Image 3D Scene Reconstruction | 图像3D场景重建
PassAudited by ClawScan on May 7, 2026.
Overview
This skill is a coherent local 3D reconstruction instruction skill with no evidence of hidden malicious behavior, but it relies on external Python packages/models and referenced project files that are not included or pinned.
This appears safe as an instruction-only skill, but it is incomplete as packaged. Before using it, verify any external project files, Python packages, and pretrained model sources, then run the reconstruction commands only on images and output folders you intend to process.
Findings (3)
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.
Following the quick start may fail or may require obtaining and trusting additional code outside the reviewed skill package.
The quick start depends on a local helper script/project setup, but the provided manifest contains only SKILL.md and _meta.json, so that helper implementation is not reviewable from these artifacts.
python3 scripts/reconstruct.py --input photo.jpg --output output/
Verify the source of any missing helper scripts or project files before running them, and prefer a packaged skill with documented, pinned setup steps.
Running the commands will execute third-party Python code that reads input images and writes reconstruction outputs.
The skill documents Python command execution. This is expected for a local ML reconstruction workflow and is presented as a user-directed example, not hidden automatic execution.
python3 -m depth_anything_3.cli image photo.jpg --export-dir output/ --export-format glb
Run the commands only in a trusted environment, ideally inside a virtual environment, and review input/output paths before execution.
Users may download or install external ML packages/models whose exact versions and provenance are not established by this skill package.
The skill relies on an external pretrained model and listed third-party packages, but the artifacts provide no pinned versions, install source, lockfile, or provenance details.
model = DepthAnything3.from_pretrained('depth-anything/DA3Metric-Large')Confirm the model and package sources, pin versions where possible, and follow the upstream project’s trusted installation instructions.