The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from various external file formats and provide the extracted content to the AI agent's context, creating a surface for indirect prompt injection.
Ingestion points: The skill reads data from Markdown and JSON outlines (keynote-build.py), existing Keynote presentation files (keynote-fill.py), and exported PDF or PPTX files (keynote-verify.py).
Boundary markers: The instructions and scripts do not implement strict delimiters or explicit warnings to the agent to ignore potentially malicious instructions embedded within the document content.
Capability inventory: The skill allows the agent to create and modify files on the local system, execute AppleScript to control the Keynote application, and perform batch file conversions.
Sanitization: The scripts perform text normalization to handle macOS-specific formatting characters (e.g., removing soft hyphens and paragraph separators), but they do not sanitize the content for adversarial LLM instructions.
[COMMAND_EXECUTION]: The skill relies on executing system commands to perform its primary automation tasks.
Evidence: The Python and Shell scripts execute osascript to run AppleScript/JXA logic and pdftotext for PDF content extraction.
Safety: The skill follows secure coding practices by passing arguments as lists to subprocess.run rather than using shell strings, which effectively mitigates the risk of command injection into the underlying shell.

keynote