The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes pdflatex and pdftoppm using subprocess.run in src/aut_sci_ppt/generator/formula_renderer.py to render mathematical formulas into high-definition images. The implementation mitigates risk by using the -no-shell-escape flag, which prevents LaTeX from executing arbitrary shell commands.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to https://latex.codecogs.com as a fallback for rendering LaTeX formulas into PNG images in src/aut_sci_ppt/generator/formula_renderer.py. CodeCogs is a widely recognized service for this specific utility.
[INDIRECT_PROMPT_INJECTION]: The skill parses content from PDF files and structured text which represents a potential attack surface for indirect prompt injection.
Ingestion points: Text extraction from PDF documents is performed using the fitz (PyMuPDF) library in src/aut_sci_ppt/enhanced_agent.py.
Boundary markers: The processing logic does not employ explicit delimiters or system instructions to ignore potential commands embedded in the document text.
Capability inventory: The skill possesses capabilities for file system modification, network communication, and execution of document processing binaries.
Sanitization: The skill extracts semantic academic structure but does not specifically sanitize the content for malicious prompt instructions before processing.
[DYNAMIC_EXECUTION]: The skill utilizes dynamic loading mechanisms for configuration and internal modules.
In src/aut_sci_ppt/config.py, it uses importlib to load shared environment settings from a relative path (../../_shared/env_config.py).
In src/aut_sci_ppt/paginator/smart_paginator.py, __import__ is used to load components from the skill's own models package at runtime.

sci-ppt