Security audit
Security checks for vulnerabilities and agentic risk
This skill handles intimate home audio/video but uses cloud APIs, silent identity persistence, and a default plaintext development endpoint that users should review carefully before installing.
Install only if you are comfortable with household audio/video being processed by a remote service and with local account/token persistence. Before use, require a production HTTPS-only configuration, explicit consent from affected household members, clear retention/deletion controls for reports and tokens, and confirmation that raw microphone audio is not uploaded contrary to the documentation.
skills/smyx_common/scripts/config-dev.yaml:1Highly Sensitive Audio/Video and Authentication Data Are Transmitted over Plaintext HTTP
The description of local conflict-intensity analysis conflicts with behaviors involving HTTP requests, token management, workspace file access, user identity resolution, and persistence. In context, this is more dangerous than a normal documentation bug because the data concerns private family disputes and in-home surveillance, so any undisclosed remote transfer or retention materially raises privacy and misuse risk.
The description of local conflict-intensity analysis conflicts with behaviors involving HTTP requests, token management, workspace file access, user identity resolution, and persistence. In context, this is more dangerous than a normal documentation bug because the data concerns private family disputes and in-home surveillance, so any undisclosed remote transfer or retention materially raises privacy and misuse risk.
The description of local conflict-intensity analysis conflicts with behaviors involving HTTP requests, token management, workspace file access, user identity resolution, and persistence. In context, this is more dangerous than a normal documentation bug because the data concerns private family disputes and in-home surveillance, so any undisclosed remote transfer or retention materially raises privacy and misuse risk.
The description of local conflict-intensity analysis conflicts with behaviors involving HTTP requests, token management, workspace file access, user identity resolution, and persistence. In context, this is more dangerous than a normal documentation bug because the data concerns private family disputes and in-home surveillance, so any undisclosed remote transfer or retention materially raises privacy and misuse risk.
The description of local conflict-intensity analysis conflicts with behaviors involving HTTP requests, token management, workspace file access, user identity resolution, and persistence. In context, this is more dangerous than a normal documentation bug because the data concerns private family disputes and in-home surveillance, so any undisclosed remote transfer or retention materially raises privacy and misuse risk.
The description of local conflict-intensity analysis conflicts with behaviors involving HTTP requests, token management, workspace file access, user identity resolution, and persistence. In context, this is more dangerous than a normal documentation bug because the data concerns private family disputes and in-home surveillance, so any undisclosed remote transfer or retention materially raises privacy and misuse risk.
The description of local conflict-intensity analysis conflicts with behaviors involving HTTP requests, token management, workspace file access, user identity resolution, and persistence. In context, this is more dangerous than a normal documentation bug because the data concerns private family disputes and in-home surveillance, so any undisclosed remote transfer or retention materially raises privacy and misuse risk.
The description of local conflict-intensity analysis conflicts with behaviors involving HTTP requests, token management, workspace file access, user identity resolution, and persistence. In context, this is more dangerous than a normal documentation bug because the data concerns private family disputes and in-home surveillance, so any undisclosed remote transfer or retention materially raises privacy and misuse risk.
The description of local conflict-intensity analysis conflicts with behaviors involving HTTP requests, token management, workspace file access, user identity resolution, and persistence. In context, this is more dangerous than a normal documentation bug because the data concerns private family disputes and in-home surveillance, so any undisclosed remote transfer or retention materially raises privacy and misuse risk.
The description of local conflict-intensity analysis conflicts with behaviors involving HTTP requests, token management, workspace file access, user identity resolution, and persistence. In context, this is more dangerous than a normal documentation bug because the data concerns private family disputes and in-home surveillance, so any undisclosed remote transfer or retention materially raises privacy and misuse risk.
The description of local conflict-intensity analysis conflicts with behaviors involving HTTP requests, token management, workspace file access, user identity resolution, and persistence. In context, this is more dangerous than a normal documentation bug because the data concerns private family disputes and in-home surveillance, so any undisclosed remote transfer or retention materially raises privacy and misuse risk.
The description of local conflict-intensity analysis conflicts with behaviors involving HTTP requests, token management, workspace file access, user identity resolution, and persistence. In context, this is more dangerous than a normal documentation bug because the data concerns private family disputes and in-home surveillance, so any undisclosed remote transfer or retention materially raises privacy and misuse risk.
The description of local conflict-intensity analysis conflicts with behaviors involving HTTP requests, token management, workspace file access, user identity resolution, and persistence. In context, this is more dangerous than a normal documentation bug because the data concerns private family disputes and in-home surveillance, so any undisclosed remote transfer or retention materially raises privacy and misuse risk.
The YARA hit suggests manifest metadata contains suspicious formatting or poisoning indicators. While the matched content here is limited, metadata poisoning in a skill manifest can manipulate downstream tooling, parsers, or agent behavior and is more concerning because the skill already has broad undeclared capabilities and sensitive-data scope.
---
name: "smyx-family-conflict-intensity-detect-analysis"
description: "Using a fixed camera with microphone in the living room, the system analyzes audio and video in real time, detecting sound intensity (dB) and the intensity of body movements (e.g., rapid hand waving, finger pointing, pushing, throwing objects). It comprehensively evaluates the family conflict intensity level (low / medium / high). | 通过客厅固定摄像头（含麦克风），实时分析音频和视频，检测声音强度（分贝）和肢体动作激烈程度（如快速挥手、戳指、推搡、摔物等）。综合评估家庭争吵的冲突强度等级（低/中/高）
The skill silently initializes, reuses, or auto-creates an internal default user identity unrelated to the stated analysis purpose. For sensitive family-conflict data, hidden identity creation and linkage can enable persistent profiling, cross-session correlation, and report access without meaningful user awareness.
The notes claim aggressive-word detection is local-only and raw audio must not be uploaded, yet the skill accepts network URLs for an external API service to download and analyze media. This contradiction is dangerous because users may rely on a local-processing privacy promise that is not actually preserved when remote URL-based analysis is used.
The tool accepts local files or remote URLs for audio/video analysis and forwards them into skill.get_output_analysis(...) without any explicit warning about potential transmission, storage, or third-party processing. Because the skill handles highly sensitive in-home surveillance data about family conflict, silent upload or remote processing could expose intimate audio/video content and create serious privacy harm.
The utility layer silently performs external account login/registration and persists returned user/token data locally, which materially exceeds the skill's declared purpose of detecting family conflict intensity from local audio/video. This creates undisclosed identity provisioning and data flow to remote services, increasing privacy, compliance, and account abuse risk if the skill runs in sensitive home-monitoring contexts.
The skill declares no explicit tool scope even though the documentation directs use of shell, file I/O, environment-derived identity, and network access. In an agent environment, that under-specification can lead to over-privileged execution and makes it harder to enforce least privilege or detect when the skill is exceeding its stated purpose.
The manifest frames the skill as real-time conflict analysis with reminders, but the documentation expands the scope to cloud history queries and report-link retrieval. That scope creep matters because it changes the privacy model from ephemeral detection to stored, retrievable domestic-conflict records.
The default trigger condition is broad enough to invoke the skill for essentially any uploaded living-room audio/video. In this context, over-triggering is risky because it may cause sensitive household media to be saved, uploaded, or linked to history without a specific, informed request for conflict analysis.
The keyword trigger list contains broad and ambiguous phrases that can cause accidental invocation. Because the skill handles sensitive family-conflict surveillance data, unintended activation increases the chance of privacy-invasive processing based on casual or non-consensual inputs.
The workflow says uploaded attachments are automatically saved locally, but that storage behavior is not highlighted up front in the skill description. For intimate household audio/video, undisclosed local persistence increases the risk of residual sensitive data exposure on the agent host.
The output heading is hard-coded in Chinese ("📊 夫妻/家庭争吵强度识别结构化结果") with no indication that the skill is region-specific or that users can opt into this locale. This is a natural-language policy concern because it imposes a specific language on all users without choice or justification.
The script includes a hidden --list mode that retrieves prior analysis records via show_analyze_list(open_id), which is outside the stated purpose of real-time conflict-intensity detection. Undocumented data-retrieval functionality increases the risk of unauthorized access to sensitive household surveillance results and weakens user consent and transparency.
Detected: suspicious.install_untrusted_source