The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches version metadata from a public GitHub Pages repository to notify the user of available updates.
Evidence: scripts/check-update.mjs retrieves a static JSON manifest from tt-a1i.github.io. This operation is for version notification only and does not execute remote code.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied diagram descriptions and Mermaid code, presenting an attack surface for indirect instructions.
Ingestion points: JSON and Mermaid source files provided as diagram specifications.
Boundary markers: Absent within the structured JSON data.
Capability inventory: File system writing (deliver command), network requests (brand-marks.mjs), and local execution of development tools like Git and Node.js.
Sanitization: The risk is managed through strict ajv schema validation, SSRF protections that block private IP ranges and non-standard ports during URL capture, and thorough HTML entity escaping for all rendered output.
[COMMAND_EXECUTION]: The skill invokes local command-line tools to verify source code evidence and open generated diagrams.
Evidence: bin/archify.mjs, bin/open-artifact.mjs, and renderers/shared/repository-evidence.mjs use spawn and spawnSync with shell: false and argument arrays. This prevents shell interpolation and protects against command injection via user-supplied file paths or brand URLs.

archify