The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's icon vendoring utility (scripts/vendor-lucide.mjs) downloads the lucide-react library from the official npm registry. The requested version is validated against a strict alphanumeric regex to prevent command injection during the installation process.
[COMMAND_EXECUTION]: The kit utilizes the gh (GitHub CLI) to retrieve authentication tokens for its shared repository-backed storage feature. This is restricted to standard developer authentication workflows and utilizes established platform tools for credential management.
[DATA_EXFILTRATION]: The skill implements a robust SSRF (Server-Side Request Forgery) protection layer in kit/net.mjs. This utility explicitly blocks network requests to loopback addresses, private networks, and cloud provider metadata services (IMDS), ensuring that extension network calls cannot reach the local host or sensitive internal infrastructure.
[INDIRECT_PROMPT_INJECTION]: The skill includes dedicated components for handling untrusted data safely. The kit/deeplinks.mjs module provides a quoteUntrusted helper to sanitize external data embedded in kickoff prompts, while kit/validate.mjs enforces strict data structures on all inputs via a dependency-free JSON-Schema validator.
[SAFE]: Minified vendor files (preact-htm-standalone.mjs) are included to provide a stable, dependency-free runtime environment and contain legitimate, well-known libraries with no malicious behaviors detected.

create-canvas-app