The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from external sources such as PDFs, Word documents, emails, and HTML files. This creates a surface for indirect prompt injection where malicious instructions hidden within a processed document could influence subsequent agent behavior or lead to data manipulation.
Ingestion points: Usage of partition() and its variants in SKILL.md (e.g., extract_invoice_data, parse_email, and build_corpus functions) to read external files.
Boundary markers: The prompt examples and code snippets do not include explicit delimiters or instructions for the agent to ignore instructions potentially embedded within the documents being parsed.
Capability inventory: The skill includes patterns for writing processed data to the local filesystem using json.dump (e.g., in the document_to_json and build_corpus examples in SKILL.md).
Sanitization: The provided code examples do not demonstrate specific sanitization, filtering, or validation of the extracted text before it is further processed or saved to disk.
[EXTERNAL_DOWNLOADS]: The skill provides installation instructions for the unstructured library and its optional dependencies (like OCR or specific document format support) from standard package registries via pip.
Evidence: References to pip install unstructured and associated extensions in the 'Installation' section of SKILL.md.

data-extractor