VTL Image Analysis

Measure compositional structure in AI-generated images using the Visual Thinking Lens (VTL) framework. Detects default-mode bias (center lock, radial collaps...

This skill appears coherent and limited to local image analysis and deterministic prompt-patching. Before installing: (1) ensure you trust the skill source (it will run Python code on images you provide), (2) review operators.yaml if you want to control what prompt patches are appended, (3) be aware the scripts read image files and write an output JSON (prompts.json) to the working directory, and (4) dependencies like opencv/scipy install native wheels—install them in a virtualenv if you prefer isolation. The code uses an AST-checked evaluator for rule triggers (a cautious pattern) and contains no network calls or requests for secrets.

Type: OpenClaw Skill Name: vtl-image-analysis Version: 1.0.0 The skill is designed for compositional image analysis and re-prompt generation. The `SKILL.md` provides clear, constraining instructions to the AI agent, including a 'HARD STOP' refusal gate for unreliable measurements. The `vtl_probe.py` script performs standard image processing without suspicious file operations or network calls. The `vtl_regen.py` script includes a robust `safe_eval` function that whitelists AST nodes and restricts built-ins, effectively preventing arbitrary code execution from `operators.yaml` triggers. The prompt modifications in `operators.yaml` are aligned with the stated purpose of improving image composition and do not exhibit malicious intent. No evidence of data exfiltration, persistence, or unauthorized remote control was found.