Security audit

Video Finder

Security checks for vulnerabilities and agentic risk

Overview

This adult-video search and download skill is mostly coherent, but it stores very sensitive preferences, history, and proxy credentials while using broad triggers and background downloads that need careful review.

Review this before installing. Use it only on a private machine/account, assume local files may reveal adult-content interests and download history, avoid saving proxy credentials unless necessary, and check or delete generated preferences.json, proxy.json, history.md, and tracking/downloads.json when done. Be especially cautious with vague requests that could trigger the skill or start search/download workflows.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (18)

Vague Triggers

Medium

Confidence: 95% confidence
Finding: The README says the skill starts when the user says nearly any keywords and gives a generic trigger, which makes accidental activation likely during ordinary conversation. For a skill that searches and downloads adult content, unintended invocation raises the risk of surprise explicit material exposure and unintended collection of preferences/history.

Vague Triggers

Medium

Confidence: 96% confidence
Finding: The example invocation uses an everyday phrase ('找点东西看看') as a trigger, reinforcing that normal chat can activate the skill. In the context of adult-video search/download behavior, accidental triggering can lead to inappropriate content retrieval, privacy-sensitive logging, or downloads the user did not mean to start.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The README advertises automatic storage of user preferences, proxy settings, feedback, and history, but does not provide a clear privacy warning, retention policy, or consent model. Because this skill handles adult-content interests and network/proxy details, the stored data is highly sensitive and could expose sexual preferences, browsing habits, and infrastructure information if accessed by others.

Vague Triggers

Medium

Confidence: 94% confidence
Finding: The trigger list includes broad colloquial phrases such as '来片', '来个片', and '有没有片', which can match ordinary conversation and cause the skill to activate unintentionally. Because this skill performs adult-content search and download workflows, accidental activation materially increases the risk of inappropriate content handling, privacy issues, and policy-violating actions.

Vague Triggers

Medium

Confidence: 89% confidence
Finding: The description embeds an informal trigger list ('找片、搜片、下片、下载视频、JAV、Pornhub 等') without clear boundaries, making activation criteria ambiguous and overly permissive. In the context of a pornography search/download skill, this raises the chance of unintended invocation and increases safety, compliance, and user-consent risks.

Vague Triggers

High

Confidence: 88% confidence
Finding: The manifest uses very broad trigger phrases such as '下载', 'download', and similar common terms, which can cause the skill to activate in unrelated conversations. Because this skill performs background downloads and file operations, accidental invocation increases the chance of unintended network access, file writes, and persistence behavior without clear user intent.

Missing User Warnings

Medium

Confidence: 92% confidence
Finding: The skill automatically reuses the last successful download directory and proceeds with background file operations without a prominent warning or renewed user confirmation. In this context, that can lead to files being written into sensitive or unexpected locations, especially on shared systems or when prior state is stale or manipulated.

Vague Triggers

Medium

Confidence: 86% confidence
Finding: The trigger list includes very broad phrases such as “反馈、评价、觉得怎么样、好看吗”, which can overlap with ordinary conversation and cause the skill to activate unexpectedly. In this skill’s context, unexpected activation is more sensitive because it can prompt discussion of explicit adult content and associated preference tracking when the user may not have intended to engage this workflow.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The file explicitly describes reading and writing persistent preference data in `preferences.json` and recording activity in `history.md`, but does not require a clear user-facing notice or consent before storing this information. In the context of an adult-video skill, persisted tags, authors, and viewing feedback can reveal highly sensitive sexual interests, making accidental or non-consensual retention significantly more dangerous.

Vague Triggers

Medium

Confidence: 81% confidence
Finding: The trigger list includes very generic terms such as “preferences”, which can easily appear in ordinary conversation and unintentionally invoke this skill. In this skill’s context, accidental activation can lead to preference collection or modification workflows that touch local files, making misfires more consequential than a normal chat-only skill.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The reset trigger relies on short natural-language phrases like “改偏好” / “重新设置”, which are ambiguous and may be said casually without intending a destructive reset. Because the documented behavior overwrites the entire preferences.json, an accidental match can cause loss of stored settings and force re-collection of sensitive adult-content preferences.

Missing User Warnings

Medium

Confidence: 90% confidence
Finding: The workflow instructs the agent to write directly to preferences.json and create history.md, but it does not clearly warn the user that local files will be created or modified. In an adult-content skill, silent persistence is more sensitive because it stores intimate preference data and leaves local artifacts the user may not expect.

Missing User Warnings

High

Confidence: 94% confidence
Finding: The full reset behavior explicitly overwrites the entire preferences.json without requiring a warning or confirmation at the point of action. This is dangerous because a mistaken or ambiguous request can irreversibly replace existing configuration, including sensitive adult-preference settings and custom exclusions built over time.

Vague Triggers

Medium

Confidence: 86% confidence
Finding: The trigger list is broad enough that ordinary user messages about networking or connectivity could invoke this skill unintentionally. In this skill's context, accidental activation is more dangerous because it can lead to handling and persistence of sensitive proxy credentials, and may steer unrelated conversations into network probing behavior.

Missing User Warnings

Medium

Confidence: 92% confidence
Finding: The skill stores proxy credentials, including passwords, in a local file but does not explicitly warn the user before persistence or obtain informed consent at the point of storage. This increases the risk of credential exposure on shared systems, backups, or through later misuse, especially since the skill is designed to handle authenticated proxies for access to adult-content sites.

Vague Triggers

High

Confidence: 90% confidence
Finding: The trigger list includes broad generic terms such as '搜索' and '搜一搜', which can cause the adult-video skill to activate during ordinary, unrelated user requests. In this context, unintended activation is especially risky because the skill handles explicit sexual content and may initiate external searches on adult sites without a clear, deliberate opt-in.

Missing User Warnings

Medium

Confidence: 85% confidence
Finding: The skill reads and later updates persistent preference data, including sexual-content preferences, but does not clearly warn the user that these preferences may be stored and modified across sessions. Because the stored data concerns sensitive adult interests, silent persistence creates a meaningful privacy risk if the device, account, or storage is later accessed by others.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The design sends full user preferences and proxy configuration to multiple parallel subagents and external search/fetch tools without a clear privacy warning or data minimization controls. In this skill, those preferences can reveal highly sensitive sexual interests, and distributing them broadly increases the chance of logging, leakage, third-party retention, or misuse.

Static analysis

No suspicious patterns detected.