The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted HTML files to generate automation scripts, creating an attack surface where malicious input could influence the generated code. A crafted HTML file could use field names or URLs containing quote marks and Python code to inject instructions into the resulting output script.\n
Ingestion points: The scripts/form_automation_builder.py script reads HTML content from user-provided files via the --html-file argument.\n
Boundary markers: There are no markers or instructions in the code generation logic to delimit or ignore potentially malicious patterns in the input HTML.\n
Capability inventory: Both scripts/form_automation_builder.py and scripts/scraping_toolkit.py utilize file-writing capabilities (Path.write_text) to save generated Python scripts to the filesystem.\n
Sanitization: The generator logic lacks sanitization for HTML attributes (e.g., input names, form actions) before interpolating them into f-string templates used for code generation.\n- [DYNAMIC_EXECUTION]: The skill performs runtime generation of executable Python scripts from predefined templates.\n
Evidence: The tools scripts/form_automation_builder.py and scripts/scraping_toolkit.py use string concatenation and text wrapping to assemble complete Python scripts that use the requests library, then write these scripts to the local disk for execution.

browser-automation