The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructions provide specific commands to extract Vercel authentication tokens from the local filesystem to facilitate automated API operations.
Evidence: The script pattern VTOKEN=$(jq -r '.token' "$HOME/Library/Application Support/com.vercel.cli/auth.json") is used in Phase 8 and references/07-vercel-deploy.md to retrieve sensitive authentication data.
[COMMAND_EXECUTION]: The workflow relies on executing a series of shell commands for project deployment, asset conversion, and API interactions.
Evidence: Uses vercel, ffmpeg, sips, jq, and curl to manage the project lifecycle, process videos and images, and communicate with the Vercel REST API.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external content provided in a markdown file which is used to perform batch replacements across the codebase, creating a potential injection surface.
Ingestion points: ~/Downloads/<brand-slug>-content.md (read in Phase 2).
Boundary markers: Absent.
Capability inventory: File-write capabilities (Edit tool), shell command execution, and network access.
Sanitization: Absent.
[EXTERNAL_DOWNLOADS]: The skill fetches resources from external domains and interacts with AI-driven services for content generation.
Evidence: Downloads assets from Google Fonts and interacts with the fal.ai platform for image-to-video generation and Topaz video upscaling.

luxury-editorial-site-builder