The Agent Skills Directory

[COMMAND_EXECUTION]: The test script scripts/formular/selbsttest_form.py uses subprocess.run to invoke the Windows mklink command. This is used solely for setting up directory junctions during testing of path-traversal defenses and uses controlled paths from temporary directories.
[INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection as it ingests untrusted data to generate forms.
Ingestion points: scripts/formular/build_form.py (via JSON data input) and scripts/formular/build_from_markdown.py (via markdown document input).
Boundary markers: Not explicitly defined in the generated HTML for LLM consumption, as the primary consumer is a human user.
Capability inventory: File system writes for HTML generation and local Tkinter UI rendering; subprocess.run is restricted to the test suite.
Sanitization: Extensive sanitization is implemented in build_form.py, including HTML escaping in render_details, URL scheme whitelisting in clean_url (blocking javascript: etc.), and path traversal prevention in the _inside function.
[SAFE]: The core logic implements strong security controls. Specifically, scripts/formular/build_form.py includes validation that prevents path traversal by blocking absolute paths, UNC paths, and parent directory references (..), while ensuring resolved paths remain within the intended data directory. All external downloads (e.g., the markdown and playwright packages) are standard tools from trusted registries.

decision-draw