The Agent Skills Directory

[DATA_EXFILTRATION]: The Preamble and Telemetry sections of SKILL.md contain shell scripts that read ~/.superstack/config.json to obtain a telemetryTier and a _CONVEX_URL. If telemetry is not 'off', it uses curl to POST system information (uname -s, uname -m), skill version, and timestamps to the retrieved URL.
[COMMAND_EXECUTION]: The skill uses multiple shell commands throughout its workflow: Discovery and configuration commands include cat, grep, sed, cut, uname, date, and mkdir; File management involves echo and cat for telemetry logging, and cp for backing up CSS files; Preview generation uses open (macOS), xdg-open (Linux), and start (Windows) to launch generated HTML previews in the user's browser.
[PROMPT_INJECTION]: The skill's workflow involves reading several project files (.superstack/idea-context.md, package.json, app/globals.css) to ground its generation process, creating an indirect prompt injection surface.
Ingestion points: SKILL.md (Prerequisites check section).
Boundary markers: None are specified to wrap or delimit the data read from project files.
Capability inventory: The skill can write to app/globals.css, app/layout.tsx, and brand.md, and can execute shell commands via curl and file-opening utilities.
Sanitization: No explicit sanitization or validation logic is described for content read from external files before it is processed.
[EXTERNAL_DOWNLOADS]: The typography preview template (references/typography-preview.md) includes a link tag pointing to fonts.googleapis.com and fonts.gstatic.com to load font previews. This uses well-known technology services for legitimate design functionality and does not contribute to verdict escalation.

brand-design