SynthClaw

Render Blender files with agent-controlled procedural parameters for synthetic data generation. Use when generating training data with controlled variations,...

This skill appears to implement the documented Blender rendering functionality, but there are a few mismatches you should address before installing or running it in a productive environment: - Ensure Blender is installed and available on PATH (SKILL.md and the code require the 'blender' binary, but the registry metadata omitted this). The skill will fail or behave unexpectedly without Blender 4.0+. - The package declares optional metrics dependencies (granatpy, lpips). If you enable compute_metrics, you must install these Python packages and their native dependencies (LPIPS typically requires PyTorch). Installing them increases the runtime footprint — consider doing that in an isolated environment. - The skill will read any file paths you pass (blend_file, output_path directory, reference_image). Be cautious about allowing the agent to pass arbitrary filesystem paths; it can read local images used for metric computation. - Tests reference assets/low.blend and assets/high.blend that are not present in the manifest. Expect some test/setup friction; verify asset availability if you plan to run tests. - Metadata/documentation gaps (missing required binary and missing declared install steps) are likely accidental but important. If you plan to use this skill, verify these items manually, run it in a sandbox/VM first, and avoid exposing sensitive files to the agent's working directory. If you want higher assurance, ask the publisher for an install spec or a small trusted release (e.g., pip package or GitHub release) and a clear list of required system packages for the metrics features before enabling this skill in a production agent.

Type: OpenClaw Skill Name: synthclaw Version: 0.1.3 The SynthClaw skill bundle is a legitimate tool designed for automating Blender renders to generate synthetic training data. It demonstrates good security practices by using `subprocess.run(shell=False)` to prevent shell injection, validating that procedural parameters are strictly numeric (int/float), and enforcing timeouts for long-running render processes. The core logic in `src/synthclaw/blender_skill.py` and the Blender-side scripts (`agent_bridge.py`, `analyze_blends.py`) is consistent with the stated purpose and contains no evidence of data exfiltration, unauthorized access, or malicious prompt injection.