The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from arbitrary external websites to generate site content, creating a potential surface for indirect prompt injection if the source site contains malicious instructions designed to influence the agent's behavior.
Ingestion points: Website content is ingested into the capture/ directory via capture scripts (referenced in references/phases.md).
Boundary markers: The skill implements human-in-the-loop verification at multiple stages, including a decision ledger in docs/plan.md, a pixel-level visual review workflow (review/all.html), and a final mandatory pull request review.
Capability inventory: The skill utilizes subprocess execution for Git and NPM operations, performs authenticated network requests for media uploads to Vercel or Cloudflare, and writes generated code to the local file system.
Sanitization: Content is parsed using the parse5 library and reproduced as "captured markup" without transformation to ensure parity, relying on manual review gates to identify and handle anomalies.
[EXTERNAL_DOWNLOADS]: Fetches project templates and toolsets from the author's repository and utilizes official package runners for standard cloud services.
Evidence: The skill clones its core components from github.com/ethanplusai/astro-conversion-kit and utilizes npx to execute official playwright, vercel, and wrangler CLI tools for deployment and auditing.

convert-site