The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it extracts untrusted text from external PowerPoint presentations and passes it to translation models without sufficient isolation.
Ingestion points: Content is extracted from slide shapes and tables in scripts/ppt_translator/pipeline.py using the python-pptx library.
Boundary markers: The provider implementations in scripts/ppt_translator/providers/ (e.g., anthropic_provider.py and base.py) define translation tasks but do not use robust delimiters or instructions to treat the input text purely as data.
Capability inventory: The skill possesses filesystem read/write capabilities to manage presentations and intermediate XML files, and it performs network requests to configured LLM provider APIs.
Sanitization: There is no evidence of sanitization or filtering of the extracted text before it is transmitted to the external LLM providers.
[EXTERNAL_DOWNLOADS]: The skill utilizes several official SDKs and standard libraries from well-known sources, including openai, anthropic, google-genai, and python-pptx.

ppt-translator