The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs a silent installation of the python-pptx library using pip install --user. This dependency is fetched from the official Python Package Index (PyPI), a well-known service.
[COMMAND_EXECUTION]: The skill executes multiple shell commands to run Python scripts for extracting text from slides, performing write-back operations to create translated files, and cleaning up temporary JSON files in the /tmp/ directory.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection because it ingests and processes untrusted data from user-provided PowerPoint files.
Ingestion points: Data is ingested in SKILL.md (Step 2) when the skill extracts all translatable content from the provided .pptx file.
Boundary markers: While the content is structured into JSON blocks for the sub-agents, the prompts do not include explicit instructions to ignore potentially malicious commands or instructions embedded within the slide text.
Capability inventory: The main agent has capabilities to execute shell commands, read/write local files, and install Python packages.
Sanitization: There is no evidence of sanitization or filtering of the extracted slide text to prevent malicious instructions from influencing the SlideTranslator or SlideClassifier agents.

pptx-translator