Security audit
Security checks for vulnerabilities and agentic risk
This is a Chinese-language cognitive-bias coaching skill with no hidden code execution, data access, persistence, or exfiltration, though its broad triggers and persuasion content deserve user awareness.
Install only if you want a Chinese-language coaching workflow for cognitive-bias self-checks. Be aware it may activate on broad everyday phrases and that its communication/influence sections include persuasion tactics; use those only transparently and avoid relying on it as medical, legal, financial, or other licensed professional advice.
The trigger list is very broad and overlaps with common conversational terms like '学习', '怎么说', '该怎么', and '继续', making accidental activation likely. Overbroad activation can hijack unrelated conversations, cause unintended workflow changes, and route user input into a scripted diagnostic flow without clear user intent.
The README presents the skill name, description, installation verification phrase, and usage guidance entirely in Chinese, including instructing the user to trigger the skill with a Chinese phrase. This indicates a language-specific experience without any opt-in, fallback language, or explanation that the skill is intentionally limited to Chinese-speaking users.
Without declared permissions the skill's intent is opaque and cannot be validated.
The skill is written to operate in Chinese throughout and does not provide a language-negotiation step, which can override the user's preferred language and reduce transparency or comprehension. While not a direct exploit vector, forced language behavior can degrade user consent and increase the chance of misunderstanding in advice-oriented interactions.
The article instructs users to trigger the skill with a very generic phrase like “帮我决策一件事,” and the metadata lists other broad triggers such as decision, learning, procrastination, persuasion, and management topics. These overlap heavily with ordinary conversation, increasing the chance the skill activates unintentionally and intercepts user requests that were not meant for this specific skill.
The article directly instructs users to run installation commands from third-party sources (clawhub, skillhub, and git clone) without any warning about source verification, reviewing code, or trust boundaries. In a skill-distribution context, this can normalize blind execution of unverified commands and lead users to install malicious or trojanized packages or repositories if the referenced ecosystem, package name, or hosting source is compromised or spoofed.
The article says the skill will automatically intervene when users say common phrases such as '我在纠结一个决定' or '学某样东西总是学不会', which are broad enough to occur in ordinary conversation. Overly broad triggers can cause unintended invocation, exposing user content to the skill unexpectedly, producing unwanted guidance, or expanding the skill's access beyond what users reasonably intended.
The skill advertises invocation via very broad, everyday phrases such as '学东西总是记不住', '一直在拖延', and '怎么谈加薪' without defining clear trigger boundaries, exclusions, or confirmation steps. In an agent ecosystem, this can cause overbroad routing where normal conversation is unintentionally captured by the skill, leading to unsolicited behavioral guidance or manipulation-oriented outputs in sensitive contexts like workplace negotiation and influencing others.
This markdown file describes how to invoke the skill using very general natural-language phrases such as "帮我做一次决策自检" and "帮我检验一下" without stating boundaries, exclusions, or a narrow activation context. Because these phrases overlap with ordinary conversation, they could cause unintended invocation in systems that route skills from free-form user text.
This markdown file is entirely written as a user-facing skill article in Chinese, and it does not indicate that the user may choose another language or that the content is intentionally limited to a Chinese-only, region-specific audience. Under the policy rule for natural-language violations, forcing a specific language without user opt-in is a reportable issue.
This markdown file presents all user-facing guidance exclusively in Chinese, and there is no indication that the skill is region-specific or that users can opt into another language. Under the policy criteria, forcing a specific language without user choice is a natural-language policy violation.
The entire skill reference is written in Chinese and does not indicate any option for the user to choose another language or locale. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale constraint is explicitly documented and justified.
This markdown file contains natural-language guidance exclusively in Chinese, and there is no indication that users may choose another language or that the skill is intentionally limited to a Chinese-language audience. Under the stated policy, forcing a specific language without user opt-in is a natural-language policy concern.
The file is entirely in Chinese and provides no mechanism to detect or accommodate a user's preferred language. This can exclude or mislead users who cannot read Chinese, causing consent, comprehension, and accessibility problems in a skill that asks users to make self-assessment choices and follow behavioral advice.
The file explicitly teaches persuasive use of loss framing and social proof as communication tactics, but does not place meaningful operational limits on when these methods become manipulative. In a skill framed as an advisor for influencing others, this can normalize coercive persuasion, especially against vulnerable users or in high-stakes contexts like sales, workplace power dynamics, or family relationships.
The 'three-step persuasion' script is a structured compliance-escalation pattern: lower defenses through validation, apply social proof and loss framing, then secure a small commitment to increase follow-through. Even though the document mentions general 'bottom lines,' this section operationalizes a reusable manipulation playbook that can be abused for undue influence, especially because it is packaged as a generic method without strong safeguards or disallowed scenarios.
This file explicitly guides users through high-stakes, partly irreversible decisions such as surgery, resigning, buying a home, and investing, yet it presents a structured decision framework without a clear boundary that it is not a substitute for licensed professional advice. Users may over-trust the framework and make unsafe medical, legal, or financial decisions based on cognitive-bias prompts alone, especially because the content sounds systematic and authoritative.
Although the file includes a brief 'three bottom lines' banner, the body of the skill still normalizes behavior-change tactics and provides examples for increasing compliance without sustained discussion of consent, power imbalance, or manipulation risk. Users are taught how to influence choices, but not given robust safeguards for when such techniques become coercive or deceptive, especially in relationships, management, or negotiations.
The diagnostic encourages users to give something small or build rapport first so later requests are more likely to succeed, operationalizing reciprocity and priming rather than merely explaining the bias. This is risky because it can be used to engineer obligation and lower defenses, especially in contexts where the target may not recognize that the initial 'value' or relationship-building is instrumental to a later ask.
This section moves beyond reflective cognitive-bias advice into operational playbooks for changing another person's behavior, including defaults, staged commitments, and scarcity/social-proof combinations. Even with the stated 'three bottom lines,' the content gives reusable persuasion tactics that can be applied manipulatively in sales, workplace pressure, or interpersonal coercion, especially because it frames influence as designing others' choice environments.
This passage explicitly teaches a foot-in-the-door escalation pattern: obtain a small commitment, then leverage consistency pressure to secure a larger one. That is a classic manipulation technique because it exploits a known cognitive bias to reduce resistance over time, and can be particularly harmful when used by managers, recruiters, sellers, or abusive partners to manufacture consent that would not be freely given upfront.
This markdown file presents all user-facing prompts and options in Chinese, which can effectively force a specific language on users. The policy allows locale constraints only when users are given a choice or when the restriction is explicitly documented and justified, neither of which appears here.
All natural-language prompts and expected behaviors in the file are specified in Chinese, and there is no indication that users are offered a language or locale choice. Under the policy rule, forcing a specific language without opt-in can be a natural-language policy violation unless the locale restriction is explicitly justified.
The trigger example uses an extremely generic phrase like “决定”, which is common in ordinary conversation and can cause unintended activation of the skill. In a multi-skill or agentic environment, this can hijack benign conversations into a structured diagnostic flow, producing unwanted state changes, confusing outputs, or accidental collection of sensitive personal context.
The manifest says the skill covers six domains and routes user questions to domain files for diagnosis and strategy output, but these test cases expect persistent session recovery features such as continuing from prior progress, restarting, and resuming from a specified question. That is a materially broader behavior than simple domain routing and advisory output, because it implies conversation state management and progress lookup not disclosed in the manifest description.
No suspicious patterns detected.
