Security audit

Dashboard Builder

Security checks across malware telemetry and agentic risk

Overview

This dashboard builder matches its stated purpose, but it gives an agent powerful database, migration, deployment, and service-key workflows that need careful review before use.

Install only if you are comfortable letting an agent create project files, install packages, generate SQL migrations, and guide deployment. Use a test Supabase project first, review generated migrations before running db push, keep SUPABASE_SERVICE_ROLE_KEY server-only and out of source control, and prefer least-privilege or backend-mediated writes where possible.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands

Findings (7)

Context-Inappropriate Capability

Medium

Confidence: 95% confidence
Finding: The skill instructs the agent to search from the filesystem root (`find /`) to locate templates, which expands file discovery well beyond the project workspace. In an agent context, this can expose unrelated files, secrets, or other tenants' data and violates least-privilege assumptions even if the goal is only to find a template directory.

Context-Inappropriate Capability

Medium

Confidence: 89% confidence
Finding: The skill recommends a sync mode where an agent writes directly to Supabase using the service role key as a general pattern. Service role credentials bypass Row Level Security, so any misuse, prompt-injection-driven action, or implementation error could lead to unrestricted reads/writes across user data.

Intent-Code Divergence

Medium

Confidence: 98% confidence
Finding: The document claims all tables will have per-operation RLS, but many schemas leave RLS generation as TODO comments rather than concrete policies. In an agent-executed build spec, these omissions are dangerous because implementers may ship tables without effective authorization controls, causing cross-tenant data exposure or unauthorized modification.

Intent-Code Divergence

Medium

Confidence: 96% confidence
Finding: The spec mandates Zod validation for API routes, but the sample webhook handler accepts arbitrary JSON and passes it directly to a skill-specific ingest handler. This creates a clear trust-boundary failure: malformed or malicious payloads can trigger unsafe logic, bad writes, denial of service, or downstream injection bugs in skill handlers.

Missing User Warnings

Medium

Confidence: 85% confidence
Finding: The README explicitly promotes a flow where an agent scaffolds a project, generates database migrations, and deploys automatically, but it does not warn users that these actions can create or modify infrastructure, schemas, and hosted applications with real-world consequences. In an agent-skill context, this is dangerous because the documentation encourages high-impact automation from a single prompt, which increases the chance of unintended deployment, destructive schema changes, or insecure default setups without informed user confirmation.

Vague Triggers

Medium

Confidence: 86% confidence
Finding: The activation phrases are very broad and match common dashboard-related requests, increasing the chance this skill will trigger in contexts not specifically intended for it. Over-broad invocation can cause the agent to scaffold projects, inspect manifests, run commands, or handle secrets when the user only asked a generic question about dashboards.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The guide instructs users to place a Supabase service-role key in `.env.local` alongside client-facing `NEXT_PUBLIC_*` values, but does not warn that the service-role key is highly privileged and must never be exposed to browser code or committed to source control. In a dashboard-generation context, readers may incorrectly use this secret in frontend pages or generated skill code, which could bypass RLS and grant full database access if leaked.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

Detected: suspicious.prompt_injection_instructions

Prompt-injection style instruction pattern detected.

Warn

Code: suspicious.prompt_injection_instructions
Location: SETUP-PROMPT.md:8

Prompt-injection style instruction pattern detected.

Warn

Code: suspicious.prompt_injection_instructions
Location: SKILL.md:22