The Agent Skills Directory

[PROMPT_INJECTION] (SAFE): No instructions were found that attempt to bypass AI safety guidelines or override agent behavior. The markdown and JSON files contain only technical instructions and data schemas.
[DATA_EXFILTRATION] (SAFE): No network requests or sensitive file access patterns (e.g., accessing SSH keys or environment files) were detected. The skill only performs local file operations necessary for reading specs and writing generated images.
[EXTERNAL_DOWNLOADS] (SAFE): Dependencies listed in scripts/requirements.txt (cairosvg, pillow) are trusted, standard packages for vector graphics and image processing.
[REMOTE_CODE_EXECUTION] (SAFE): No instances of eval(), exec(), or untrusted remote script execution were found. The code uses standard library modules for its operations.
[INDIRECT_PROMPT_INJECTION] (SAFE):
Ingestion points: The skill reads JSON specifications (scripts/svg_cli.py) and parses SVG XML data (scripts/svg_skill/validate.py).
Boundary markers: Not explicitly defined, but the skill processes structured data rather than unstructured natural language.
Capability inventory: Limited to writing SVG and PNG files to the local filesystem and performing pixel-level comparisons.
Sanitization: Uses xml.etree.ElementTree for XML parsing; while not hardened against all theoretical XML-based attacks (like Billion Laughs), there is no evidence of malicious intent or exposure to untrusted external network data.
[DYNAMIC_EXECUTION] (SAFE): The use of sys.path.insert in svg_cli.py and self_test.py is a standard developer practice for local module resolution and does not constitute a security risk in this context.

svg-precision