The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-provided notes, external PowerPoint files, or existing HTML decks and renders that content into a persistent HTML file. This creates a surface for indirect prompt injection if the generated file is later processed by an agent or user browser.
Ingestion points: User outline and notes (Phase 2), content extracted from .pptx files during conversion (Phase 1/3), and existing .html decks (Phase 1/4).
Boundary markers: The skill does not explicitly mention the use of delimiters or instructions to ignore embedded content when generating the final HTML file.
Capability inventory: The skill uses the Write tool to create the presentation and the Bash tool to run validation scripts and open the file in the default OS browser.
Sanitization: While the implementation uses a templating syntax (e.g., {{heading}}), it does not include explicit instructions for HTML escaping or content validation to prevent script injection from the source data.

frontend-slides