The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes external command-line tools for image manipulation and optimization.
Specifically, scripts/crop.py calls magick and scripts/compress.py calls pngquant using subprocess.run to process user-provided screenshots.
These subprocess calls use a list of arguments rather than a shell string, and do not enable shell execution (shell=True), which prevents typical shell injection attacks from user-provided parameters such as image dimensions or directory names.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data extracted via OCR from user-provided screenshots, which creates an attack surface for indirect prompt injection.
Ingestion points: Data is ingested from YAML files in the 03-ocr directory, which are created by the LLM based on its interpretation of contact list screenshots.
Boundary markers: The workflow relies on the structured YAML format to isolate data, but lacks explicit instructional delimiters to warn the agent against following instructions embedded in the data itself.
Capability inventory: The skill's capabilities are limited to local file system operations (read/write) and execution of specifically named local binaries (magick, pngquant).
Sanitization: The scripts scripts/analyze.py and scripts/merge.py correctly use yaml.safe_load() for parsing external data. This is a robust security control that prevents the execution of malicious Python objects that might be embedded in the OCR output.

im-contact-sorter