@adityaaria/spark@6.0.20

No blocking static signals were detected.; previous stored version diff introduced dangerous source

User runs `spark dashboard` or `spark ui`

A local web page or reachable network client could create or potentially path-traverse skill files in the project, influencing later AI-agent behavior.

unauthenticated CORS-enabled local dashboard writes user-controlled skill content

If a user starts the dashboard, src/dashboard/server.js listens on port 4321 with permissive CORS and no authorization. Its POST /api/skills route accepts arbitrary name/content and writes SKILL.md under a path derived from that name, creating a route for cross-origin or network-triggered AI-agent skill planting/path traversal. This requires explicit dashboard launch and is not install-time execution.

Static source inspection does not show lifecycle execution, credential theft, exfiltration, or staged payloads, so a malicious block is not justified. The dashboard write API is nevertheless a concrete critical vulnerability affecting AI-agent control surfaces, so warn/suspicious is appropriate.