The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a suite of Node.js scripts located in the scripts/ directory to perform asset generation. These scripts are intended to be executed by the agent via the shell as part of its operational phases. The scripts handle tasks such as loading brand data, rasterizing HTML to PNG/PDF using Playwright, and generating design tokens.
[EXTERNAL_DOWNLOADS]: The skill requires standard, well-known development dependencies (playwright-core and pptxgenjs) from the official NPM registry. These are used for the stated purpose of generating slide decks and performing browser-based rendering of assets.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an inherent surface for indirect prompt injection because its primary function is to ingest user-provided "brand kits" (containing JSON and Markdown files like design.md and editorial/voice.md) and follow the creative instructions contained within them. An attacker who provides a malicious brand kit could potentially embed instructions designed to manipulate the agent's output during the content generation phase.
Ingestion points: Untrusted data from the brand kit's assets and editorial files (e.g., brand/<id>/editorial/*.md) are read and provided to the agent as context in Phase 3.
Boundary markers: The skill uses template placeholders (e.g., {{SLOT}}, {{TOKEN}}) for content interpolation but does not employ hardened delimiters or "ignore embedded instructions" warnings for the agent when processing user-provided guidelines.
Capability inventory: The skill has the capability to write generated assets to the local filesystem and execute internal scripts.
Sanitization: Standard HTML escaping is performed for web-based templates; however, no specific filtering is applied to the instructional markdown text ingested from the brand kit.
[DATA_EXPOSURE]: The video-brief.mjs script includes the current working directory path (process.cwd()) in the generated video brief JSON. This is used for locating the codebase for video production and is considered a low-risk exposure of local environment metadata.

use-brand