The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and summarize external research papers provided as PDFs or DOIs. This creates a surface for indirect prompt injection where malicious instructions embedded in a paper's text could influence the agent's behavior during the analysis or slide generation phase.
Ingestion points: The skill reads external paper content in Phase 0 (Paper Analysis) and Phase 1 (Supporting Research) via the /search-lit tool, and extracts text/figures from PDFs in Phase 3 using utility scripts like scripts/extract_pdf_figures.py and scripts/check_text_overflow.py.
Boundary markers: Absent. The instructions do not explicitly warn the agent to ignore potential instructions embedded within the research material.
Capability inventory: The skill possesses significant capabilities including executing local shell commands via subprocess, writing files to the project directory, and performing network operations through agent tools like /search-lit and /make-figures.
Sanitization: No explicit sanitization or filtering of the text extracted from research papers is performed before it is used to generate the speaker script or slide content.
[COMMAND_EXECUTION]: The skill automates several document processing tasks by executing local binaries and Python scripts via the shell.
Evidence: Multiple scripts, including scripts/check_text_overflow.py and scripts/extract_pdf_figures.py, utilize the subprocess.run method to invoke external utilities such as pdftoppm, pdftotext, and soffice. While these calls generally follow safe practices (using argument lists rather than shell strings), they represent an execution surface that relies on the integrity of the input file paths.
[EXTERNAL_DOWNLOADS]: The skill uses tools to search for and download academic papers from external research databases to provide context and supporting evidence for the presentation.

present-paper