等）。使用场景：(1) 生成短视频文案后自动检测违禁词，(2) 用户要求检查某段文字是否有问题，(3) 抖音/...

Security Scan

The skill's code, instructions, and network usage are consistent with a local sensitive-word checker that updates wordlists from public GitHub raw URLs; it does not request unrelated credentials or perform unexpected actions.

Purpose & Capability

Name/description (Douyin sensitive-word detection) match the implementation: Python scripts perform local substring matching and merge open-source wordlists. Required binary is python3 which is appropriate.

Instruction Scope

SKILL.md instructions and scripts stay within scope: reading/writing a local data/ cache, performing substring matching, and (on daily-first-use or --update) fetching raw text files from the three listed GitHub repositories. The skill does not read unrelated system files, environment variables, or transmit user content to external endpoints.

Install Mechanism

No install spec; code is bundled as plain Python scripts. The only network access is to raw.githubusercontent.com (explicitly declared). No downloads from opaque URLs or archive extraction are present.

Credentials

The skill requests no environment variables or credentials. Its file writes are limited to its own data/ directory under the skill path, which is proportionate to maintaining a local cache.

Persistence & Privilege

always is false, the skill only writes to its own data/ directory and update state file. It does not modify other skills or system-wide settings.

Assessment

This skill appears coherent and implements what it claims: local detection plus optional daily-first-use updates from public GitHub raw files. Before installing, consider: (1) the skill will make outbound requests to raw.githubusercontent.com on first use each day to fetch upstream wordlists — if you require strict offline use, populate data/sensitive_words.txt and set data/.update_state.json last_update to a future date or avoid running updates; (2) the upstream sources are public GitHub repos listed in the SKILL (inspect them yourself if you want to verify content and license); (3) the skill writes its cached wordlist and a small state file under the skill's data/ directory — ensure you trust the skill directory owner; and (4) because the skill is from an unknown registry owner, prefer to review the bundled scripts yourself or run them in an isolated environment if you have any doubt.

抖音违禁词检测 Skill（开源词库版）

本地词库 + 每日自动更新，无需 API Key，离线可用。

脚本路径

scripts/
  check.py         # 主检测脚本（入口）
  update_words.py  # 词库更新模块（每天首次自动触发）
data/              # 运行时生成，词库缓存目录（.gitignore 排除）
  sensitive_words.txt
  .update_state.json

常用命令

SKILL=~/.agents/skills/douyin-sensitive-check

# 检测一段文案
python3 $SKILL/scripts/check.py "今天给大家推荐史上最好用的护肤品，加我微信领优惠券"

# 检测文件
python3 $SKILL/scripts/check.py -f /path/to/script.txt

# 管道
echo "文案内容" | python3 $SKILL/scripts/check.py

# 强制更新词库
python3 $SKILL/scripts/check.py --update

# 查看词库状态
python3 $SKILL/scripts/check.py --status

工作流

1. 每天首次运行 → 自动调用 update_words.py 从 3 个 GitHub 开源词库拉取最新内容合并
2. 加载本地 data/sensitive_words.txt（去重合并，含数万词条）
3. 对输入文案做全文子串匹配（长词优先）
4. 输出：🔴 违禁词（必改）/ 🟡 广告极限词（建议改）+ 上下文标注
5. 根据结果帮用户改写文案，改完后再次检测直到通过

词库来源

• konsheng/Sensitive-lexicon：广告、政治、暴恐、色情、涉枪涉爆、补充词库
• bigdata-labs/sensitive-stop-words：广告、政治、色情、涉枪涉爆
• jkiss/sensitive-words：广告、政治、色情

更新机制

• data/.update_state.json 记录最后更新日期
• 每天第一次使用自动触发，当天内后续使用直接读缓存
• 网络失败时保留本地缓存，不影响使用
• 手动强制更新：--update

重要提示

• 开源词库以通用违禁词为主，抖音平台的部分特有限流词（如"私信"、"加微信"）已内置在 check.py 的 CATEGORY_PATTERNS 中补充
• 匹配策略是子串匹配，可能有误报；如需精确匹配可编辑 data/sensitive_words.txt 删除误报词
• 改写建议：被标注词优先用谐音、符号分割、同义替换等方式规避