The Agent Skills Directory

External Data Ingestion Surface: The skill reads and interprets content from Google Slides decks to perform edits and summaries. This is an essential function but involves the processing of potentially untrusted data from slide content.
Ingestion Points: Presentation metadata and slide content are retrieved via tools such as mcp__codex_apps__google_drive_get_presentation and mcp__codex_apps__google_drive_get_slide (referenced in references/reference-google-slides-mcp-discovery.md).
Boundary Markers: The agent is instructed to use visual thumbnails and structural readbacks to verify that its modifications match intended patterns.
Capability Inventory: The skill can execute writes via batchUpdate and performs local processing in a node_repl environment.
Sanitization: The instructions emphasize maintaining structural fidelity and verifying rendered output to prevent unintended content changes.
Local Execution Environment: The skill utilizes a node_repl surface for data processing and small JavaScript utilities. This environment is restricted to non-connector tasks to ensure a clear separation between utility logic and document management.
Visual Verification and Network Operations: To ensure high-quality output, the skill uses curl to fetch rendered slide thumbnails from Google-managed content URLs. This allows the agent to visually inspect its work, ensuring no text is clipped or overlapping before finalizing the task.

google-slides