ClawHub

Agent Credential Wallets: Verifiable Intent & Delegation Chains

Security checks across malware telemetry and agentic risk

Overview

This is a documentation-only GreenHelix guide with disclosed, purpose-aligned API examples, but users should treat the examples as live state-changing code if they run them.

Install only if you want a GreenHelix-focused guide for agent credential wallets and agent commerce. Use sandbox credentials or a least-privilege key, verify GREENHELIX_API_URL before running any snippet, and do not reuse the verifier examples in production without adding real DID resolution, signature, disclosure, expiration, revocation, and constraint-chain checks.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Description-Behavior Mismatch

Medium
Confidence
97% confidence
Finding
The guide explicitly frames itself as non-executing educational content, yet the included Python examples perform real HTTP POST requests and invoke state-changing tools such as create_intent, create_dispute, submit_metrics, and check_compliance. That mismatch increases the risk that users will run the examples against real environments or production credentials without understanding they can trigger external side effects, including financial or audit-impacting operations.

Intent-Code Divergence

Medium
Confidence
88% confidence
Finding
The AP2Adapter.verify_counterparty docstring claims DID resolution and disclosure validation, but the implementation only checks the protocol/format fields and fetches reputation data. This can mislead integrators into believing cryptographic identity and selective-disclosure proofs are being verified when they are not, enabling acceptance of forged or malformed presentations.

Missing User Warnings

Medium
Confidence
94% confidence
Finding
The examples demonstrate payment execution through create_intent and budget tracking but do not prominently warn that running the sample against a configured production environment may initiate real financial intents or spending workflows. In a commerce-focused skill, omission of an explicit safety warning materially raises the chance of accidental transactions and unintended operational impact.

VirusTotal

62/62 vendors flagged this skill as clean.

View on VirusTotal