The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external PDF documents which constitutes an attack surface for indirect prompt injection. A malicious PDF could contain hidden instructions designed to influence the agent's behavior during analysis.
Ingestion points: The skill ingests data from PDF files, as described in SKILL.md and tested in evals/evals.json.
Boundary markers: There are no explicit instructions or delimiters defined to separate the untrusted PDF content from the agent's system instructions.
Capability inventory: The skill has access to Read, Write, Edit, and shell execution via Bash(python:*) and Bash(uv:*).
Sanitization: No specific sanitization or filtering of the PDF content is described prior to it being processed for Q&A or data extraction.
[EXTERNAL_DOWNLOADS]: The skill references several external Python dependencies for its core functionality.
Evidence: SKILL.md and references/pdf_explore_usage.md mention pymupdf (fitz), pdfplumber, and pytesseract (Tesseract OCR). The use of uv run suggests these packages will be retrieved from standard registries at runtime.

alterlab-pdf-explore