The Agent Skills Directory

[SAFE]: The skill serves as documentation for the Docling document parsing tool. All code examples and references use legitimate, well-known libraries and services.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is parsing external, untrusted documents (PDF, HTML, etc.) and converting them to structured formats. This represents an inherent indirect prompt injection surface where a malicious document could contain instructions intended to influence the agent processing the output.
Ingestion points: Functions like converter.convert(source) and converter.convert_all(sources) in SKILL.md and references/parsing.md allow the ingestion of local files and remote URLs.
Boundary markers: No explicit instructions for the agent to ignore embedded instructions within processed documents are provided in the examples.
Capability inventory: The skill documentation demonstrates file-writing capabilities (save_as_markdown, save_as_html, save_as_json in references/output.md) and network access for document retrieval.
Sanitization: No specific sanitization or filtering of document content is described prior to conversion.
[EXTERNAL_DOWNLOADS]: The skill mentions downloading model artifacts from Hugging Face (ds4sd/SmolDocling-256M-preview) and installing standard packages like docling-core. These are operations involving well-known and trusted technology providers.

docling