Midscene Automations Skills for HarmonyOS

Security checks across static analysis, malware telemetry, and agentic risk

Overview

The skill fits its HarmonyOS automation purpose, but it needs review because it can broadly control a connected device, including raw device shell commands, and may process screenshots through external model APIs.

Review this skill before installing if you will connect a real device. Prefer a dedicated test device and test accounts, trust and pin the Midscene package/provider, and instruct the agent to ask before any raw HDC shell command, deletion, account change, purchase, message sending, or other irreversible action.

Static analysis

No static analysis findings were reported for this release.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Risk analysis

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

#
ASI02: Tool Misuse and Exploitation
High
What this means

If used on a real device, the agent could run low-level commands or make device changes beyond simple visual testing, especially if the user gives an ambiguous task.

Why it was flagged

The skill gives the agent Bash access and documents an arbitrary raw HDC shell command path to the connected device, which is broader than scoped UI automation and is not paired with clear confirmation or command-scope limits.

Skill content
allowed-tools:\n  - Bash\n...\n### Run a Raw HarmonyOS Shell Command\n...\nnpx -y @midscene/harmony@1 runhdcshell --command "hidumper -s RenderService -a screen"\n\nThis is forwarded to `hdc shell` on the connected device.
Recommendation

Use only with trusted test devices, require explicit approval for raw shell commands or destructive actions, and avoid connecting personal devices with sensitive data.

#
ASI04: Agentic Supply Chain Vulnerabilities
Low
What this means

The user is relying on the external Midscene npm package and its updates to safely handle device control and screenshots.

Why it was flagged

The skill's main behavior depends on downloading and executing an external npm package at runtime. This is central to the stated purpose, but the package code is not included in the provided artifacts and the version is pinned only to major version `@1`.

Skill content
Automate HarmonyOS NEXT devices using `npx -y @midscene/harmony@1`.
Recommendation

Install only if you trust the Midscene package source, and consider pinning or reviewing the exact package version in controlled environments.

#
ASI03: Identity and Privilege Abuse
Low
What this means

A model API key may incur costs or grant access to the user's model provider account if mishandled.

Why it was flagged

The skill requires a model provider API key and related configuration. This is expected for Midscene's visual model use, but it is sensitive credential material and is not declared in the registry requirements.

Skill content
MIDSCENE_MODEL_API_KEY="your-api-key"\nMIDSCENE_MODEL_NAME="model-name"\nMIDSCENE_MODEL_BASE_URL="https://..."\nMIDSCENE_MODEL_FAMILY="family-identifier"
Recommendation

Use a limited-scope API key where possible, store it securely, and remove it from shared project directories or logs.

#
ASI07: Insecure Inter-Agent Communication
Medium
What this means

Screenshots can contain private messages, account details, app data, or other sensitive information that may be sent for model inference depending on the provider configuration.

Why it was flagged

The skill's visual automation relies on screenshots and external model configuration, implying that screen contents may be analyzed by the configured model provider. This is purpose-aligned, but users should be aware of the data boundary.

Skill content
Operates entirely from screenshots — no DOM or accessibility labels required.\n...\nMidscene requires models with strong visual grounding capabilities.
Recommendation

Use test accounts and test devices when possible, avoid displaying secrets during automation, and review the privacy terms of the configured model provider.