The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs multiple types of external network operations:
assets/chart-skills/archify/scripts/check-update.mjs fetches an update manifest from a non-whitelisted external domain (tt-a1i.github.io).
assets/chart-skills/archify/renderers/shared/brand-marks.mjs includes logic in captureRemoteBrand to fetch site icons from arbitrary user-provided HTTP(S) URLs. While it implements IP-based filtering (isPrivateBrandAddress) to prevent access to internal networks, it still represents an intentional Server-Side Request Forgery (SSRF) surface area.
[COMMAND_EXECUTION]: The skill facilitates the execution of various system commands and subprocesses:
The main SKILL.md instructs the agent to use msedge.exe in headless mode to take screenshots of generated HTML files, which provides the agent with a powerful way to interact with the local filesystem and system resources.
assets/chart-skills/archify/bin/open-artifact.mjs uses child_process.spawnSync to execute platform-specific openers like open, xdg-open, and powershell.exe to launch local files.
assets/chart-skills/archify/bin/archify.mjs and assets/chart-skills/archify/bin/preview.mjs spawn Node.js subprocesses to manage the diagram rendering and live preview server cycles.
[INDIRECT_PROMPT_INJECTION]: The skill defines a large attack surface for indirect prompt injection:
Ingestion points: It ingests untrusted data from Mermaid or JSON specifications provided by users or external sources.
Boundary markers: It employs JSON Schema validation (assets/chart-skills/archify/renderers/shared/validator.mjs) to enforce input structure, which helps mitigate accidental obedience but may not stop determined adversarial input.
Capability inventory: The skill possesses significant capabilities, including file system writes (fs.writeFileSync, fs.renameSync), network requests (http.request, fetch), and arbitrary process execution (spawn).
Sanitization: Input is validated against a schema before processing, though the resulting rendered output (HTML/SVG) is then used in contexts (like headless browsers) where crafted content might attempt further exploitation.

swiss-shot-demos