The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection as it ingests and processes untrusted data from multiple external sources.
Ingestion points: Data enters the agent's context through Product Requirements Documents (PRDs), content crawled from external sites using scripts/crawl.py, and visual analysis of user-provided screenshots.
Boundary markers: The skill implements a 'spec-first' workflow where a DESIGN.md file is generated for user review and approval (Phase B) before any implementation code is generated (Phase C). This manual review acts as a security boundary.
Capability inventory: The skill performs subprocess calls to execute local Python scripts (crawl.py, extract_tokens.py, fetch_images.py), makes network requests for crawling and API access, and writes generated code files to the project directory.
Sanitization: The skill instructions do not specify any automated filtering or sanitization of the external content ingested from crawled websites or PRDs.
[EXTERNAL_DOWNLOADS]: The skill utilizes external packages and services to support its design workflow.
References the installation of standard, well-known Python packages playwright and beautifulsoup4 via pip.
Fetches image assets from the Unsplash API using a developer-provided access key.
Provides instructions to clone the skill's source repository from GitHub.
[COMMAND_EXECUTION]: The skill documentation describes the execution of utility scripts to automate parts of the design process.
Commands are provided for running crawl.py to extract design tokens from URLs.
Commands are provided for running extract_tokens.py to analyze static files.
Commands are provided for running fetch_images.py to retrieve assets from Unsplash.

xiaopu-web-design-skill