The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external documents (PDFs, text files, and Markdown) which represents a surface for indirect prompt injection if the processed data contains malicious instructions.
Ingestion points: Files processed via xberg extract and text strings passed to the xberg chunk command as shown in SKILL.md examples.
Boundary markers: The skill uses --- chunk N --- as output separators in text format, but does not implement specific boundary markers or instructions to ignore embedded prompts within the source content.
Capability inventory: The skill utilizes the xberg CLI for data extraction and chunking, along with shell utilities like cat and jq.
Sanitization: No explicit sanitization or validation of the input text content is described in the skill instructions.
[EXTERNAL_DOWNLOADS]: The skill downloads tokenizer configurations from HuggingFace (e.g., Xenova/gpt-4o) when the --chunking-tokenizer flag is used to enable token-based sizing.

chunking