Trap Detector
Scan 8 signals to detect pump-and-dump scams and rate risk from safe to highly suspicious.
The skill explicitly requires evidence URLs and checklists, no implicit execution commands; however, no permission boundaries, no user confirmation steps, and external searches (DuckDuckGo) and domestic platform scraping (Weibo/Douyin) present data-flow transparency risks, no rollback. Deduction: missing permissions and confirmation mechanisms.
Clear steps and output format, but depends on external data sources (web search, fetch_financials etc.) whose availability is uncertain; no handling of abnormal input or failure feedback; tests only cover renderers, not detection logic. Deduction: happy path plausible but tests and error handling thin.
Trigger conditions clear and Chinese keyword support, but no non-fit scenarios or boundary conditions; relies on overseas services (DuckDuckGo) potentially unreachable on mainland networks. Deduction: environment fit incomplete.
Structured docs, version, author, license present, but missing install/dependency notes (how to obtain fetch_* functions), no FAQ/known limitations, changelog absent. Deduction: hidden assumptions and governance incomplete.
Task defined clearly, output format specified, but no actual run results or sample outputs to verify direct usability. Deduction: no evidence supporting output usability.
Checklists and signal definitions exist, but no tests or third-party verification; relies on author descriptions. Deduction: lack of reproducible verification.
- This skill depends on external search (DuckDuckGo) and domestic platform scraping, which may fail due to network restrictions; ensure reachability or provide alternatives.
- No user confirmation or rollback mechanisms; recommend permission review before use.
- Detection logic lacks test coverage; verify reliability in real environment.
What does this skill do, and when should you use it?
Trap Detector is a dedicated skill within the UZI-Skill repository that detects typical 'pump-and-dump' (杀猪盘) manipulation patterns in A-shares. When users mention keywords like 'friend recommended', 'group chat said', 'teacher advises', or 'insider news', it triggers an 8-signal scan covering mass low-quality account recommendations, templated hype phrases, paid community funneling, fundamental-valuation disconnect, unusual K-line movements, guru persona promotion, cross-platform coordinated promotion, and fake research reports. Based on the number of signals hit, it outputs a risk rating from 🟢 safe to 🔴 highly suspicious. The skill gathers evidence via web search and produces a structured JSON report with specific evidence URLs and clear recommendations. It is part of the larger UZI-Skill repository (MIT licensed), installed via plugin marketplace, requiring Python and network access.
This skill activates when user input contains trigger phrases like 'friend recommended', 'group chat said', or 'insider news'. It performs 8 signal scans: 1) searches for homogeneous low-engagement accounts recommending the stock; 2) searches for templated phrases like 'about to explode' or 'main force building positions'; 3) searches for WeChat groups, live streams, VIP rooms for paid promotion; 4) calls fetch_financials and fetch_sentiment to compare fundamentals with hype; 5) calls fetch_kline to check for abnormal price surges before the recommendation wave; 6) searches for 'name teacher' or 'name stock god' persona promotion; 7) searches multiple platforms (Xiaohongshu, Douyin, Bilibili, Zhihu) for cross-platform coordination; 8) searches for 'name rumor', 'name denial', 'name fake' for false information. Each signal is marked as hit/miss/insufficient data, then aggregated into a risk rating, output as JSON including ticker, trap_score (1-10, higher is safer), detailed hit signals with evidence URLs and severity, user keyword boost, recommendation, and warning phrases.
- An investor receives a 'guaranteed rise' tip from a friend or group chat and wants to quickly verify its safety.
- A trader sees a stock heavily promoted on social media and suspects a pump-and-dump, requesting a risk check.
- A user executes `/scan-trap 002273` command to run a one-click scam screen.
- An analyst uses trap-detector as a pre-screening step before deep analysis to exclude suspicious tickers.
- A financial content creator checks a stock for manipulation signs before recommending it to followers.
- A casual investor encounters exaggerated 'no-risk profit' claims and wants to know if fake reports or fabricated news are involved.
What are this skill's strengths and limitations?
- Covers 8 typical pump-and-dump signals with comprehensive detection dimensions
- Outputs structured JSON with evidence URLs for each signal, enabling verification
- User keywords automatically boost risk severity for higher sensitivity
- Intuitive risk rating (traffic light) with clear recommendations
- Integrates with other UZI-Skill analysis tools for a complete workflow
- Relies on web search quality; DuckDuckGo Chinese search can return irrelevant results, affecting evidence reliability
- Requires network connectivity, and some data sources (e.g., Xiaohongshu) may need extra configuration
- Primarily designed for A-shares; support for Hong Kong/US stocks is limited
- No standalone test suite; the skill itself is untested in isolation
- Some commands require additional setup or configuration (e.g., Playwright fallback)
How do you install this skill?
Trap Detector is part of the UZI-Skill repository; first install the whole repo. For Claude Code: /plugin marketplace add wbh604/UZI-Skill, then /plugin install stock-deep-analyzer@uzi-skill. After installation, trap-detector is automatically included. Alternatively, git clone the repo and configure the Python environment as per README.
How do you use this skill?
Once installed, simply tell Claude Code: 'Check if this stock is a pump-and-dump' or 'scan for risk', or type a message containing keywords like 'friend recommended', 'group chat said', 'insider news'. You can also use the command /stock-deep-analyzer:scan-trap <ticker> or /trap-detector <ticker> (with the stock-deep-analyzer: prefix). The skill will execute the 8-signal scan and return a JSON rating report.
How does this skill compare with similar options?
As part of the UZI-Skill package, trap-detector complements other skills (like deep-analysis) but has no separate alternative. The UZI-Skill repo broadly references methodology from anthropics/financial-services-plugins, but trap-detector focuses on fraud detection rather than valuation.