Tp2
- Category
- MCP Tool Poisoning
- Confidence
- 85% confidence
- Finding
Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.
- Content
Security audit
Security checks for vulnerabilities and agentic risk
This is a markdown-only video-editing reference skill with broad triggers, but it does not include code execution, sensitive data access, persistence, or privileged behavior.
Install only if you want a lightweight Chinese video-editing reference skill. Be aware that generic trigger words like video editing or 2025 may invoke it unexpectedly, but the inspected artifact does not perform actions on your system.
Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.
The manifest trigger string includes generic topic language and the year '2025年', which is too ambiguous for safe dispatch because many unrelated conversations may contain those words. This increases the chance of accidental invocation and misrouting within the agent framework.
The trigger conditions are broad enough to match ordinary user discussion such as mentioning '视频剪辑' or '2025年', which can cause unintended activation of the skill. In an agent system, overly generic triggers create prompt-selection confusion and can route users into irrelevant or lower-quality instructions without clear consent.
The nested skill section repeats broad trigger logic using generic help-request language and short topic phrases, which can activate during normal conversation rather than explicit selection. Repeated ambiguous triggers across merged sections compound the risk of incorrect skill resolution.
This merged section again uses generic requests like needing help on the topic, which are not safe boundaries for tool or skill activation. Such broad language can unintentionally pull this skill into unrelated sessions and degrade reliability of agent behavior.
The repeated ambiguous trigger language broadens the activation surface of the skill and makes false activation more likely, especially because the file contains multiple merged variants of similar names. In an orchestration environment, this can cause unintended context injection and selection of the wrong skill branch.
The final merged section still relies on ordinary topical discussion as an activation signal, so the agent may invoke the skill when a user merely mentions video editing rather than requesting this specific capability. Because the file is a merged aggregate with many overlapping sections, the contextual danger is higher than for a single isolated broad trigger.