The Agent Economy Architect

• Outcome: Review
• Latest audit: May 29, 2026 at 9:33 PM UTC
• Version:
  1.3.1

Overview

This is a non-executable educational guide, but it asks for sensitive payment and signing credentials with inconsistent sandbox versus production guidance.

Review before installing. Treat this as a high-impact payment and identity guide, not just passive reading material. Use sandbox or test credentials only, do not provide production Stripe keys or private signing material unless you intentionally plan to run the examples, and require explicit human approval before deposits, escrow releases, subscriptions, API-key issuance, webhook registration, or production launch steps.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (4)

Intent-Code Divergence

Medium

• Confidence: 95% confidence
• Finding: The guide explicitly says examples use the sandbox with no API key required, but the sample code immediately reads GREENHELIX_API_KEY from the environment and posts to the production-style api.greenhelix.net endpoint. This mismatch can cause users to expose real credentials and interact with live infrastructure when they believe they are in a harmless sandbox context.

Intent-Code Divergence

High

• Confidence: 98% confidence
• Finding: The manifest asks users to provide AGENT_SIGNING_KEY via environment variables, while the guide later says private keys should never be stored in environment variables. Encouraging storage of signing keys in environment variables increases the chance of leakage through process listings, logs, crash reports, shells, CI/CD output, and inherited subprocess environments.

Intent-Code Divergence

Medium

• Confidence: 86% confidence
• Finding: The trust scoring pipeline uses submit_metrics with an empty payload as if it were a read operation. Reusing a write-style endpoint for reads can trigger unintended state changes, audit noise, authorization bypass assumptions, or data integrity issues if the API later interprets empty submissions differently.

Missing User Warnings

Medium

• Confidence: 83% confidence
• Finding: The guide discusses cryptographic signing keys and credential use but does not give clear, immediate handling guidance near the examples that introduce them. In practice, users may paste private signing keys into insecure storage or logs, especially because the surrounding material normalizes environment-based credential supply.

VirusTotal

64/64 vendors flagged this skill as clean.