The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses system utilities to perform document conversion and comparison tasks:
Executes soffice (LibreOffice) for converting presentations to PDF and for headless validation in ooxml/scripts/pack.py and scripts/thumbnail.py.
Executes pdftoppm (Poppler) to convert PDF pages into JPEG images in scripts/thumbnail.py.
Executes git diff to perform character-level comparison of document revisions in ooxml/scripts/validation/redlining.py.
Launches a Chromium browser instance via playwright to render HTML slides in scripts/html2pptx.js.
[EXTERNAL_DOWNLOADS]: The skill relies on several standard packages from official registries (PyPI and NPM), including playwright, sharp, pptxgenjs, and markitdown.
[PROMPT_INJECTION]: The skill contains patterns that override default agent tool behaviors:
Explicitly instructs the agent to ignore tool range limits (e.g., "NEVER set any range limits") when reading documentation and inventory files to ensure full content consumption.
[PROMPT_INJECTION]: There is a potential surface for indirect prompt injection (Category 8) due to the ingestion of untrusted external content:
Ingestion points: Text extraction from user-provided .pptx files via markitdown and scripts/inventory.py; rendering of user-influenced HTML slides in scripts/html2pptx.js.
Boundary markers: Missing explicit delimiters or instructions to ignore embedded commands within extracted slide content.
Capability inventory: Significant capabilities including file system write access, system command execution, and browser rendering.
Sanitization: Lacks explicit instructions for sanitizing content extracted from slides before it is processed by the agent's reasoning engine.

pptx-processing-anthropic