The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs remote network requests for two legitimate purposes: checking for updates via a manifest on the author's GitHub Pages domain and fetching brand icons from user-specified URLs. These operations are implemented with significant safeguards: strict timeouts, response size limits (32KB for manifests, 1MB for images), and a validation layer that blocks requests to private, loopback, or non-standard port addresses, effectively mitigating SSRF risks.
[PERSISTENCE]: The skill includes instructions for the AI agent to update global agent configuration files (such as ~/.claude/CLAUDE.md) to save user preferences. This is a standard practice for agentic skills and is explicitly described as requiring user consent before the agent performs the modification.
[INDIRECT_PROMPT_INJECTION]: The skill processes structured JSON or shorthand diagram definitions into HTML and SVG. To prevent cross-site scripting (XSS) or prompt injection via data, all user-provided labels and descriptions are sanitized using a dedicated HTML escaping utility before being rendered into the final output files.
[COMMAND_EXECUTION]: The build and validation pipeline involves executing internal helper scripts and local browsers (e.g., Chrome or Edge) using execFile and spawn. These executions are limited to hardcoded internal paths, and the skill performs thorough path verification to ensure that output files are written to safe, intended locations without clobbering existing system files.

archify