The Agent Skills Directory

[SAFE]: All identified operations are local, transparent, and strictly aligned with the skill's stated purpose for engineering and slide design auditing. The skill operates on files provided by the user within the local environment.
[COMMAND_EXECUTION]: The skill utilizes local Python scripts (scripts/extract_pptx.py and scripts/verify_layout.py) to process user-provided PPTX files. These scripts extract metadata and enforce layout invariants via the python-pptx library, which is a standard approach for this use case.
[EXTERNAL_DOWNLOADS]: The skill identifies a dependency on python-pptx, a well-known and trusted library available through the official Python Package Index (PyPI). Instructions are provided for the user to install it locally.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided HTML and PPTX content, which represents a potential attack surface for indirect prompt injection if the source files originate from untrusted parties.
Ingestion points: Source HTML decks and generated PPTX files are ingested and analyzed as ground truth (SKILL.md Step 1 and 2).
Boundary markers: Data is extracted into JSON or a markdown audit table; while the skill doesn't specify explicit delimiters for the slide text, the workflow is focused on geometric layout rather than executing slide content.
Capability inventory: The skill uses subprocess calls to execute the provided Python extraction and verification scripts.
Sanitization: Content is parsed using the structural rules of the python-pptx library, which provides a layer of validation for the underlying OOXML data before it is interpreted by the agent.

pptx-html-fidelity-audit