The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests raw material, notes, and external Google Docs/Slides to generate slide content. This processing of untrusted data constitutes a surface for indirect prompt injection attacks.
Ingestion points: User-provided text, notes, and files retrieved from Google Drive via MCP.
Boundary markers: The skill implements mandatory human-in-the-loop (HITL) checkpoints for storyline finalization and final output review to mitigate accidental obedience to embedded instructions.
Capability inventory: The agent has access to the local file system, Google Drive operations, and shell command execution.
Sanitization: Content review is delegated to human approval gates at key stages of the relay.
[DYNAMIC_EXECUTION]: The skill generates and executes temporary Python scripts for slide manipulation and quality assurance (e.g., using python-pptx to fill template fields and pdftoppm for visual verification). These operations are transient and performed within the local environment.
[COMMAND_EXECUTION]: Utilizes shell commands including cp, npx, gh, and soffice to manage file assets, install dependencies, report feedback to GitHub, and convert document formats for QA.
[EXTERNAL_DOWNLOADS]: Fetches visual brand tokens from the vendor's domain (zynkr.ai) and installs required skill components from the author's official GitHub repository.

zynkr-slide