The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local project and external URLs during visual testing, which could allow malicious content to influence agent behavior.
Ingestion points: The input and target arguments in the Python scripts (scripts/run_visual_tests.py, scripts/update_baselines.py, scripts/analyze_diffs.py) allow the skill to read and analyze arbitrary directories and files.
Boundary markers: The skill lacks explicit delimiters or instructions to ignore embedded commands within the files or web pages it processes.
Capability inventory: The skill has access to Read, Write, Grep, Glob, and Bash tools, and the Python scripts perform directory traversal and file metadata analysis.
Sanitization: There is no evidence of content sanitization or validation of the data read from the target project files or the screenshots captured from external URLs.
[METADATA_POISONING]: The skill contains minor documentation inconsistencies that could be misleading regarding its implementation implementation.
Evidence: The scripts/README.md file and the docstrings in the Python files (e.g., scripts/run_visual_tests.py) refer to shell script extensions (.sh) while the actual provided files use Python extensions (.py).

testing-visual-regression