Skillv1.0.0

ClawScan security

travel-destination-brochure · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

BenignFeb 11, 2026, 9:41 AM

Verdict: benign
Confidence: medium
Model: gpt-5-mini
Summary: The code and runtime instructions are consistent with the stated purpose (fetch public images, build a manifest, and call the vlmrun CLI to generate video/itineraries), but the install instructions ask users to run remote installer scripts and there are a few operational risks to be aware of.
Guidance: This skill appears to do what it says: fetch public images for a city, save them locally, and call the vlmrun CLI (if you provide its API key) to generate a video and itinerary. Before installing/running: 1) Avoid blindly running the suggested remote installer commands (curl | sh or PowerShell irm | iex). Prefer installing packages from your normal package manager (pip install uv) or review the install script from https://astral.sh before executing it. 2) Verify the origin and trustworthiness of the 'vlmrun' CLI/package you install (it will send image data to an external service). 3) Run the scripts inside an isolated virtual environment or sandbox to limit accidental system changes and to control network/file access. 4) Be aware images are downloaded to disk (check output directories) and may contain large files; check disk/quota. 5) If you care about privacy, note that images and the images themselves are sent to the external vlmrun service when generating video/plan; do not provide private or sensitive imagery. If you want more confidence, you can audit the Python files (they are included) and test them with --no-download or --skip-commons/--skip-osc to observe behavior without external downloads.

Review Dimensions

Purpose & Capability: okThe name and description match the actual files and behavior: scripts geocode a city (Nominatim), download public images from Wikimedia Commons and OpenStreetCam, assemble them into an images/ manifest, and call the vlmrun CLI to generate video and a travel plan. Required credentials are limited to an optional VLMRUN_API_KEY for vlmrun; no unrelated cloud credentials or secrets are requested.
Instruction Scope: noteSKILL.md instructs the user to create a venv, install dependencies, download images, and run the provided scripts. The runtime steps stay within the stated purpose and the scripts only read/write files within the output directories. One noteworthy instruction-level detail: SKILL.md suggests running remote install commands (curl | sh and PowerShell irm | iex) to install 'uv' — see install_mechanism. The docs also instruct users how to set VLMRUN_API_KEY in environment or .env. The scripts do not attempt to read other environment variables or system config paths.
Install Mechanism: concernThere is no formal install spec, but SKILL.md recommends installing the 'uv' tool using remote installer scripts (https://astral.sh/uv/install.sh and install.ps1) with curl/PowerShell piped to shell/iex. Executing remote installers with ExecutionPolicy Bypass or piping shell scripts from the network is a higher-risk operation and should be treated with caution. The rest of the dependencies are normal Python packages (requests, vlmrun[cli]).
Credentials: okOnly an optional VLMRUN_API_KEY (and optional VLMRUN_BASE_URL in .env_template) is referenced to enable vlmrun-based video/plan generation. No other secrets are requested. The scripts themselves do not read arbitrary environment variables beyond what the user supplies for vlmrun; they default to public APIs for image fetching.
Persistence & Privilege: okThe skill is instruction-only and does not request always:true. It creates local files and suggests creating a virtual environment (.venv) inside the skill directory — normal behavior. It does not modify other skills or system-wide agent settings.