Security audit
Security checks for vulnerabilities and agentic risk
This driving-video analysis skill is coherent in purpose, but it silently creates and reuses user identities, stores tokens locally, and sends user media/history requests to a cloud service without clear user control.
Install only if you are comfortable sending driver videos or video URLs to lifeemergence.com services and allowing the skill to create or reuse an internal identity, query cloud-stored report history, and persist account tokens in a local workspace SQLite database. Prefer a version that clearly documents identity, token storage, retention, deletion, and consent behavior before processing sensitive footage.
The combination of external API requests, token/header handling, user-account creation or retrieval, file I/O, and workspace-context detection goes well beyond expected single-purpose video analysis. In a skill that processes potentially sensitive driver footage, undisclosed identity and workspace handling makes misuse, data leakage, and unintended linkage materially more dangerous.
The combination of external API requests, token/header handling, user-account creation or retrieval, file I/O, and workspace-context detection goes well beyond expected single-purpose video analysis. In a skill that processes potentially sensitive driver footage, undisclosed identity and workspace handling makes misuse, data leakage, and unintended linkage materially more dangerous.
The combination of external API requests, token/header handling, user-account creation or retrieval, file I/O, and workspace-context detection goes well beyond expected single-purpose video analysis. In a skill that processes potentially sensitive driver footage, undisclosed identity and workspace handling makes misuse, data leakage, and unintended linkage materially more dangerous.
The combination of external API requests, token/header handling, user-account creation or retrieval, file I/O, and workspace-context detection goes well beyond expected single-purpose video analysis. In a skill that processes potentially sensitive driver footage, undisclosed identity and workspace handling makes misuse, data leakage, and unintended linkage materially more dangerous.
The combination of external API requests, token/header handling, user-account creation or retrieval, file I/O, and workspace-context detection goes well beyond expected single-purpose video analysis. In a skill that processes potentially sensitive driver footage, undisclosed identity and workspace handling makes misuse, data leakage, and unintended linkage materially more dangerous.
The combination of external API requests, token/header handling, user-account creation or retrieval, file I/O, and workspace-context detection goes well beyond expected single-purpose video analysis. In a skill that processes potentially sensitive driver footage, undisclosed identity and workspace handling makes misuse, data leakage, and unintended linkage materially more dangerous.
The combination of external API requests, token/header handling, user-account creation or retrieval, file I/O, and workspace-context detection goes well beyond expected single-purpose video analysis. In a skill that processes potentially sensitive driver footage, undisclosed identity and workspace handling makes misuse, data leakage, and unintended linkage materially more dangerous.
The combination of external API requests, token/header handling, user-account creation or retrieval, file I/O, and workspace-context detection goes well beyond expected single-purpose video analysis. In a skill that processes potentially sensitive driver footage, undisclosed identity and workspace handling makes misuse, data leakage, and unintended linkage materially more dangerous.
The combination of external API requests, token/header handling, user-account creation or retrieval, file I/O, and workspace-context detection goes well beyond expected single-purpose video analysis. In a skill that processes potentially sensitive driver footage, undisclosed identity and workspace handling makes misuse, data leakage, and unintended linkage materially more dangerous.
YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).
---
name: "drive-analysis"
description: "Analyzes videos of vehicle drivers to identify unsafe driving behaviors. It generates professional analysis reports to help enhance road safety awareness. | 安全驾驶行为分析工具，针对机动车驾驶人员的驾驶行为进行视频分析，识别不安全驾驶行为，输出专业分析报告，提升道路交通安全意识"
version: "1.0.21"
license: "MIT-0"
---
# 🚗 Safe Driving Behavior Analyzer | 安全驾驶行为分析工具
> **智能分析中枢** · 图片/视频智能分析 · 结构化报告 · 历史报告云端查询
---
## 🧭 技能概览 | Ove
This file defines a generic user DAO and stores identity and authentication-related fields such as username, realname, email, token, and open_token, which are not justified by the stated purpose of driver-behavior video analysis. In a least-privilege review, unnecessary credential/PII storage increases privacy, breach, and misuse risk, especially because local SQLite storage is created automatically.
The model implements local persistence for personal data and tokens despite the skill being described as a road-safety video analysis tool. This capability mismatch is dangerous because it expands the attack surface and enables covert or unnecessary collection of sensitive data unrelated to the advertised functionality.
The code generates synthetic user identities, reads identity from a workspace file, and persists fallback identities in a local database. Creating or reusing identities unrelated to the skill's declared function can mask accountability and trigger network actions under automatically selected accounts, which is dangerous in an agent environment.
This utility file performs account lookup/provisioning, token management, and authenticated API access that are not necessary for a driver-video analysis skill's stated purpose. Such hidden identity and network side effects expand the trust boundary, can cause undisclosed data transmission, and may let the skill act under persistent credentials without informed user consent.
The skill advertises and instructs use of shell, network, local file, and environment-backed capabilities, but the manifest does not declare any explicit tool scope or allowed-tools constraints. That makes the skill overprivileged by default and weakens reviewability, increasing the chance that a caller or agent executes sensitive operations not clearly authorized by the manifest.
The skill automatically triggers cloud history-report queries based on keywords, yet provides no clear privacy warning or user-consent checkpoint for accessing cloud-stored historical reports. Because the skill also describes internal automatic identity association, this can expose prior report metadata without sufficiently transparent user intent confirmation.
The skill says uploaded attachments or video/image files are automatically saved locally, but it does not clearly warn the user about local writes, storage location, or retention. For media that may contain faces, driving habits, and other sensitive personal data, silent persistence increases privacy and data-handling risk.
The script's help text, errors, progress messages, and formatted results are consistently hard-coded in Chinese, including CLI descriptions and runtime output. Because the file provides no user opt-in, locale selection, or justification that the tool is China-specific only, this is a natural-language locale policy violation.
The skill advertises video driving-behavior analysis, but the CLI also performs hidden identity resolution and can enumerate a user's prior analysis records via the --list path. That expands the data-access scope beyond the stated purpose and can expose user-linked history or metadata without clear disclosure or explicit consent, especially because open-id handling is suppressed from help output.
Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.
result_json = JsonUtil.parse(result_json_pure_text, result_json_pure_text)
result_json_common_ai_response = result_json.get("commonAiResponse") if isinstance(result_json,
dict) else result_json
if result_json_common_ai_response:
result_json = result_json_common_ai_response
Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.
result_json = JsonUtil.parse(result_json_pure_text, result_json_pure_text)
result_json_common_ai_response = result_json.get("commonAiResponse") if isinstance(result_json,
dict) else result_json
if result_json_common_ai_response:
result_json = result_json_common_ai_response
The manifest describes a tool for analyzing videos of vehicle drivers, which naturally implies processing supplied video content, but this implementation also directly accepts arbitrary HTTP(S) URLs as input sources. Fetching and processing remote resources is an additional capability not stated in the manifest and is not clearly required by the declared purpose.
The skill reads arbitrary local file contents or forwards remote video URLs directly to the analysis API without any visible consent prompt, disclosure, or confirmation in this code path. This can cause unintentional exfiltration of sensitive local media or third-party URLs to an external service, especially when invoked by an agent on behalf of a user.
The add, edit, and delete methods provide generic remote modification primitives that are not inherently limited to the stated purpose of analyzing driving-behavior videos. In a skill ecosystem, such broad write/delete functionality can be abused to alter or remove remote data through arbitrary backend endpoints if higher layers pass attacker-influenced URLs or payloads.
This file exposes generic arbitrary HTTP wrappers (http_post, http_put, http_get, http_delete) that can call caller-supplied URLs, making the skill usable as a general-purpose network client rather than a narrowly scoped driver-video analysis component. In an agent/skill environment, this broad capability increases the attack surface for unauthorized outbound requests, data exfiltration, SSRF-style access to internal services, or policy bypass if untrusted inputs can influence the URL or request parameters.
Detected: suspicious.install_untrusted_source