The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The share command fetches the cloudflared utility from its official GitHub repository to enable temporary public tunneling. This uses a well-known and trusted service provider for its intended purpose.
[COMMAND_EXECUTION]: The CLI tool performs local process management to run the HTTP server and the sharing tunnel, utilizing standard utilities like tar for setup and archive extraction.
[INDIRECT_PROMPT_INJECTION]: The skill generates code-heavy HTML content based on user-provided concepts. This represents a known surface where malicious input could influence generated output.
Ingestion points: Concept names and descriptions provided by the user are interpolated into HTML templates within new.mjs.
Boundary markers: The current templates do not employ explicit delimiters between user content and boilerplate code.
Capability inventory: Generated pages include local JavaScript for interactivity but lack capabilities for network operations or file system access.
Sanitization: Post-generation safety is managed by the check utility in check.mjs, which performs static analysis on the output to validate file structure, title hierarchies, and script syntax.
[SAFE]: The project demonstrates defensive programming techniques, including path traversal protections in the serve.mjs implementation and strict origin validation to block administrative API access from public tunnel endpoints.

aha